Lens

• Fixed images failing to resize before upload on servers whose ImageMagick build uses more memory per pixel. The image memory limit now follows the server's free memory, up to 384 MB, where it was previously fixed at 64 MB. (#3)
• Images in a format the provider does not accept, such as AVIF, now fail with a clear message when they cannot be converted. Previously the original was sent and the provider returned an HTTP 400 error.
• Raised the Claude upload limit from 3 MB to 7 MB to match Claude's current 10 MB limit for encoded images.
• Image decode failures now log the ImageMagick version, its area limit, and the server's available memory.

• Added a Custom Instructions setting for site context and style guide rules. They are applied to every analysis, including translations. (#2)
• The AI is now told to only mention objects, people, text, and details that are visible in the image.
• Long descriptions now open directly with what is happening and no longer list what is missing from the image.
• Added the Claude Sonnet 5 and Opus 5.5 models alongside the existing Claude 4 models. New installs default to Sonnet 5; existing sites keep whatever model they are already set to.
• The Claude and Gemini model menus are now grouped by generation, like the OpenAI one.
• Added the Gemini 3.8 Flash, 3.1 Pro, and 3.5 Flash Lite models alongside the existing Gemini 2.5 models. New installs default to Gemini 3.8 Flash; existing sites keep whatever model they are already set to.
• Gemini costs now include thinking tokens, which Gemini bills as output.

• Added the GPT-5.6 OpenAI models (Sol, Terra, Luna) alongside the existing GPT-5.4 models, and refreshed pricing for both. New installs default to GPT-5.6 Luna; existing sites keep whatever model they are already set to.
• The OpenAI model menu is now grouped by generation, and the GPT-5.4 Mini option shows its correct price.

• Fixed Image Assessment failures on some ImageMagick 7 servers.
• Image Assessment is now optional and disabled by default; enable it on servers with Imagick and sufficient memory.
• Reduced image-analysis memory usage, removed low-resolution upload substitutions, and added recovery for interrupted queue jobs.
• Duplicate hashing is now Imagick-only. After updating, run php craft lens/duplicates/rebuild-hashes once to refresh existing hashes.
• Fixed bulk completion status and limited failed-asset retries to the selected volume.

• First stable release.

• Fixed Image Assessment section disappearing on ImageMagick 7 builds.
• Database logging is now opt-in via the LENS_LOGS env variable, with richer diagnostic context on each entry.
• Old Lens log entries are now auto-deleted daily. Retention is configurable on the Settings page (default 15 days).
• Disabled-volume notice on the asset edit page now offers a shortcut to edit the volume's field layout alongside the link to Lens settings.
• Refreshed the setup banner with clearer copy and direct links to each volume's field layout.

• Fixed Image Assessment section not appearing on some Imagick builds

• Recalibrated sharpness scoring: replaced the hard-cliff floor and worst-patch penalty with an absolute-energy ramp and best-patch rescue for more accurate scores on textured and partially soft images
• Suppressed Craft's "asset has been updated" reload banner after Lens apply and inline-edit actions
• Focal point editor now shows the asset's current focal point alongside the AI suggestion when they differ
• Fixed the confidence badge disappearing when reverting a detection toggle to its AI value

• Improved sharpness scoring accuracy; brightness and contrast now run on a downscaled working buffer for speed without changing scores
• Cancelling a bulk run now clears Pending records so cancelled assets immediately show "ready for analysis" again
• UI polish: removed the dead Bulk Concurrency setting from settings

• Per-asset wall time roughly halved: one job per asset, single shared image decode for metrics and AI preprocessing
• Added Sync Pro Metadata backfill for filling Pro-only fields on Lite-analyzed assets, with progress and cost tracking
• Duplicate detection now surfaces as a dashboard attention card and a dedicated Pro asset source
• Stability and UI polish

• Retired the custom Asset Browser in favor of the native Craft asset index, with Lens filters, quick-access sources (Needs Review, Not Analysed, Missing Alt Text, NSFW Flagged, etc.), sort options, and table columns merged into the native experience
• Added image preprocessing before AI upload (resize, EXIF stripping, decode-cost guard) to shrink payloads for faster requests and stay under provider upload limits (notably Claude's ~5MB cap)
• Search now works across every configured site's base language, so a query in any language hits assets analyzed in any other language
• AI provider calls now retry on connection timeouts and 500/504 responses for better resilience

• Onboarding overhaul: explicit volume opt-in, dashboard checklist with locked prerequisites and severity-tinted lock icons, per-asset volume-not-enabled panel state, reprocess-controller guard
• Image-quality rework: histogram-based too-dark/too-bright verdict with shadow/highlight clipping, removed overallQualityScore aggregate and Low Quality filter, replaced Web Readiness umbrella with standalone File Too Large rule
• OCR as regions: extractedTextAi stores an array of region objects instead of a single flattened string
• Various UI and stability fixes

• Cancel in-flight analyses from the asset page or Queue Manager with checkpoint-based cancellation
• Rewritten quality engine: blur-decay sharpness, percentile-spread contrast, improved compression analysis
• Updated AI providers to GPT-5.4 and Claude 4.6 model families with current pricing
• Simplified safety detection UI and refined watermark and brand detection prompts
• Quality of life updates

• Search index now includes AI original values, site translations, and language-specific stemming
• "Analyze with Lens" element action replaces "Find Duplicates" (supports re-analysis)
• Mobile-first setup banner, improved onboarding flow
• Auto-apply AI metadata to all sites after analysis
• Alt proxy field styled as native field (blue accent)
• Various UI and stability fixes

Added

• Initial alpha release of Lens, an AI-powered image analysis plugin for Craft CMS.
• AI image analysis: alt text, titles, descriptions, tags, OCR, quality scoring, and focal point detection.
• Content detection: people/faces, NSFW scoring, watermarks, brand logos, and stock photo identification.
• Duplicate detection via perceptual hashing with similarity scores.
• Three AI providers: OpenAI GPT, Google Gemini, and Anthropic Claude.
• Control Panel: Dashboard, Asset Browser with filters and search, Review Queue (Focus/Bulk modes), and Bulk Processing.
• Analysis Panel in the asset editor with inline editing and "Revert to AI" support.
• Configurable review-before-apply workflow.
• Automatic processing on asset upload and file replace.
• Console commands for batch processing, duplicate scanning, and statistics.
• Asset query extensions and condition rules for filtering by analysis data.
• Twig variable for accessing analysis data in templates.