Dubovoy: Artificial Intelligence - Automatic Product Generator

Update "Dubovoi: Artificial Intelligence" 17.0.19.

What's new:

- Improved matching with existing catalog properties in AI mode. Explicit symbolic codes and unambiguous exact matches are not replaced by model assumptions.

What has been improved:

- A false "no license" message that appeared after an action error in the directory has been fixed. Action errors are now handled separately from license verification.

- Saving directory characteristics uses the original value without premature splitting.

- Processing of batch requests for background job statuses has been improved.

- Image and page loading is protected: connections are directed only to verified public IPs.

- HTML preview and draft cleaning has been strengthened without removing user notes.

- Changing the API mode and address is restricted to administrative rights. Partial saving of settings does not reset missing fields.

- Creation of a weak internal token upon failure of the system random number generator has been prevented.

- File logs are allowed only outside the site's public directory.

Why it is important to update:

- The update prevents incorrect replacement of selected characteristics and false "no license" messages.

- Protection of administrative settings, external file loading, and generation result display has been strengthened.

What to do after updating:

- Reload the settings and background generation pages.

- Check the selected profile and characteristic matching in the directory.

- Before mass processing, save the generation result for one product and check its card.

Update "Dubovoi: Artificial Intelligence" 17.0.18.

What's new:

- Searching for ready-made images more accurately matches technical product variants by Du/DN and Ru/PN.

- Equivalent pressure designations in bar and MPa are supported without loss of the decimal part.

What has been improved:

- False matches of PN2.5 with PN25 and mixing of parameters from different variants on the same page have been eliminated.

- An image with an explicit code of another model is no longer accepted due to matching alt text or adjacent JSON-LD.

- Cached cards are re-verified by brand, model, numeric parameters, and mandatory article number.

- Article numbers are compared as separate values: SKU120 does not match SKU1200, and PN40 is not replaced by RU40.

- An empty article property is no longer replaced with the product name.

- Explicit anti-bot verification is recognized even in 404/410 responses and independently of the order of HTTP headers.

Why it is important to update:

- Existing profiles, search modes, and old client settings continue to work.

- Ordinary 404/410 responses are not considered blocking only due to the general Cloudflare header.

- Opaque file names from a confirmed gallery are supported, but explicit signs of another product are rejected.

What to do after updating:

- Update the generator page and background results page so that the browser loads the current handlers.

- For accurate search, it is recommended to pass the article number via the configured product property.

Update "Dubovoi: Artificial Intelligence" 17.0.17.

What's new:

- Bulk saving no longer discards images without service confirmation of the source.

- The "Save all images displayed on the page" mode truly saves all available result images.

- Images created by the neural network are saved with a single common button alongside found images.

- The erroneous message stating that auto-save is disabled by protective checks has been removed.

What has been improved:

- In "Save all images" mode, all displayed valid images are saved.

- In "Save only selected images" mode, only images marked by the user are saved.

- With moderation enabled, automatic background recording still does not execute until the user makes a decision.

Why it is important to update:

- Profile settings are once again the main and predictable rule for image saving.

- The bulk button no longer ends the operation with a message about the absence of allowed images.

- Access rights, session, and CSRF checks are preserved.

- Checks for the validity of remote URLs, type, and content of uploaded files are preserved.

- Product identification by article, supplier, and other configured fields remains unchanged.

- Result format and compatibility with existing profiles are preserved.

What to do after the update:

- Refresh the background generation page so the browser loads the current JavaScript.

- For automatic bulk saving, select the "Save all images displayed on the page" mode in the profile.

Update "Dubovoi: Artificial Intelligence" 17.0.16.

What's new:

- After temporary unavailability of the licensing server, the module normally updates the outdated signed cache during the next access check.

- When specifying symbolic codes for characteristics, allowed values are now bound to the exact code and property ID.

What has been improved:

- The Control Center, settings, and generation pages use a single actual license status.

- The issue has been resolved where the Control Center displayed an active license while generation remained blocked by the old cache state.

- Catalog properties with the same name are no longer mixed when preparing lists of allowed values for the neural network.

- When saving, the exact code and property ID take precedence: a dictionary entry with the same name but a different code can no longer redirect the value to another property.

- Ambiguous selection of properties with the same name is no longer automatically confirmed.

Why it is important to update:

- Generation is automatically restored after the central licensing server resumes operation.

- Checks for signature, domain, license validity, and update rights have not been weakened.

- Existing restrictions on repeated requests and protection against parallel license updates are preserved.

- The AJAX response format and compatibility with existing installations are preserved.

- The mode for selecting characteristics by name is retained for previously configured profiles.

- Old unbound directory entries continue to work as a fallback by name if they do not have their own code or ID.

What to do after updating:

- Open the Control Center or the generation page: if necessary, the module will update the license status automatically.

Update "Dubovoi: Artificial Intelligence" 17.0.15.

What's new:

- Characteristic names can now be specified using property names or property symbolic codes.

- Code support works when uploading Excel files, manual entry, and selecting products from the catalog.

- A list interpretation mode has been added to the profile: automatic, codes only, or names only.

- A safe format and separator selection have been added for allowed values of lists and directories.

What has been improved:

- Values containing slashes within the text, such as "Entrance/exit", are saved in full when using the recommended safe format.

- Values with rearranged words, such as "Internal threads", are matched with the existing variant "Internal threads" without creating a duplicate.

- In case of ambiguous matching, the module does not randomly select a property or value.

Why it is important to update:

- Old profiles and lists specified by names continue to work.

- The automatic mode allows using both codes and names simultaneously.

- The old separator "/" is preserved as a separate compatible mode.

What to do after updating:

- In the profile settings, if names are identical, it is recommended to select the "Property symbolic codes" mode.

- For allowed values of lists and directories, it is recommended to keep the safe format selected by default.

- Before a mass launch, check one product with a characteristic containing a slash and one product with a value consisting of multiple words.

Update "Dubovoi: Artificial Intelligence" 17.0.14.

What's new:

- AI analysis of section properties takes the selected API mode into account. In "Dubovoi API" mode, the request is sent via the central service and does not require a separate OpenAI key on the site.

What has been improved:

- Direct mode with own keys and proxy continues to use the previous route. Yandex AI requests with own keys also remain on the direct route.

- The central service receipt for AI analysis of properties is saved in the local cost history with the model and amount specified.

- If the central service fails, its reason is displayed instead of the generic provider code.

Why it is important to update:

- The error section_props_ai_error: openai_provider_error has been fixed during AI analysis of properties on sites using Dubovoi API.

- The update does not change existing information block properties and drafts: only the route for the new AI request is affected.

What to do after the update:

- Check the selected API mode in the module settings and repeat the AI analysis of one section.

- If OpenAI via Dubovoi API is selected, ensure that the central service receipt appears in the request history.

Update 'Dubovoi: Artificial Intelligence' 17.0.13.

What's new:

- Log settings, log file, and request history settings are now shared across all profiles. Switching profiles no longer changes logging. Obsolete profile-specific copies of these settings are ignored and cleared upon saving settings or switching profiles; shared values and logs are preserved.

What's improved:

- Fixed a JavaScript imageEvidenceManualLabel error when saving images: instead of freezing, an instruction for manual selection is displayed. Image security checks remain in place.

- Repeated manual saving of characteristics is no longer replaced by an old success receipt. The new user action receives a separate operation identifier.

- The fix applies when saving generation results and editing an existing product.

- An action identifier has been added to the background results interface to prevent duplicate delivery of the same request.

- When rejecting a record via property control, handlers return an error instead of an unconditional success message.

- Replacement of multiple dictionary values is performed under a single record control: a failure does not clear previous values, and re-delivery of a manual request does not trigger a new clearing.

Why it is important to update:

- The update resolves the freezing issue when saving images and prevents skipping repeated manual recording of characteristics. Diagnostics are no longer disabled due to profile changes.

- Existing settings, property values, tasks, and operation history are not deleted.

- Rules for skipping filled properties and prohibiting the creation of new values remain unchanged.

- Background task checks and supplier identification fixes from 17.0.12 are preserved.

- The old interface without an action identifier is supported: each manual request is treated as a new save.

What to do after updating:

- Open settings and save the shared logging parameters. Old profile-specific copies will be cleared, while active settings and logs will be preserved.

- Open the background generation page and verify the saving of one result.

Update of "Dubovoi: Artificial Intelligence" 17.0.12.

What's new:

- Added support for the numeric variant of the supplier property "Link to user", which is used in specific Bitrix configurations.

What has been improved:

- Supplier input modes by exact user name and by ID now correctly convert the value to a user ID in the same way.

- For an unknown or ambiguous name, the module does not perform an erroneous write and does not clear the already saved supplier.

Why it is important to update:

- When generating characteristics, descriptions, and images, the module saves the correct supplier and correctly distinguishes products with the same article by the "article + supplier" combination.

- Compatibility with the string variant of the "Link to user" property and with ordinary string, numeric, and list properties is preserved.

- The update does not change catalog settings, tasks, or other data.

What to do after the update:

- Verify the saving of one product with identification "article + supplier" first in the supplier mode "Name", and then, if necessary, in "ID" mode.

Dubovoi: AI version 17.0.11

This update enhances the administrative interface, feature recommendations, and the presentation of related services, Tovaroved and Pixel.

What's new:

- Separate pages for the Tovaroved service and the Pixel module have been added, featuring current capabilities, examples, and links to order or install.

- Banners for related solutions have been added to the control center and product generator, with the option to hide the block on the generator page.

- The Pixel page has been restructured around results: before-and-after examples, card formatting, image unification, presentation scenes, and file weight optimization.

What has been improved:

- A clear sequence of actions has been established in feature recommendations, and server-side confirmations for bulk operations have been added.

- The feature directory, section recommendations, navigation, buttons, and adaptive layout of the administrative interface have been improved.

- The installer safely publishes administrative pages and resources, excluding backup and system files.

- The central API accepts an optional provider timeout limit only as an integer value from 1 to 300 seconds.

Why it is important to update:

- Administrators receive a clearer and safer sequence for preparing section features without modifying existing catalog data.

- The update does not change existing settings, tasks, or data; older clients continue to work without the new timeout parameter.

What to do after updating:

- After installation, open the product generator, Pixel page, feature directory, and section recommendations, then verify the interface and saved profiles.

Update "Dubovoi: Artificial Intelligence" 17.0.10.

This update switches the search for ready-made images to current models, improves the reliability of background processing, and makes module settings clearer for administrators.

What's new:

- Added precise AI search for product cards and full galleries via Luna, Terra, and SOL with brand, model, and article verification.

- Added product identification by the "brand + supplier" combination.

- For supplier properties of type "Link to user," both user IDs and the exact value of the NAME field are supported.

What has been improved:

- Strengthened protection against erroneous images: the source and identification features are saved along with the candidate, and suspicious results are directed to manual moderation.

- Increased reliability of saving previews, detailed images, and galleries; added controlled retries and protection against duplicates.

- Added time limits, background task leasing, safe retries, and provider fuses for stable operation on virtual hosting.

- Strengthened compatibility of the direct API and Dubovoi API: receipts, debits, and repeated responses are processed idempotently; older client versions continue to work via the previous protocol.

- Fixed the update of cached Dubovoi API tariffs.

- Updated menus and settings pages: two-level structure, unified tabs, clearly visible active state, and adaptive wrapping without truncation.

- HTML descriptions of the most recently created products are displayed formatted via a secure server-side filter.

Why it is important to update:

- The module more accurately finds images for specific product modifications and less frequently requires manual correction of galleries.

- Long requests and temporary provider errors have less impact on site availability and do not leave tasks hanging.

- Existing settings and previously created tasks are preserved, and compatibility with older API clients is not broken.

What to do after updating:

- Check one product in both synchronous and background modes.

- Perform an exact gallery search via Luna, Terra, or SOL and confirm article correspondence.

- Open the "Tariffs" and "Write-offs" tabs and compare the test request receipt.

- When using a supplier, check the string property or user binding.

Dubovoi: Intelgen 17.0.9 update.

Changes:

- Fixed parsing of characteristic lists from manual input, Excel, and section settings. Semicolons and line breaks are used as primary separators, while commas within names such as "Pressure, bar" and "Diameter, (mm, inch)" are preserved.

- Compatibility with old comma-separated lists is maintained: the module matches fragments with actual information block properties and directory entries without splitting known property names.

- All permissible values for list and directory properties are passed to the prompt without the previous 20-option limit. Properties without a symbolic code and known properties containing service words are supported.

- For limited list properties, the neural network result is additionally validated after generation. Values outside the allowed list are not saved in the final HTML.

- Excel, CSV, TSV, and TXT import has been redesigned: rows are saved in a unified structured format that preserves semicolons, commas, quotes, line breaks, and the value 0.

- The option to skip the first row is applied exactly once to recognized rows before removing empty product rows, including the header with an empty first cell.

- Article number, supplier, and brand are saved together with the main import row. When saving a product, the current task file is used, not the last found service file from another import.

- Reading, deleting rows, and clearing imports are limited to files of the current user/session. Writing is performed atomically under lock; active imports are not deleted by background cleanup.

- Compatibility with previously uploaded files of the old format and their service identification files is maintained.

Update Dubovoi: Intelgen 17.0.8.

What has changed:

- Fixed HTTP 403 error when creating a text task from a browser after strengthening CSRF protection. The request to the generation handler now passes the standard Bitrix sessid in the URL, which is accessible to check_bitrix_sessid().

- Retrieving sessid is compatible with BX.message(), BX.bitrix_sessid() and server-side fallback, so the fix works on older Bitrix administrative templates as well.

- Mandatory CSRF check for browser requests is preserved. Bypass via internal token remains available only for server-side background workers with a correct internal_token.

Update Dubovoi: Intelgen 17.0.7.

What has changed:

- Fixed processing of the characteristics list for generation if the property name contains a comma. Now properties such as "Pressure, bar", "Diameter, (mm, inch)" and "Maximum operating temperature t max, °C" are not split into fragments.

- For the characteristics list, the main separators are now a line break and a semicolon. This applies to manual input, Excel/mass generation, and the properties list of the UF_INTELGEN_PROPS section.

- During mass generation, the characteristics list from catalog sections is passed to the generator line by line, not concatenated with commas.

- Hints for allowed values for list and dictionary properties are now deduplicated by property name, so that duplicate blocks such as "Material", "Control type", "Passage", etc. do not enter the prompt.

- Fixed the hint for dictionary properties S:directory: the property name is passed to the prompt, not the last value of the highload dictionary.

- The default list of general special instructions has been converted to JSON format so that legacy fallback does not require splitting by comma.

- The default properties dictionary remains disabled: the dictionary control mode uses "No dictionary control", and AI analysis of the dictionary is not enabled automatically.

- The "Task for external AI" section has been hidden from the module menu, as this scenario is not yet ready for regular operation.

- Fixed saving of list and string property values that look like numbers but should be saved as text. For example, the value "up to 80 kg" is no longer converted to "80" for a non-numeric property.

- Fixed saving of single values with separator signs. If the found Bitrix property is not multiple, the value "Universal, for left and right doors" is saved as a single value, not split into two parts.

- For multiple properties, the previous logic of splitting values by comma, semicolon, and other separators is preserved.

Update Dubovoi: Intelgen 17.0.6.

What has changed:

- The default properties directory remains disabled: the directory control mode uses "No directory control," and AI analysis of the directory is not enabled automatically.

- The "Task for external AI" section has been hidden from the module menu, as this scenario is not yet ready for general operation.

- Fixed saving of list and string property values that resemble numbers but must be saved as text. For example, the value "up to 80 kg" is no longer converted to "80" for a non-numeric property.

- Fixed saving of single values with separator characters. If the found Bitrix property is not multiple, the value "Universal, for left and right doors" is saved as a single value, not split into two parts.

- The previous logic for splitting values of multiple properties by commas, semicolons, and other separators has been preserved.

- The same logic has been applied to both synchronous saving and callback saving of generation results.

Update Dubovoi: Intelgen 17.0.5.

What has changed:

- Improved model normalization when searching for images for products with Cyrillic and Latin designations such as "MR-27M", "MR 27M", "MR-27M" and "MP-27M".

- Added support for the MR/MP alias only for codes with a numeric model part, so the module can find cards with an alternative model recording.

- Direct OpenAI image search now ranks and filters images by exact variant/model identity before falling back to general product-score.

- OpenAI Pages now preserves exact card URLs of the form /mr-27m/ and does not discard them due to a long search query.

- Strengthened protection against adjacent models: results for MR-155, MR-153, MP-18MN and similar models are not accepted for the MR/MP-27M query.

- Packaging and quantity designations such as 4x12, 4H12 and 4H12GR are not considered strict model identity and do not break filtering.

- Previously added checks for Pettric, K26S IMX415, NORD/HIBERG and Yandex Images marketplace fallback have been preserved.

- Fixed text generation via YandexGPT in Dubovoi API mode: when working via the central API, the module no longer requires locally filled `api_yandex_ai` and `yandex_ai_folder_id`.

- For YandexGPT via external API, the formation of the model URI has been clarified when the project folder id is available on the site side.

Dubovoi: Intelgen 17.0.4 update.

What has changed:

- Centralized saving of properties via the properties catalog: the "Section Properties" section no longer blocks saving on its own, but is used as a hint and priority mapping.

- Fixed saving of properties in synchronous mode and in callback/background mode: properties missing from the section list are no longer discarded with reason=section_props_restriction.

- For mismatches with the section list, a diagnostic record save_product_property_section_hint / not_in_section_props_allowed_list has been added without stopping the save.

- AI analysis of the catalog and bootstrap of new properties modes no longer depend on the "Section Properties" whitelist; the decision to save is made by the catalog/DecisionLayer.

- Improved relevance of the "Search for ready-made images via neural network (OpenAI)" mode for products with model, sensor, and resolution in the name.

- Resolution designations such as 4K/4K/2.5K are no longer concatenated with model tokens like IMX415 and do not turn them into false mandatory matches.

- The filter for weak product/packaging tokens for feeds and treats has been further refined to avoid accepting other products based solely on weight, taste, or packaging.

- The fallback calculation for web-search billing for direct OpenAI image search has been refined so that OpenAI error responses do not lose data from search calls.

- Strict relevance checks are preserved: similar images are not accepted without a match on model, brand, article number, or key product attributes.

Update Dubovoi: Intelgen 17.0.3.

What has changed:

- Fixed the relevance filter for the mode "Search for ready-made images via neural network (OpenAI)".

Dubovoi Update: Intelgen 17.0.2.

Changes:

- Fixed the relevance filter for the "Search for ready-made images via neural network (OpenAI)" mode.

- Packaging labels such as "4x12 g" and "40 g" are no longer converted into artificial mandatory model tokens like KOSHEK4H12GR or YAGNENKA40GR.

- Relevant product cards with exact names and packaging are no longer discarded solely due to a composite packaging token.

- However, a match based only on weight or packaging is not considered sufficient: a different product with the same "40 g" should not pass the filter without a match in product words, brand, or article.

- The duplicate hint for selecting the image search method has been removed from the settings page.

Update Dubovoi: Intelgen 17.0.1.

What has changed:

- The "Properties and Values Directory" mode with AI analysis has been improved.

- If the directory is empty, paid AI requests are allowed, and the limit has not been exhausted, the first characteristic and value are accepted as initial reference data without offering them to the operator.

- If no suitable characteristic is found, the mode with AI analysis may accept a new characteristic and value as initial directory data without manual offering to the operator.

- If the characteristic is known but the list/directory of values is empty, the first new value is accepted as initial before a paid AI request.

- Ordinary disputed cases, low confidence, type conflicts, or unavailable AI analysis are still sent to the operator for offering.

- A runtime issue in empty directory branches has been fixed: normalization is now performed via a Normalizer instance, without static calls to non-static methods.

- Tooltips and the behavior table on the directory page have been updated to match the new logic for auto-saving in the mode with AI analysis.

What to check after the update:

- On the "Properties and Values Directory" page, select the mode with AI analysis.

- Allow paid AI requests and set a positive limit.

- Check saving a product in a test catalog with an empty directory: the characteristic and value should be saved without offering to the operator.

- Check an unknown characteristic in a non-empty directory: with available AI analysis, it should be accepted as initial, not sent to offers.

Dubovoi: Intelgen 17.0.0 update.

What has changed:

- A new method for obtaining images has been added: "Search for ready-made images via neural network (OpenAI)".

- The new mode uses GPT-5 and web search to find direct URLs of existing product images; new images are not generated.

- If a brand property code is specified in the settings and the product has a brand, the brand is added to the search query.

- The names of image search methods in the settings and setup wizard have been updated: generation by prompt, search via neural network, search and parsing from specified sources.

- Tooltips and selection scenarios for the mode have been updated to separate the generation of new images from the search for ready-made product images.

- The cost calculator has been updated: generating new images now takes into account the selected number of images, and the new neural network search is estimated based on GPT-5 tokens and web search.

- An approximate cost display for the new mode has been added to the pricing page without creating a separate central pricing branch.

Dubovoi: Intelgen 16.0.22 update.

What has changed:

- Fixed saving images from background generation into an existing product when identifying by article.

- If, as a result of background image generation, identification data is located in the nested JSON field result/payload/response/data, the module now correctly extracts identification_info.

- For identification modes "Article" and "Article + Supplier", a safe fallback has been added: if the separate article field is empty but the article value arrived as product_name, it is used as the article value for searching the existing product.

- When a product is found, the background result popup transmits a positive product_id and saves images via the existing-product endpoint, without creating a new product.

- Auto-saving of background images without manual moderation uses the same resolver and also saves images to an existing product if it is found by article.

What is important to check after the update:

- Reopen the background results page so the browser does not use old JS/cache state.

- In article identification mode, open an old or new background image generation result and save the images.

- Ensure that images are added to the existing product with the correct article, and that no new product with an article-as-name is created.

Dubovoi: Intelgen update 16.0.21.

What has changed:

- Fixed saving images with full processing via PhotoRoom when using the site's own API keys.

- The local PhotoRoom script no longer requires the internal Intelgen/Dubovoi API token if the site operates with its own keys.

- Fixed a PhotoRoom error when saving JPGs without an explicitly set background color: an opaque white background is automatically used for JPGs, as JPEG does not support transparency.

- Compatibility of background image saving via `dubovoi_intelgen_save_product_second.php` has been preserved, including the scenario without manual moderation where changes are immediately applied to the product.

- Enhanced security of the image saving endpoint: regular browser/admin POST requests now require a Bitrix sessid.

- Internal-token bypass remains only for genuine server-to-server requests with a valid internal token.

- The weak fallback for generating an internal token via `md5/uniqid/mt_rand` has been removed; tokens are now created only via CSPRNG `random_bytes`.

- If CSPRNG is unavailable, the internal token is not created and the token-bypass is not silently enabled.

Update Dubovoi: Intelgen 16.0.20.

What has changed:

- Improved cleaning of AI-generated product descriptions before saving.

- Automatically removes service tails indicating sources from generated HTML, such as `[Source: ...]`, `[Sources: ...]`, `[Additionally verified ...]`, `Source: ...`, `Sources: ...`.

- Cleaning is applied both in the general HTML sanitizer and in the legacy `stripOpenAILinks()` path to prevent extra source/citation notes from appearing in product cards.

- Useful characteristics containing the word `Source`, such as `Power source: external block`, are preserved.

- Added smoke control for real source/citation tail formats from generation logs.

- Resolved additional remarks from the 1C-Bitrix automated security check: safe output of license status and center notifications, safe output of logs, strengthened checks for the background diagnostic endpoint and image sources in the PhotoRoom/Dubovoi API flow.

- The updater now cleans up the incorrectly included file of another module `admin/options/dubovoi_sku_settings.php` if it is present inside `dubovoi.intelgen`.

What to check after updating:

- Generate or save a product description where the neural network may add lines with sources at the end of the text.

- Ensure that service lines `[Source: ...]` and `[Additionally verified ...]` do not appear in the product card.

- Ensure that ordinary characteristics and text blocks of the product are not truncated.

Important:

- This package applies only to the `dubovoi.intelgen` module.

- Previously released update folders have not been changed; fixes are included in the new version.

What has been changed:

- SSRF risks in processing remote images, external AI packages, OpenAI Pages reference URLs, and background local callback requests have been mitigated.

- The ability to pass arbitrary local files to the PhotoRoom/OpenAI image flow has been blocked: local files are accepted only from the controlled `/upload/dubovoi.intelgen/...` directory.

- Trust in the `Host`/`SERVER_NAME` header from the task during server-side callback requests has been removed; only the strictly verified `web_url_client` from Intelgen settings is used.

- Stored XSS vulnerabilities in settings and result popups have been closed: values are escaped, JS configuration is generated via JSON, and preview HTML is sanitized.

- Transmission of the secret API token to client-side JavaScript has been removed.

- CSRF vectors for clearing logs/history, text verification, and the public background tool have been closed.

- The external endpoint no longer accepts `proxy_server/proxy_auth` from the payload.

Update Dubovoi: Intelgen 16.0.18.

What has changed:

- Expanded extraction of OpenAI Pages images from product asset galleries of the type /assets/images/products/....

- Added priority for large variants lg over mdm/md so that the module selects full-size product photos.

- Improved deduplication of different size variants of the same image, including lg/mdm/md and double extensions such as .png.jpg.

- Updated smoke tests for MAUNFELD-like asset galleries.

Update Dubovoi: Intelgen 16.0.17.

What has changed:

- Improved extraction of OpenAI Pages images from lazy/Swiper galleries of product cards.

- Added support for Bitrix paths of the type /upload//iblock/..., including galleries with lazy-loaded images.

- Strengthened the filter for irrelevant generic images: brand images without a direct match to the model should no longer override product photos.

- Updated identity guard checks for cards where product photos are stored in data-src and in non-standard upload paths.

Dubovoi: Intelgen 16.0.16 update.

Changes:

- Fixed image processing via PhotoRoom when saving images from background search results in Dubovoi API mode.

- Full PhotoRoom modification now passes size settings, margins, shadows, and background color to the central executor.

- A unified contract for transferring PhotoRoom parameters is used for background results with an already created product and for the product creation/search scenario.

Dubovoi: Intelgen 16.0.15 update.

Changes:

- Improved image search via Yandex Images for products with complex alphanumeric models and industrial designations.

- Units of measurement and technical parameters, such as kW and rpm, are no longer used as strict model indicators.

- Added multi-level result selection: exact match, then weak match based on the base model, then verified original Yandex Images results as the final fallback.

- Adjacent models with similar numbers are not promoted to the weak selection solely due to matching numeric suffixes.

Dubovoi: Intelgen 16.0.14 update.

What has changed:

- Fixed storage of source images for background generation via the Dubovoi API: found images are now locally copied to the client site in the result history.

- The background generation popup no longer depends on temporary sandbox links that may be cleared after a day.

- Added a clear message for old results where temporary images are no longer available and no local copy existed yet.

Dubovoi: Intelgen 16.0.13 update.

What has changed:

- OpenAI Pages messages have been clarified: the module no longer displays "product card found" if a similar page was read but the exact card did not pass the model or brand verification.

- A separate flag for a confirmed product card has been added before image extraction, so the operator can see the actual reason for an empty result.

- The "Ready" filter on the background generation page has been fixed: the legacy status `success` is now correctly included in the selection of completed tasks.

- The filter for completed tasks no longer excludes successful responses due to service JSON fields and correctly handles empty error values.

Dubovoi: Intelgen 16.0.12 update.

What has changed:

- Improved search priority for specified sources: if an official domain is set, the module first performs sitemap/slug precheck on it and only then returns to the cache of third-party stores.

- Enhanced ranking of sitemap candidates considering the brand: a similar model series on another brand site no longer receives high priority without a brand match.

- Added diagnostic markers to check cache/precheck behavior in OpenAI Pages logs.

Dubovoi: Intelgen 16.0.11 update.

What has changed:

- Improved exact product card search via sitemap/slug precheck: the module better selects the correct category based on adjacent cards of the same model series.

- Added a universal SEO slug extension by product type to find cards with refined addresses such as gas-electric stoves and gasoline generators.

- Fixed brand extraction for OpenAI Pages: common words in the product name are not treated as a brand before the actual brand.

- Strengthened control of `meta/og` images: technical preview images without a direct brand or model indicator are no longer accepted as product photos.

Update Dubovoi: Intelgen 16.0.10.

Changes:

- Enhanced search for OpenAI Pages images for products with exact model codes and close modifications.

- Refined deterministic sitemap/slug precheck: the module can build and verify an exact product card based on `model_variant`, even if a separate `model_hard` was not identified.

- Added protection against pages from other brands when model codes match: explicit brand guard is applied in cache, precheck, and page scoring.

- Improved handling of long letter series with a single-character model suffix, such as `NRCB 405 W`, without converting standard units of measurement like `230 V` into a model variant.

- Generic `meta/og` preview images (`ogimage`, `opengraph`, placeholder, and similar) are filtered out before downloading to prevent saving abstract images instead of product photos.

Update Dubovoi: Intelgen 16.0.9.

Changes:

- Enhanced multi-layer search for OpenAI Pages images for products with exact model codes.

- Added a universal discovery layer: the module builds safe URL candidates from the product name and the site's sitemap structure to find the exact product card on the official domain faster.

Update Dubovoi: Intelgen 16.0.8.

What has changed:

- Enhanced universal verification of image correspondence to the original product in OpenAI Pages and fallback via Yandex Images.

- The logic is not tied to specific brands, domains, models, or product categories: the module analyzes the model code structure in the product name.

- For alphanumeric models, an anchor-token is now mandatory: a close model with a common suffix must no longer pass as the correct product.

Update Dubovoi: Intelgen 16.0.7.

What has changed:

- Improved OpenAI Pages fallback via Yandex Images for rare precise product modifications: the module now additionally attempts the exact model in quotes, for example "NORD NRB 264NFD SG".

- For queries with a strict model code, the Yandex Images fallback disables automatic typo correction so that Yandex does not substitute SG with a close variant S.

- Yandex Images fallback now checks an extended pool of candidates but returns only the requested number of images to the operator after strict model verification.

Update Dubovoi: Intelgen 16.0.6.

What has changed:

- Limited search via Yandex Images for specified domains and platforms has been added: you can now specify sources such as market.yandex.ru, ozon.ru, wildberries.ru and search for images only within their index.

- For OpenAI Pages, a fallback via Yandex Images for platforms has been added: if the specified source does not provide an image, the module can attempt to find ready-made images in the Yandex index without directly opening the protected marketplace page.

- The Yandex Images fallback within OpenAI Pages also works in background generation: a separate state step has been added, along with protection against re-launch and clear trace events.

- Image search for OpenAI Pages for products with very similar modifications has been improved: the module can now attempt a strict adjacent URL if the found card differs only by an incomplete model suffix.

Dubovoi: Intelgen 16.0.5 update.

What has changed:

- Improved OpenAI Pages image search for products with very similar modifications: the module can now attempt a strict adjacent URL if the found card differs only by an incomplete model suffix.

- Fixed a case class where for products `NRB ... SG` the search saw a close page `NRB ... S` but did not attempt the page `NRB ... SG`.

- A close base modification is still not accepted as the correct product: `NRB 262NFD S` does not pass validation for `NRB 262NFD SG`, and `CN 500` does not pass for `CN 500 B`.

- Enhanced validation for tiles with similar series: `GG 9064 YR` must not be accepted for the query `GE 9064 YR`.

- Improved product card search in sitemap for models where the code in the name and URL is written with different separators: for example `i-RFS` in the name and `i_rfs` in the page address.

- Page filtering by sitemap now compares both the original URL path and the normalized version without `_`, `-`, `/`, while maintaining strict model modification checks.

- Enhanced OpenAI Pages image search for products with similar modifications: the module now filters out unsuitable options at early stages of page selection, including sitemap, precheck, cache, and URL filters.

Improved search for product cards in the sitemap for models where the code in the name and URL is written with different separators: for example `i-RFS` in the name and `i_rfs` in the page address.

Enhanced search for images from OpenAI Pages for products with similar modifications: the module now filters out unsuitable options at early stages of page selection, including sitemap, precheck, cache, and URL filters.

- Improved image search via OpenAI Pages: the module more strictly verifies that the found image relates to the original product, not to similar products, recommendations, catalog lists, or technical images of the page.

- Added protection against erroneous selection of images from "similar products" blocks, product tiles, section sliders, and preload/loader images.

- Added strict verification of model variants: products of the same line are now distinguished by model suffixes, for example `NRB 162NF S`, `NRB 162NF W`, `NRB 162NF SG`, `RFQ 510 GW`.

- Improved protection against substitution of similar modifications: the page `NRB 162NF W` must not be accepted for the request `NRB 162NF S`, and `RFQ 550 GB` must not be accepted for `RFQ 510 GW`.

- Model variant verification is performed carefully: units of measurement such as `230 V`, `12 KG`, `500 ML` do not become rigid model identifiers.

Image search via OpenAI Pages has been improved: the module now more strictly verifies that the found image relates to the original product rather than to similar products, recommendations, catalog lists, or technical images of the page.

Protection against incorrect selection of images from "similar products" blocks, product tiles, section sliders, and preload/loader images has been added.

Update Dubovoi: Intelgen 16.0.0.

What has changed:

- Full external AI processing mode added: the module prepares a ZIP task for the neural network, receives the result, validates the structure, displays a preview, and applies data in batch via the standard save-flow.

- External AI product selection filter expanded: tasks can be generated by information block, sections, properties, and other parameters, not just by manual ID list.

- Property and value dictionary added: a unified database of standard characteristics, values, synonyms, suggestions, and decision history.

- Dictionary integrated into product card saving and External AI apply: confirmed matches are saved automatically; disputed values are sent to suggestions and do not overwrite the catalog without control.

- Dictionary operation modes added: strict control, working mode, no-control mode, and AI-assisted groundwork for subsequent intelligent matching.

- Dictionary operator tools added: import from catalog, dry-run/apply, search, bulk actions, safe cleanup, deletion, merge groundwork, quality reports, and history tab.

- External AI and dictionary interface improved: more hints, localization, summaries, documentation, and clear operator scenarios.

What to check after the update:

- External AI section: product selection, ZIP download, result upload, validation report, preview, and batch apply on a test product.

- Dictionary: import from real catalog iblock, mode switching, property/value search, suggestions, history, and safe bulk actions.

- Product card save-flow: confirmed values are saved; unknown/conflict values go to suggestions; history is created.

- Documentation and Russian localization of new interface terms.

Important:

- This package applies only to the `dubovoi.intelgen` module.

- A final server check in an authorized Bitrix admin panel is recommended before publication.

Improved background operation of OpenAI Pages for external page sources `Google` and `Yandex`.

Improved multi-layer image search functionality

What has changed:

- Fixed the popup preview of images on the background generation page.

- Large images now correctly fit into the popup window.

- Styles in settings have been improved

Added quick entry points to documentation on key pages

Fixed documentation styling

Improved navigation in the built-in documentation

Updated built-in module documentation.

What has changed:

- Improved documentation structure, formatting, and navigation.

- Links to DOCX and PDF versions of the guide have been preserved.

Improved documentation handling - navigation, quick links, and formatting

Module documentation improved

Terminology in the module configuration wizard for specific tasks improved

New version of documentation implemented

Documentation now opens as a single page within the module

- Documentation translated to a unified Markdown source.

- Documentation page interface improved:

- Download links for the manual added to the documentation:

- in DOCX format;

- in PDF format

- A more prominent built-in banner with a link to the documentation and quick access to downloadable manual versions added to the product generator.

Why it is important to update:

- Built-in documentation becomes significantly more convenient for module users.

- Help is now available both on the documentation page itself and directly from the product generator.

Fixed saving of product descriptions and characteristics in scenarios where the module uses draft section properties.

Fixed 500 error when saving data

Improved generation of temporary image URLs in Dubovoi API mode

Images are now stored locally on the server

Improved image generation functionality at the Dubovoi API level

Improved module AI functionality

Improved module security

Module operation improvements

Module security improvements

Background request processing in the module has been strengthened

Dubovoi API performance has been improved

What's new:

- In this version, the background generation page and the associated backend for large queues have been enhanced.

Key changes:

- Load on the background generation results page has been reduced.

- Pop-up window stability and automatic result saving have been strengthened.

- Queue maintenance and the auxiliary background task table schema have been strengthened.

Why it is important to update:

- The risk of hangs and overload when working with a long background queue is reduced.

- Predictability of the operator interface when opening, checking, and saving results is improved.

Module version 15 update

Before updating, be sure to create a backup of the website.

In version 15, we developed our own API service that helps clients quickly launch the module without the need to configure API keys, purchase a proxy server, pay for each service with a foreign bank card, and constantly monitor the balance in each service.

Now, a single balance for all API functions is possible, and the module can be launched quickly without manually assembling the technical infrastructure. We take care of keys, proxies, request routing, and service support.

No need to manually configure API keys for each service.

No need for a foreign bank card to top up neural network balances.

No need to track a separate balance in each service.

No need to buy proxies and monitor their functionality.

No need to increase server limits for heavy module requests.

No need to manually connect new services as the module evolves.

No need to separately connect OpenAI, Google, Yandex, and other services.

No need to figure out which service and which key are needed for a specific module function.

Major tasks included in version 15:

- Implementation of the `Dubovoi API` mode as a separate module operation scenario.

The module now has a full choice between manual API mode and the centralized `Dubovoi API` mode. A connection screen, a separate page with explanations, account status display, balance, available services, and tariffs have been added. This is important because the client receives a simpler and more controlled way to work with neural networks and external services.

- Migration of supported operations to the central execution and billing infrastructure.

For scenarios supported by `Dubovoi API`, request execution, billing calculations, and part of the background tasks have been moved to the licensing center side. This is important because the client sees the final billing for their tariff, not the internal cost of individual APIs.

- Complete overhaul of tariff logic and user cost benchmarks.

In version 15, the `Tariffs` page, cost calculator, and tariff display for `Dubovoi API` have been redesigned. More transparent cost benchmarks have been introduced, along with calculations for various scenarios, separate logic for manual mode and centralized mode. This is important because it is easier for the client to understand the budget before starting generation.

- Enhanced search, generation, and saving of images.

Image search settings have been restored and improved, including minimum image width, product identification during separate search, local temporary saving of found images, fallback for outdated temporary files, and improved diagnostics for temporary storage errors. Scenarios `openai_images` have been separately strengthened and their operation via the license center has been improved. This is important because more suitable images end up in product cards, and the process has become more predictable.

- Interface operator and moderation improvements.

Branch 15.x includes a major set of UX changes: unification of `Dubovoi API` screens, improvement of the generation page, improvement of tariff and debit screens, alignment of the interface for text and image generation, as well as more understandable statuses and user hints. This is important because the module has become easier for daily manual work.

Minor but important improvements:

- Added and stabilized the ability to delete product rows after loading from a file and when working with product lists.

- Improved handling of imported data: empty rows no longer hinder generation and do not clutter the table.

- Restored and stabilized import of product lists from Excel files in scenarios where it is supported by the current distribution.

- For synchronous generation, the product identification block in image search has been unified.

- Unnecessary manual moderation elements that hindered the operator and served no useful purpose have been removed.

- Fixed the behavior of the automatic result expansion option so that the interface works exactly as configured in the settings.

- Corrected top alignment of rows and eliminated specific visual regressions in the generation table.

For synchronous generation, the product identification block in image search has been standardized, the unnecessary manual query field has been removed, top alignment of rows has been restored, and the behavior of the results expansion option has been fixed.

Generation via the Dubovoi API has been improved in both synchronous and background modes.

Improvements:

- For new launches, display and saving of client write-offs from the licensing center have been aligned.

- Repeats, timeouts, and progress tracking for synchronous generation via the Dubovoi API have been fixed.

- The background queue, pause functionality, result retrieval, and related runtime scenarios have been fixed.

- Image search functionality has been restored and strengthened: minimum width requirements, product identification, local temporary storage, error diagnostics, and fallback handling for expired temporary files.

The write-off function has been improved for operation via the Dubovoi API.

Dubovoi API generation improved

- The generation result on the screen is now immediately displayed in a clean format without service links.

- The display of single generation progress has been fixed so that extra requests are not shown.

Why it is important to update:

- The risk of repeated paid API calls when generating a single product is reduced.

- The text on the screen immediately looks the same as it does after saving to the catalog.

Increased reliability of requests via Dubovoi API

What has changed in this version:

- Improved stability of the module when working with `Dubovoi API`

- Fixed the service logs page

- Improved protection of service data in logs and handling of error responses.

Dubovoi: Intelgen 15.0.8 update.

What has changed in this version:

- Improved background generation via `Dubovoi API`. Now tasks are more reliably queued and do not start prematurely.

- Fixed image search via OpenAI. The module now more accurately uses the settings selected on your site.

Added the ability to delete rows after filtering products via an information block or uploading via EXCEL

Improved product loading from EXCEL

Improved accuracy of neural network model billing

Improved calculator functionality in settings

Improvements to Dubovoi API functionality

The update improves the generation cost calculator in settings and unifies tariffs within the module

Improvements to Dubovoi API functionality

Improvements to Dubovoi API functionality. Mandatory update.

Improvements to Dubovoi API functionality

A proprietary API service has been implemented to help clients quickly launch the module without the need to configure API keys, purchase a proxy server, pay for each service with a foreign bank card, or monitor the balance in each service. The API can be replenished in rubles via your company's bank account or via a bank card. We take care of all the complexities. You simply replenish a single balance for all API functions and start generating.

We do not profit from this service; the markup covers only the actual costs:

- fees for using foreign bank cards

- payment system commissions

- taxes on API replenishment inflows

- purchase of proxy servers

- currency exchange rate differences

The difference is obvious:

- You do not need to independently delve into the technical details of configuration

- You do not need to independently configure API keys for each service

- You do not need a foreign bank card to replenish neural network balances

- You do not need to constantly monitor the balance in each service

- You do not need to purchase proxies and monitor their operability under conditions of regular blocks

- You do not need to configure the server and increase server limits for the module to work

- You do not need to keep up with new features in the module and connect each service manually

You use a single API for all services. You replenish the API within the module via a bank account or by card in rubles and immediately start generating.

At the same time, the previous manual mode of operation is preserved, and independent configuration of all API keys and proxies remains possible.

The information block selection function for working with products has been improved

Major module update from version 14.0.1 to 14.0.29.

Before installing the update, be sure to create a full backup of the entire website.

Key new features of the 14th version of the module:

- Image search via Yandex Images has been implemented, and the capabilities for selecting images from various sources have been expanded.

- Generation modes for product descriptions, characteristics, and other data using OpenAI and Yandex AI have been added.

- A solution setup wizard has been implemented to quickly prepare the module for specific website tasks and generation scenarios.

- A function has been added to prepare a list of characteristics for catalog sections, followed by AI-based generation and application.

- Image search and selection modes have been expanded, including scenarios with search engines, alternative names, and analysis of source pages.

- Background data and image generation have been improved, along with result control before applying them to the catalog.

- New source settings, domains, search restrictions, and professional parameters have been added for fine-tuning the module.

- Image validation, filtering, and saving mechanisms have been improved, including handling problematic and unavailable images.

- Diagnostics, logging, and quality control tools have been strengthened for solution setup, testing, and maintenance.

- Stability of key AI scenarios, the wizard interface, and internal result processing mechanisms has been increased.

What the update to the 14th version provides:

- More options for automating catalog population with products, descriptions, characteristics, and images.

- More flexible module configuration for different niches, data sources, and project requirements.

- More convenient initial setup and maintenance of the module without manual modification of each basic scenario.

- Expanded capabilities for professional configuration and phased development of generation for a specific catalog.

Update Dubovoi: Intelgen 14.0.28.

Minor clarifications and improvements in the module

Update Dubovoi: Intelgen 14.0.27.

What was fixed in this version:

- The remaining sync defect of the first click on the Show more button after the initial image loading has been fixed: the module now immediately saves the calculated next_num_start in the DOM/state and does not repeat the first batch of images.

- The native branch of synchronous image generation has been fixed: after the first response from Yandex or Google, the loading cursor is updated to the next page even before the user's first manual click.

Update Dubovoi: Intelgen 14.0.26.

- The residual defect of the first click on the Show more button has been fixed, which previously caused the first batch of images to be displayed again in the Yandex Search API.

Dubovoi: Intelgen 14.0.25 update.

What was fixed in this version:

- Fixed the image loading cursor for the Yandex Search API.

- Fixed the initial initialization of "Show more" in background results: the popup no longer restarts loading from the first page if the first batch has already been displayed.

- Fixed sync image loading: removed the misalignment between the local UI state and the actual num_start, which caused new images to appear only on the second click.

- Added next_num_start to image search logs so that pagination behavior can be accurately diagnosed from the server log without manually reconstructing the step.

- Strengthened the retry strategy for AI analysis of section properties: the module now uses up to four attempts, retains web-tools on repeated empty responses, and writes ai_attempts to background diagnostics.

Update Dubovoi: Intelgen 14.0.24.

- Improved pagination for image search via Yandex Search API

- Added filtering of broken and unavailable images for Google and Yandex before displaying results in the interface and before subsequent saving.

- Enhanced retry strategy for AI analysis of section properties: the module now uses up to three attempts and correctly repeats the request after an empty response without manual task retry.

Update Dubovoi: Intelgen 14.0.23.

- Normalized handling of AI responses during section property analysis: empty and schema-invalid structured JSON responses no longer become false properties.

- Fixed problematic cases in the property generator

- Refined logic for repeated AI requests: a valid short list is no longer considered a reason for retry; repetition remains only for truly empty or truncated responses.

Dubovoi: Intelgen 14.0.22 update.

- Control elements for found images have been restored in synchronous image generation: saving images, the Show More button, image selection, and handlers for changing the order.

- Fixed the passing of the Yandex Search API key to the image search handler.

- The section name for which the AI analysis was performed has been added to the results of background property generation for sections.

- On the Section Properties page, actions for applying drafts, mapping, and binding properties no longer run without selected drafts.

- Improved binding of properties to sections

Dubovoi: Intelgen 14.0.21 update.

What is fixed:

- Image generation and search via OpenAI Images, Google Images, and Yandex Images have been updated.

- Synchronous text generation has been improved.

- Generation of attribute lists in sections has been improved.

- Draft property generation for sections via AI has been fixed.

- Bulk actions on the results page have been clarified: buttons now clearly operate "on this page," and tooltips and localized status messages have been added.

- Tooltips and names in settings and the wizard have been updated: API keys, graphic services, PhotoRoom, OpenAI Images, Yandex AI Studio, Google/Yandex Search, ZenRows, and tariff explanations have been standardized to a single terminology.

- Tariff tooltips and the calculator have been fixed: OpenAI Images is displayed as a calculation per 1 image, Yandex AI Studio is calculated in rubles, and OpenAI and Yandex are separated into different billing models.

- The tariff page has been clarified: different services show calculations in their own unit of measurement — product, image, request, characters, or 1000 tokens.

- The import policy has been updated: active import accepts TXT/CSV/TSV; XLS/XLSX files are not parsed in the browser, and the user is prompted to save Excel files as CSV, TSV, or TXT.

Update Dubovoi: Intelgen 14.0.20.

What was fixed:

- Background product generation progress has been improved

- Synchronous text generation has been improved

- AI analysis of section properties has been improved

Update Dubovoi: Intelgen 14.0.19.

What changed:

- Background processing of section property analysis tasks has been improved

- API error messages have been expanded to now include the selected service

- Calculation and logging of Yandex AI Studio costs have been fixed

- Image search diagnostics have been improved

- Solution setup wizard instructions have been expanded

Update Dubovoi: Intelgen 14.0.18.

What changed:

- Improved file logging.

- Enhanced diagnostics for image search via Google, Yandex, OpenAI Images, and OpenAI Pages: the file now includes engine context, HTTP responses, cURL/JSON errors, service messages, and the number of found images.

- Fixed termination of synchronous text generation: the result string expands hidden cells, stops the loader/animation, and correctly switches the button to the result state even when saving the result fails.

- Fixed recording of Yandex AI Studio costs in the database log: ruble charges no longer appear as `OpenAI API cost: $...` if the actual billing record originated from Yandex.

- Improved stability of the background queue for analyzing section properties: when `section_props_*` tasks are queued, the module enables background mode and reactivates the canonical agent `\Dubovoi\Intelgen\EventHandlers\BackgroundHandler::processQueue();`.

- Expanded the wizard instructions for API keys for OpenAI, Yandex AI Studio, Google Custom Search, Yandex Search API, ZenRows, and PhotoRoom: added hints regarding VPN, foreign bank cards, Yandex scopes, and identity verification for OpenAI Images.

- The final wizard step "Server Settings and Launch" now always displays limit and background queue settings so they are not hidden before launch.

- A separate block for image processing via PhotoRoom has been added to the images step, with explanations of Base/Basic and Pro/Plus modes.

- Wizard values for the "OpenAI Image Background" parameter have been localized.

- In the final settings hint, the OpenAI Images mode no longer receives recommendations about ZenRows and source domains, as they apply to OpenAI Pages.

Update Dubovoi: Intelgen 14.0.17.

What changed:

- Improved background queue operation

- Improved synchronous search

- Fixed image search via Yandex Search API when passing source domains

- Expanded file diagnostics for image search for Google, Yandex, OpenAI Images, and OpenAI Pages: the log now includes the engine, request type, HTTP status, number of images found, service message, and error.

- Improved processing of background tasks for analyzing section properties: `section_props_*` tasks are processed by the internal `BackgroundHandler`, and a direct call to the service endpoint returns a clear diagnostic error instead of a generic `invalid_type`.

- In the "Section Properties" section, added the ability to delete selected analysis drafts.

- In the setup wizard for the "Generate images from scratch" task, the image source is fixed as OpenAI Images and does not offer to select Google or Yandex.

Dubovoi: Intelgen 14.0.16 update.

What changed:

- Fixed profile settings switching on the profiles tab: profile selection no longer depends solely on the jQuery handler.

- Improved resetting of the selected products file: resetting a file already deleted from within the protected import directory no longer returns a false error, while protection against escaping the sandbox is preserved.

- Neural network responses based on Yandex are cleaned of technical markdown wrappers such as code block markers with an html label, closing code block markers, DOCTYPE, empty code block markers in the source, empty code block markers in the source, and empty code block markers in the source, so that unnecessary full document markup does not appear in the result.

- Restructured expense storage: TOTALRUB is used for rubles, a separate TOTALUSD field has been added for USD, and USD estimates are additionally converted to rubles using the module's exchange rate.

- Added a safe transition of the billing table to the new RUB/USD scheme: upon the first transition, the old expense journal is cleared because previous rows might have stored USD in the TOTALRUB field and lacked sufficient data for accurate recalculation.

- Localized background tasks for analyzing section properties and improved the display of completed section_props_* tasks on the background generation page.

Update Dubovoi: Intelgen 14.0.15.

What changed:

- Fixed reset of the selected products file: attempting to delete a file that no longer exists within the protected import directory no longer returns a false error.

- Strict path security validation preserved: reading a file still requires the file to exist, and the missing file mode is permitted only for the reset operation within `/upload/dubovoi.intelgen/`.

- Fixed AJAX requests for the generator and background processing where standard payload requests were incorrectly sent as `processData:false`.

- Fixed product generation startup in scenarios where the global jQuery `$` is unavailable on the page: text quality checks and image helper handlers no longer break text generation after startup.

- For scenarios without image generation, image save and upload handlers are no longer loaded; if jQuery is unavailable in an image scenario, the module displays a controlled error instead of a `ReferenceError`.

- Budget calculation clarified: scenarios using only OpenAI Images do not inherit text model settings; mixed OpenAI/Yandex costs are converted to a single estimate in rubles using the module's exchange rate.

- Correct OpenAI name used in tariffs and tooltips; added lines for OpenAI Images, Yandex AI Studio, and Yandex Search API.

- Enhanced the "Section Properties" page: added manual property exclusions, a warning when starting without selected sections, and tooltips for apply, map, and bind property actions.

- Property analysis made safer: file properties, bindings to elements/sections, and standard marketing properties are excluded; empty bound properties are no longer recommended solely due to section linkage.

- Background property analysis jobs now pass manual exclusions and display more accurately on the background generation page, including results of completed `section_props_*` tasks.

Regression smoke tests for the generator, sandbox file reset, budget, tariffs, and section property analysis have been strengthened.

Update Dubovoi: Intelgen 14.0.14.

What changed:

- The setup wizard after server verification has been improved: an example Excel/CSV file has been added, and hints regarding product identification, Yandex AI Studio, catalog characteristics, catalog properties, and server limitations have been clarified.

- On the API keys step, the wizard now displays only services relevant to the selected scenario; for Yandex-only scenarios, OpenAI/proxy fields are not shown.

- Separate instructions for ZenRows and PhotoRoom have been added, including terms of use, cost risks, and recommendations for initial testing.

- Yandex Search API has been switched from an IAM token to a permanent API key; hints now specify the required roles and key scopes for Yandex Search API and Yandex AI Studio.

- Queue processing on the "Section Properties" page has been fixed: JSON tasks are no longer corrupted by SQL escaping; section search, mode explanations, and a link to background generation after task submission have been added.

- Generation launch on the product generator tab has been fixed to operate without dependency on global jQuery.

- Reading of selected products from the information block has been fixed: AJAX file loading no longer sends the standard payload as `processData:false`.

- The quick start route on the generator page has been clarified: it now reflects actually enabled functions, including OpenAI Images and OpenAI Pages scenarios.

- An estimated cost calculation for generating images via OpenAI Images has been added.

- Budget calculation for OpenAI Pages has been fixed: when the page source is OpenAI, Google Search API costs are no longer added.

- Regression smoke tests for the wizard API steps and OpenAI Pages budget calculation have been strengthened.

Update Dubovoi: Intelgen 14.0.13.

What has changed:

- The setup wizard has been improved: hints regarding profiles, product identification, manual moderation, catalog properties, and server limitations have been clarified.

- Accessibility of OpenAI Pages settings in the wizard has been fixed for the single task "Search for images via search engine".

- When OpenAI Pages is selected as the image source, parameters for page analysis and ZenRows settings for protected sites are now available.

- For single image search, the availability of OpenAI Images parameters has also been aligned if the image generation mode via OpenAI is selected.

- Visual readability of the wizard has been improved: the active step is highlighted with a background, scenario descriptions are not bolded, and HTML hints within steps are formatted more neatly.

- The hint regarding the token limit for YandexGPT has been clarified: cautious initial values have been added, along with a recommendation to increase the limit only after testing.

- Internal validation in the wizard has been refined to prevent recurring contradictions between the selected image source and the available settings for that source.

Update Dubovoi: Intelgen 14.0.12.

What changed:

- The setup wizard has been improved: the separate first step "Operating Mode" has been removed, and parameters for the background queue, pass time, number of products, and domain have been moved to the final step of server settings.

- Settings that should not be fixed in the wizard profile have been removed from the wizard: catalog/info block selection, CSV/Excel separator, skipping the first file row, and manual selection of batch processing mode.

- The product identification step has been aligned with the logic of the settings page: identification variant names now match the settings, and a detailed hint has been added regarding manual moderation and testing on 20-50 products.

- The task selection step has been divided into single and complex scenarios; the selected task and the currently saved scenario are now visually distinct.

- The wizard now shows only steps relevant to the selected task: price, characteristics, and images do not appear if the scenario does not use them.

- The "Show all step parameters" and "Hide irrelevant parameters" buttons have been removed. Irrelevant fields within the current step remain visible but are disabled and visually dimmed so that the interface does not jump when dependent parameters change.

- Selection of the OpenAI / Yandex AI Studio text neural network service has been added in the wizard and on the settings page. The list of models is filtered by the selected service; Yandex models are marked as an experimental option for simple tasks.

- Outdated Temperature and Top P parameters have been removed from the wizard. Hints have been added for the number of search queries and the depth of web context.

- Hints for catalog properties have been expanded: safe mode "only specified characteristics," transition to the list of characteristics generation section, and warning about the risks of arbitrary property creation for the smart filter.

- Instructions for OpenAI, Yandex AI Studio, Google/Yandex Search API, and required roles for service accounts have been added to the API key step.

- The final step now saves settings and immediately redirects to the product generator; unnecessary final buttons have been removed.

- Fixed the reuse of master profiles: when returning to an already created scenario, the master switches to the existing task profile instead of creating duplicates.

- Removed the outdated launch_mode from the master's active state so that the old step does not appear in progress and recommendations.

Dubovoi: Intelgen 14.0.11 update.

What changed:

- The module setup wizard has been completely redesigned: it now guides the administrator step by step, displays one active step at a time, and contains the full set of module settings without separating ordinary and expert parameters.

- The wizard is adapted to tasks: settings for descriptions, characteristics, price, images, rewriting, API keys, and server mode are shown based on the selected task and dependent parameters.

- A summary has been added to the final step of the wizard before launch: what the module will do, which services will be required, and where there are risks related to cost, catalog properties, server load, and mass processing.

- Wizard hints have been strengthened for OpenAI Pages, OpenAI Images, Google/Yandex image search, average price search, catalog properties, smart filter, background mode, hosting, and timeouts.

- The wizard logic and settings page are synchronized via a common SetupWizardService: a unified policy for excluding secret API keys from profiles, unified normalization of conflicting settings, and a common contract for the wizard's final summary.

- A common manifest contract has been added for critical settings: profile secrets, normalization of saving, summary parameters, allowed values, and key dependencies.

- Protection against contradictory configurations has been improved: scenarios involving price, rewriting, description, characteristics, image search, OpenAI Pages, multi-query search, and price rounding are brought to a safe final state.

- The wizard's final summary now correctly accounts for OpenAI dependency when searching for images via Google/Yandex if alternative search names or multi-query search are enabled.

Dubovoi: Intelgen 14.0.10 update.

What is fixed:

- OpenAI API key validation on the API tab has been corrected: the button now displays the validation progress and result without reloading the page and no longer depends on the jQuery alias.

- An unnecessary background keep-alive request to dubovoi_intelgen_ajax.php when simply opening the API tab has been removed; background scripts continue to use keep-alive only where it is truly needed.

- License status display on the settings page has been corrected: the request has been brought back to the working format of previous versions and should no longer show the false message 'Unable to determine license status'.

- The 'Safe Launch' block has been renamed to 'Module Setup Wizard' to avoid tying the text to a fixed number of steps.

- The 'Current Configuration Summary' block and the 'Module Setup Wizard' block have been separated into two columns of the settings table: the left block no longer remains empty.

- The 'Main Risk' text in the 'Current Configuration Summary' block has been rewritten: it now takes into account the actual number of web calls, context depth, inclusion of characteristics, price search, and image mode.

- OpenAI image model parameters are shown only when the image generation mode via OpenAI is selected.

Dubovoi: Intelgen 14.0.9 update.

What is fixed:

- The OpenAI API key check on the API tab has been improved: the button immediately shows the checking process, the result is displayed without page reload, and the handler no longer depends on the jQuery alias.

- An unnecessary background keep-alive request to dubovoi_intelgen_ajax.php when simply opening the API tab has been removed; background scenarios continue to trigger keep-alive only where it is truly needed.

- The 'Summary of current configuration' block has been redesigned: it now accounts for the actual number of web calls, context depth, inclusion of characteristics, price search, and image mode, without contradicting current settings.

Dubovoi: Intelgen 14.0.8 update.

What is fixed:

- Fixed display of settings page tabs: the profile block no longer causes a 500 error due to a namespace call to CUtil, and profile links retain the active tab.

- Restored and pinned styles for the "Select in information block" button and the product filter form, including the old blue style for the choose_iblock button.

- Enhanced scrolling in the "Select in information block" window: scoped Bitrix popup content receives internal scroll and layout/scrollbar properties without globally modifying all popup-window-content.

- Fixed calculation of OpenAI write-offs: citations and search results are no longer counted as separate web-search tool calls, so GPT-5 costs should not be inflated multiple times.

- Fixed AJAX check for the OpenAI API key and added protection against checks without saved proxy settings.

- Clarified labels for the OpenAI API key for text, search, and images.

Dubovoi: Intelgen 14.0.7 update.

What is fixed:

- Fixed display of settings page tabs: the profile block no longer causes a 500 error due to a namespace call to CUtil, and profile links retain the active tab.

- Restored and pinned styles for the "Select in information block" button and the product filter form, including the old blue style for the choose_iblock button.

- Enhanced scrolling in the "Select in information block" window: scoped Bitrix popup content receives internal scroll and layout/scrollbar properties without globally modifying all popup-window-content.

Module UX/UI improvements

Module setup wizard improved: scenarios are divided into independent and complex tasks, settings are displayed according to the selected task, and the launch guides the user sequentially from the working mode and catalog to services and the first launch.

What has been improved:

- A separate module configuration wizard has been added to help select a task and sequentially configure the catalog, AI services, search, or image generation, as well as perform a secure initial launch.

- Documentation for main scenarios has been expanded.

- The section properties page has been improved: section selection, statuses, and working with catalog characteristics are now clearer and more convenient.

- The wizard logic for selecting image sources has been refined: the wizard now displays only current user configuration options.

Why it is important to update:

- The new wizard simplifies the initial launch and reduces the risk of enabling unnecessary functions without verification.

- Documentation helps quickly understand which keys, services, and modes are needed for a specific task.

- Configuring characteristics by section has become clearer for large catalogs and helps prevent cluttering the information block with unnecessary properties.

After installation:

- Open the configuration wizard, test the required scenario on 1-3 items, and only then start mass processing.

- If using characteristics, check the section properties page and applied characteristic sets before mass saving.

The functionality for generating properties in sections, based on which product characteristics can subsequently be generated, has been improved. The ability to generate a list of properties at the level of any section has been implemented. Working with section characteristics becomes more intuitive when configuring large catalogs.

Module security has been improved and potential vulnerabilities have been eliminated.

Mandatory security updates for the module.

Update to Dubovoi: Intelgen 14.0.0.

What has changed:

- Support for Russian neural networks based on Yandex (image search, text generation) has been added.

- The ability to generate images via OpenAI has been implemented.

- A first-run wizard has been added, which shows the readiness of settings for safe testing and mass launch.

- Built-in documentation and a section on working with section properties have appeared in the admin panel.

- The properties section now helps prepare a controlled list of characteristics by section: based on already filled products, with AI suggestions for empty sections, matching infoblock properties, and linking to sections.

A bug has been fixed where product identification in the background was incorrect if the name used apostrophes.

A bug has been fixed when filtering products if the product name contains an apostrophe.

General improvement of module operation

Update of module hints and language files

Update 13.0.0

Significantly improved image search via neural networks.

Added the ability to specify image search sources at the catalog section level, which allows saving budget on neural networks.

Added automatic extension of the administrative session during long-term work with generation and background results.

Added detailed accounting of write-offs by operations and external services with saving records in the module.

Added support for bypassing certain protected sites via ZenRows, including sites with Cloudflare.

Full list of features and improvements:

- Added display of write-offs by operations and external services in the "Write-offs" section.

- Added integration with ZenRows to bypass external blocks when searching for product pages and images.

- Implemented step-by-step image search via neural networks with continuation of the background task and display of intermediate progress.

- Product page search has been switched to using the domain sitemap instead of brute-forcing non-existent URLs in precheck.

- Fixed the fallback page search mode: when restrictions are relaxed, requests are rebuilt without the site filter.

- Improved selection of the relevant product card: strengthened checks for strong model tokens and brand, reducing the number of false matches.

- Improved parsing of galleries on sites with slick, swiper, owl, and fancybox markup, and added support for additional image paths in Bitrix structures.

- Fixed an infinite loop of re-scanning thumbnails and thumbnail containers during image search.

- Added support for bypassing Cloudflare and other restrictions for reading product cards and loading images via the browser access provider.

- Implemented a live progress indicator for background tasks, displaying the current phase and partial completion.

- Restored the logic for searching or creating a new product when saving images if the product does not yet exist in the catalog.

- Unified image saving for the module's main page, background generation page, and automatic background application.

- Improved Photoroom Pro operation: when a temporary local file is available, processing is now performed on the local file instead of the external donor site URL.

- Added fallback scenarios for saving original images if Photoroom does not return a processed result.

- Expanded logging for Photoroom and OpenAI Pages to diagnose image sources, positioning parameters, and background processing stages.

- Fixed interface messages and operation mode tooltips: the text regarding the background generation page is no longer displayed in synchronous mode.

- Updated Beta mode interface tooltips with an explanation of reasons for empty results when searching for images on specific sites.

Technical fixes:

- Adjusted image source priority: temporary local files and temporary URLs are now used before external donor URLs.

- Added support for local imageFile submission to the Photoroom API Plus.

- Added additional vertical positioning parameters for images in Photoroom for complex cases.

- Added additional diagnostic logs for internal unauthorized calls and Photoroom parameters.

Video demonstration of image search via neural networks with new capabilities

Improved image search via neural networks

Image search via neural networks has been moved to a step-by-step background mode to prevent heavy tasks from hitting server timeouts. Display of phases and progress for the background image search has been added to the results table. Product card selection has been tightened and only images are extracted from the gallery to reduce the risk of irrelevant images appearing.

Mandatory security update for the module. Fixes potential module vulnerabilities.

Module protection against hacking has been strengthened; mandatory update for security.

Module protection against hacking has been strengthened; mandatory update for security.

Improved product name search by article number.

Macro support implemented in background generation

Error handling in background generation improved

Macro support implemented in background generation

Background generation of the module improved

Background generation of the module improved

Automatic reconnection to MySQL added upon connection breaks

Result output in background generation mode has been improved - automatic removal of unnecessary quotation marks and escape characters has been implemented

Module installer has been improved

Major update!

Brief announcement:

Background mode for the module has been implemented (Beta version)

Module security has been improved

Image generation via neural networks has been improved (multi-layer image search)

__

New working mode — "background generation"

The module can now operate fully autonomously. Generation is launched in the background, results are saved to the database and are available for review at any time.

This is convenient when:

- generation takes a long time

- a large volume of products needs to be processed securely

- subsequent moderation before publication is important

What this provides:

- results are not lost even when the page is reloaded or when exiting the admin panel;

- results are not lost when errors occur;

- you can view the history, edit and approve results;

- a convenient moderation system directly in the interface.

What background generation can do:

Autonomous operation without browser dependency

Generation continues even if you close the page or exit the admin panel.

Saving results to the database

All responses are saved in the system; they can be opened and checked at any time.

Moderation and editing before application

Received texts and images can be corrected manually and only then saved to the product.

Separate tasks for text and images

If both options are selected, independent tasks are created — you can control each result separately.

Convenient list of all background tasks

Statuses, errors, date, results — everything is available in one interface.

Restarting tasks

Any task can be restarted without a full rebuild.

Automatic pause when errors occur

If any error occurs, the process stops for correction; the launch can be repeated after making changes.

Manual pause

The process can be stopped at any time and resumed later.

Clear progress control

The number of created and completed tasks is displayed.

Convenient filter for working with specific products

Filter results on the page by name, section, generation date, status, or generation type (text/images)

A task status system has been implemented

All, in queue, in progress, paused, ready, saved, moderated, error

A convenient data moderation system

Edit text or change images; data is automatically saved to the database

The result of each task can be quickly set to the "Moderated" status to view already processed tasks

Additional tools for working with tasks

You can select specific tasks in bulk and restart generation or delete them from the database

Bulk data saving

After moderation, you can run bulk saving of text or images on the page (for multiple items at once)

__

Module security has been improved

Key protection measures have been implemented:

Internal token (internal_token)

Automatically created and used for all internal requests, protecting service calls.

Callback endpoint protection

Public endpoints are no longer accessible without a token or user rights — protected from external access.

SSRF protection during image upload

localhost and private/reserved IP addresses are blocked to prevent access to internal server resources.

Strict SSL verification

Disabling TLS validation in admin checks has been removed — data transmission is now strictly over a secure connection.

Secure data processing

Risks of injecting malicious objects during processing of saved settings and properties have been eliminated.

__

Image generation via neural networks has been improved

A multi-layered image search system via Google + OpenAI + parsing has been implemented

The 'Multi-layered Search' mode first finds product pages via Google (not images, but actual product pages / cards). Then the module opens the found pages and extracts images from galleries. If there is no exact card among the URLs, OpenAI is connected: the neural network selects the correct product page and helps filter out irrelevant sources. Thanks to this, images are found even on sites with complex structures. The function has been tested on several hundred products from various categories; we continue to improve the smart image search.

Module Improvements

A permanent button to check the API key and proxy server has been added. Recently, there have been frequent cases of proxy blocking by internet service providers. Be sure to check the proxy before working.

Error handling in the module has been improved.

A new product identification option has been added: Name + Brand. This is used when the site has duplicate products by name but from different brands. In this case, product search and identification will be performed using the product name + brand combination. If the brand is not filled in for a product, generation will not be performed. The allowed property type is string, non-multiple list, dictionary, or link to elements of another information block. If you are uploading products from EXCEL, the brand must be filled in column 7 (G).

In the EXCEL file, the first row can now contain headers for convenience.

A new option "Do not process the first row when uploading EXCEL" has appeared in the module settings.

Example of an actual EXCEL file for upload

Mass processing of products in the module has been improved.

Error handling has been improved.

The module's operation has been optimized to improve query execution on virtual hosting.

A new operating mode has been added in the module settings - sequential query execution. Now, each subsequent query is launched only after the previous one completes. This allows more efficient query processing without the need for manual timeout management. If any error occurs, the query is automatically retried; if the error repeats, generation stops to allow adjustment of the module settings. If an empty response is received from the neural network, the module automatically increases the maximum number of tokens in the settings by 1000 and retries the query.

To switch the module to the new operating mode, select the value "Sequential" in the module settings under the "Mass generation mode" parameter.

The module's behavior has been improved for cases where the neural network returns invalid HTML markup with unclosed tags and/or "cut off" text. This is related to token shortage during query execution. In such cases, the module will automatically increase tokens by 1000 and retry the query.

The logs now record the budget spent on the neural network for each query.

Command operation has been optimized when using profiles.

Extended module operation logging has been implemented for rapid diagnosis of emerging issues.

Improved profile handling

Added the ability to change the module's predefined commands

Update is mandatory for installation - module language files used in commands have been updated

Minor improvements in module operation

Image Search via Neural Network with Maximum Relevance (Beta)

The module offers a new way to search for product images — via a neural network with intelligent page analysis.

The OpenAI + page parsing function operates in beta mode. We continue to improve algorithms and result quality.

How the search works

Query Generation

The module generates a set of keywords based on the product name, considering possible variations on different websites.

Search for Relevant Pages

The neural network selects the most suitable pages containing the product.

Parsing and Image Extraction

After identifying candidates, the module analyzes the pages and extracts product images.

Search Priority by Source Websites

If the module settings specify search sources (e.g., manufacturer or official distributor domain names), the module first searches for the product via sitemap.xml of the specified websites.

In this scenario, the neural network is used to generate keywords, as product names on the website and in the catalog may differ.

If the product page is not found on the specified website or images cannot be extracted, an alternative search via the neural network is initiated (this option can be disabled).

Cost and Usage Features

This is the most expensive but most relevant method for searching images.

The actual cost of API usage depends on how easily a product from your catalog can be found on the web. If you use common product names that frequently appear online, API costs will be lower.

The price depends on the maximum number of neural network search calls configured in the module settings. If no search results are found, it is recommended to increase this parameter to locate more websites and, consequently, images of the required product.

On the Logging tab, you can see the actual cost of the executed request, for example: Success OpenAI+Parser API cost: $0.0690.

How to Reduce API Costs

To minimize expenses, we recommend specifying the websites of manufacturers in the module settings.

Specify one or several domains where the sought-after product is likely to be located (for example, products of the same brand).

At the time of testing the first version of the function, the average cost of image requests was:

with the manufacturer's domain specified: $0.01

without the manufacturer's domain specified: $0.07

The cost may vary depending on the module settings and the complexity of the search.

Optimized JS script performance in the module.

Fixed an error where bulleted lists were incorrectly removed from descriptions when multiple lists existed on the page (during property block identification).

Improved the function for saving product characteristics into a catalog HTML property as a single block. The module now supports saving all product characteristics at once into a catalog property of HTML type. If the function for parsing and saving individual characteristics into properties is disabled, you can specify a CSS class for any block containing the product characteristics. Previously, the module only supported working with bulleted lists (ul). For example, if you want product characteristics to be saved into a separate HTML property based on a specific HTML structure (which does not contain a ul list), you can define an arbitrary HTML structure for such a block in the commands and save it into a separate HTML property. This way, characteristics are removed from the product description (if necessary) and saved into a separate property with an arbitrary structure.

Added an exception for links in text; now microdata links (schema.org) are allowed.

Attention! After installing module version 8.2.0, all created profiles must be saved again.

The module's UI part has been improved. Working with the module is now more intuitive and logical. Many parameters in the module are automated.

The ability to run individual module functions has been implemented, namely: generation of characteristics, search for the average market price of a product, generation of product descriptions, and generation of product images. In the new version of the module, all these functions are independent and can work separately. For example, you can run the function to search for and fill in product characteristics without filling in the description.

The function for searching the average market price of a product has been improved:

- sources for price search can be specified

- websites with prices can be searched automatically

- a method for calculating the average price has been added (arithmetic mean, median, trimmed mean)

- the ability to add a markup to the average price has been implemented

- the ability to round the price has been implemented

Supported price calculation methods:

Arithmetic mean - sum all prices and divide by their count. If there is a very expensive or very cheap price, it will significantly distort the result. When is it suitable? When prices are approximately the same and there are no sharp outliers. Or when all possible prices in the market need to be taken into account.

Median - the price "in the middle of the market". How is it calculated? Sort prices from lowest to highest. Take the price from the middle of the list. If the number of prices is even, take the average of the two central ones. With this calculation method, expensive and cheap outliers have almost no effect. To understand the real market price, as "on average for how much it is sold".

Trimmed mean - the average value without extremes. How is it calculated? Remove obviously outlying prices (too expensive or too cheap). Calculate the average from the remaining ones. Combines the logic of the mean and the median - the result is stable and realistic. For analytics, pricing, marketplaces, automatic calculations.

Now the function for searching the average price runs only separately and cannot work simultaneously with other functions to improve the quality of price search.

Image search via SerpAPI (using the "Bing" engine) has been implemented as an alternative to Google search. Currently, the function is inactive; its launch is expected in future updates.

Improved image search

Multi-threaded image search for products with different names has been implemented. Now you can launch multiple simultaneous image search requests at once, with each request using a different product name (the module finds names via neural networks on the web). This allows obtaining a more relevant and extensive set of images at once.

A function to change the product name when loading the next batch of images via the "Show More" button has been implemented. Thus, if the initial image search results do not meet your needs, the module will search for images with a new product name when executing new requests. This allows maximizing the diversity of search results and finding suitable product images. Additionally, a new working mode with the module has been implemented, which allows quickly selecting the necessary images without the need to delete unnecessary ones.

New module settings options:

- When clicking "Show More", search for images with an alternative product name

- Perform multiple simultaneous image search requests with different product names

- Number of simultaneous image search requests

For example, if multi-threading is enabled in the module settings and 3 simultaneous requests are selected, the module will search for images using the original product name and two of its variations (found via neural networks).

Improved image handling mechanism

An additional image handling option has been implemented. Now you can select specific images to save by clicking on them. Previously, you had to delete all unnecessary images on the page to keep only the required ones. The module supports both image handling modes.

The module settings include an "Image Save Mode" option with two modes:

- Save all images displayed on the page

- Save only selected images

Improved module settings handling

Now, in all fields with commands and prompts, you can use any characters, including double quotes, commas, and HTML tags, which significantly simplifies configuring the module for specific tasks.

Image search has been improved.

If no image search results are found, the module can now automatically use a neural network to select an alternative product name and initiate a re-search for images (up to 3 possible search iterations). The alternative product name can be saved in the information block property.

New settings have been added to the module configuration:

- If images are not found, search for an alternative name via a neural network

- Search for an alternative name until the specified number of images is found

- Minimum number of images for an alternative name

Video presentation of the new function's capabilities

Added the ability to select the OpenAI model: GPT-5, gpt-4o-search-preview, GPT-4o-mini, GPT-5-mini, GPT-5-nano

Added the ability to save the product description to an arbitrary HTML property

Added the ability to save product properties as a bulleted list in an arbitrary HTML property

In the module settings, you can choose where to save the product description, and you can also check a box to save product characteristics to a property

Implemented a mechanism for automatic retry of requests during temporary failures (30 seconds, up to 3 attempts per product) without queue freezing.

Improved stability of mass generation: the queue does not block on temporary errors.

Improved handling of server errors.

Adapted module operation for virtual hosting (shared).

Improved property handling; fixed errors in incorrect identification of numeric properties.

Implemented automatic conversion of the Cyrillic letter yo to the Cyrillic letter ye in property names and values to avoid potential duplicates.

Implemented additional options in module settings:

- Show generated properties in the smart filter

- Show generated properties in the product card

Implemented normalization of hyphens in property names and values.

Implemented the ability to save empty text in the product description (saving a product with empty text removes the product description).

Implemented a prohibition on saving the product description if it contains a neural network clarification (when no generation results are available).

Added the ability to search for images by property value (by default, the image search query is performed by product name or article; the new function allows taking the prompt from a property and expanding the image search query).

The ability to configure module access for different user groups has been added.

Note! In addition to module access settings, you must grant the user group rights to edit the information block with products, as well as provide access to the /bitrix/admin/ directory - this is configured outside the module.

Improved processing of numeric properties.

Added the ability to immediately open the result for all blocks during mass generation (new setting parameter "Expand all blocks with generation results").

The module settings now allow enabling or disabling the "reverse inclusion" function for properties during identification. The module includes the function "Enable strict property search by name". When this option is enabled, the module searches for properties by exact name match. When disabled, the module searches for similar properties if no property is found with an exact name match.

Property identification by inclusion

For example, if a neural network generates the property "Lamp Power", while the information block has the property "Lamp Power, W". When the option is disabled, the module identifies this property and fills it, since the information block does not have a property named "Lamp Power", but has a property "Lamp Power, W" which contains the generated property name.

Property identification by reverse inclusion

The reverse inclusion function works oppositely. If the module generates the property "Lamp Power, W", but the catalog only has the property "Power", the module will fill this property, since the word "Power" is present in the name generated by the module. This function is useful in many catalogs, however, with a certain data structure, the filling logic may be violated; use the option carefully and with mandatory data verification.

Full scheme for finding properties on the site:

1) Search for an exact match of the property name, for example, if the module generates the property "Lamp Power, W", the first step is to search for a property with the same name.

2) Matching by cleaned key: the module will search for the "Lamp Power" property instead of "Lamp Power, W".

3) Matching by direct name entry

4) Matching by reverse entry (when the option is enabled)

Improved processing of numeric properties; a "concatenated numbers" filter has been implemented.

The module can identify numeric properties even if the value does not start with a number. For example, the property "Power: up to 50 kW" is identified as numeric and will save the value 50. A property like "Face layers: 16K" will not save the value "16" even if a numeric property with the same name is found in the catalog. The module includes a concatenated numbers filter: if letters are attached after digits that are not units of measurement from the whitelist, the property is not considered numeric.

Processing of numeric properties with ranges has been added. A new parameter in the module settings is "Processing of numeric properties with ranges". Two options are available:

- Save the upper limit of the range

- Save the lower limit of the range

For example, the module generated the property "Heated room area: 50-70 sq.m.". If a numeric property with this name is already defined in the catalog, the module will save either 50 or 70 depending on which range limit is selected in the module settings. If the property in the catalog is defined as a "String" type, the module will save the entire range. If no property with this name is defined, the module will create a property of type "String" and fill it with the entire range.

Note! You can adjust the format of generated property values via prompts in the module settings. For example, you can prohibit the generation of values with ranges for all properties or only for specific properties.

Improved identification of existing properties "by inclusion".

The mechanism for identifying existing catalog properties by inclusion has been refined, which improves matching accuracy even when full names do not match.

If the strict search option is disabled, the module performs intelligent property selection based on partial keyword matches.

Examples of the updated algorithm in action:

Generated: "Lamp power"

An information block contains the property "Lamp power, W" — a correct match after removing the unit of measurement.

Generated: "Case Material"

In the information block: "Material" — found via reverse keyword inclusion.

Generated: "Height, mm"

In the information block: "Height" — match after trimming measurement units.

Generated: "Tank Volume"

In the information block: "Volume" — found via partial match of the base term.

First, the module searches for the property in the catalog by exact name match. If the property is not found, a search by inclusion is performed (unless the "Strict property name search" option is enabled in the module settings).

Full algorithm for searching properties in the catalog:

1) Search for exact property name match; for example, if the module generated the property "Lamp Power, W", the first step is to search for a property with the same name.

2) Search for match by cleaned key; the module will search for the property "Lamp Power" instead of "Lamp Power, W".

3) Search for name by inclusion (direct or reverse).

User interface improved — added a button for quick saving of the generation result

Significant improvement in product attribute processing

New options and capabilities have been added:

- It is now possible to restrict the generation of attribute values only to those values already specified in properties of the list or catalog type. Before generation, the module collects all possible property values in the information block and adds them to the prompt. Thus, the module will select the required value from the already specified list, solving the problem of creating duplicate values.

In this case, properties that should not be added to the prompt (for example, system properties) can be specified.

- Implemented the ability to prohibit filling a property if the value is already set; when the "overwrite property" option is enabled, the property will not be overwritten.

- Creation of numeric properties with units of measurement. For example, the string "Length: 38 mm" is automatically converted upon saving to the property "Length, mm" and filled with the number 38. If a property named "Length" is already specified in the information block, the name will not change.

- Improved identification of multiple properties using various separators (comma, semicolon, slash).

- Improved processing and identification of various properties in different formats.

- The module now always checks the type of existing properties and adjusts the values to match this type.

Image processing has been improved.

The ability to save the preview image and the detailed image as separate properties (including in WEBP format) has been added.

The ability to select the image format (WEBP, PNG, JPG) has been implemented. When attempting to save WEBP images to standard fields (preview image and detailed image), the found image is automatically converted to a format supported by the site. This eliminates possible errors during saving.

Automatic image resizing functionality has been implemented within the module itself (in settings, you can select the maximum width and height of the image). The settings option is "Proportionally crop found images".

Automatic renaming of files has been implemented; Cyrillic characters, spaces, and other prohibited characters that may prevent the image from being saved on the site are automatically removed from the filename.

Automatic deletion of broken images has been implemented.

Forced deletion of obvious duplicate images has been implemented.

The ability to start generation simultaneously on all pages for all selected products has been implemented. A new option has been added in settings: "Automatic transition to the next page" (Continue mass generation on the next page without reloading until products run out). Without enabling this option, by default the module processes products only within the 1 page. We recommend running generation no more than 1 time(s) for no more than 200–300 products (across all pages), otherwise there is a risk of losing generated data if browser memory runs out or an error occurs on the page. To process a large number of products in 1 time(s), server-side storage of generation results must be used (the function will be available in future updates).

Minor console errors in the module have been fixed.

Image search has been optimized, and automatic removal of broken images has been implemented.

Functionality for the new Photoroom API has been updated.

A bug with the progress loading indicator has been fixed.

A bug causing an extra display of the "Show next items..." button when pagination is absent has been fixed.

Fixed a bug where the announcement text was being overwritten with zero when the function for generating the announcement description in items was disabled.

Improved profile functionality with module settings

Minor module operation improvements