What gets extracted
| Extracted field | Where it appears in the UI |
|---|---|
| Title | Image title in the gallery card and detail modal |
| Description | Description section in the detail modal |
| Tags (5–10) | Color-coded tag badges in the modal and table view |
| Colors (5 hex codes) | Clickable color swatches in the modal and color filter in table view |
| Visual style | Stored internally (e.g. 35mm Film, CGI, VHS) |
| Category | Category badge in the modal; used for gallery filtering |
| Group | Group badge (e.g. Film, Commercial, Music Video); used for filtering and AI prompting |
| Project name | Groups images into project rows in the gallery |
| Moodboard name | Sub-label within a project |
Model configuration
The VLM runs via OpenRouter. The model is configurable through theOPENROUTER_VLM_MODEL environment variable. The default model is:
When analysis runs
Direct uploads
Direct uploads
Analysis starts immediately after the image is uploaded and its Convex storage URL is available. The image enters
aiStatus: processing as soon as the upload mutation completes.Discord imports
Discord imports
Discord-imported images land in a
pending moderation queue with aiStatus: queued. Analysis does not run automatically. It is triggered only after you manually approve the image from the Upload → Discord queue.Re-analysis
Re-analysis
You can re-trigger analysis from the Edit Image modal using the Re-analyze button. This resets
aiStatus to processing and schedules a new VLM run. Any existing AI-generated fields are overwritten with the new results.AI status states
Images move through the followingaiStatus values:
| Status | Meaning |
|---|---|
queued | Waiting for approval (Discord imports only) |
processing | VLM analysis or variation generation is in progress |
completed | Analysis finished successfully |
failed | Analysis failed (e.g. missing API key, VLM error) |
processing state. Images older than 18 hours in processing are automatically moved to failed on the next page load.
Viewing results
Analysis results appear in the image detail modal. Open any image from the gallery to see its extracted metadata in the panel below the image.If you provided a title, description, or tags during upload, the VLM may overwrite them. To keep your manual values, edit the image after analysis completes or re-run analysis with your preferred values pre-filled.