---
name: visual-source-identification
description: Identify the exact source/provenance of an image or visual artifact from a user-provided crop, screenshot, scan, or low-context image. Use for art/manuscript provenance, historical illustrations, screenshots of unknown origin, watermarked/overlaid images, and requests like “find exactly where this comes from.”
---

# Visual Source Identification

Use this skill when Alex asks where an image comes from, what it depicts, or requests exact provenance for a visual artifact. The goal is not just a plausible subject label; it is a **source-backed identification** with repository/object IDs, page/folio/location, direct image links, and concise context.

## Workflow

1. **Inspect the image first**
   - Use vision on the provided local file/URL.
   - Explicitly ask vision to ignore modern UI overlays/watermarks when the user says to disregard them.
   - Extract search handles: visible text/rubrics, labels, script style, iconography, colors, creature/action, margins, page layout, and any crop boundaries.
   - Distinguish literal depiction from tempting modern interpretations (e.g. “rider” may actually be an animal attacking a beast).

2. **Search by both text and iconography**
   - Combine likely subject terms with visual features: `"basilisk" "weasel" "biting its neck" manuscript`, not only generic labels like `dragon rider`.
   - Search exact visible phrases/rubrics when available.
   - Try alternate historical names/synonyms: basilisk/cockatrice, bestiary/medieval manuscript, folio/f., shelfmark/MS.

3. **Prefer institutional/catalog sources over reposts**
   - Reposts (Tumblr, Pinterest, X, blogs) are useful leads but not final evidence.
   - Verify against museum/library/catalog entries where possible.
   - For manuscripts, look for: repository, shelfmark, folio, date, origin, language, catalog description, and IIIF manifest/canvas/image-service links.

4. **Use IIIF manifests for exact page verification**
   - When a digital manuscript/object has IIIF, fetch the manifest and locate the relevant canvas/page/folio.
   - Verify the exact folio/page label and direct image service.
   - If a search result says “fol. 79r,” confirm the manifest canvas label and image service ID before finalizing.
   - Direct image URLs can often be formed from IIIF image service IDs, e.g. `/full/1200,/0/default.jpg` for a shareable view.

5. **Return a concise but complete provenance answer**
   - State the identification first.
   - Include source links and exact object identifiers.
   - Explain what the image depicts and correct common misreadings.
   - Include rights/licensing if known and relevant.

## Output template

- **Identification:** [object/scene title]
- **Source:** [repository, collection/manuscript/object]
- **Exact location:** [folio/page/canvas/object ID]
- **Date/origin/language/material:** [if available]
- **What it depicts:** [literal scene]
- **Source links:**
  - Catalog/object page
  - IIIF manifest if available
  - Exact canvas/page if available
  - Direct image service/view if available
- **Notes:** [short context + any correction of likely misinterpretation]

## Pitfalls

- Do not stop at “it looks like X.” The task is exact provenance.
- Do not trust social repost captions unless confirmed by institutional metadata.
- Do not over-index on a cropped image’s apparent subject: a “knight/rider” can be an animal, demon, child, marginal figure, etc.
- Do not omit the exact folio/page/object ID when it can be found.
- If the source is a manuscript, “the manuscript” is not exact enough; provide the folio/canvas and direct image reference.

## References

- `references/ashmole-bestiary-basilisk-weasel.md` — worked example: identifying a cropped medieval bestiary image as Bodleian Library MS. Ashmole 1511, fol. 79r, basilisk attacked by a weasel.