Batch-rename images and produce a mapping table

When to use

Images collected from a camera, a chat app, and design exports, named IMG_2034 and the like. You want content-based names, and you need to still know which original each one was.

Preparation

Host version

To fill (record the exact rc version when testing)

Model

To fill

Mode / permissions

Standard mode + modify-within-workspace is enough to start

Plugins required

None (add name & version after testing if used)

Input materials

Keep in a separate workspace; back up originals first

Full instruction

Process the images under images/, rename them by content, and produce a mapping table at rename-map.csv.

[Input]
- .jpg / .jpeg / .png / .webp under images/
- Naming: grouped by subject, as subject-index.ext (for example teapacking-01.jpg)

[Output]
1. Renamed files (copy rather than move; the original images/ directory must not change by a single byte)
2. rename-map.csv with columns: original name | new name | dimensions | what you actually saw | confidence (high/medium/low)

Requirements:
1. 'What you saw' must describe the actual content ('a tea cake on the left, 357g printed on the wrapper'), never just 'looks like tea'.
2. For medium or low confidence, do not assign a meaningful name; use unknown-index as a placeholder. I will rename those myself.
3. Dimensions are actual pixels (width times height), never estimated.
4. Unreadable images (corrupt, unsupported) keep a row with 'not processed' in the new-name column and the reason.
5. Report at the end: total images, renamed, placeholders, not processed. The total must equal the sum of the three.

Copy full instruction

Run log

Output

After a successful run, list files and screenshots here, and note whether originals changed.

Acceptance checklist

Tap to tick each item off as you verify.

• The original images/ directory hash is unchanged (not 'looks unchanged')
• Every row describes what was actually seen, with no 'looks like' filler
• Dimensions are measured pixels, not estimates
• Medium and low confidence rows use unknown- placeholders
• Total = renamed + placeholders + not processed

Failures & boundaries

• Copy rather than move is the key clause: a wrong rename combined with a move destroys traceability. Verify the source directory hash instead of trusting 'unchanged'
• The confidence column tends to come back all 'high'. That means the evidence column is vague — fix that column first
• Content judgement is inherently subjective. Supply the subject names yourself and let the model only assign them, or you get inconsistent naming
• Blocked on: Needs a set of images with real content, including a corrupt one, before 'what you actually saw' can be judged. Synthetic colour blocks turn confidence into guesswork.

Claim this recipe Back to recipes