Check AI Images for Extra Fingers & Duplicates

How it works

AI Image Set QA Checklist is for the moment a batch of AI-generated images lands in your folder and you have to decide whether to accept it, pay for it, or send it back. It runs the same six checks on every image, then looks across the whole set for repeated expressions and scenes. It never edits, crops or uploads anything.

Phase 1 — Intake. Whether it can actually view the images or only read your descriptions, the brief, the file list and sizes, and which image is the reference for "same character".

Phase 2 — Six checks per image. Arms/hands/fingers, face and artifact integrity, watermark/stray text/crop, same-character consistency, fit to the brief, and safety flags (apparent-minor look, possible real-person lookalike as an unverified suspect flag only). Each is PASS, REWORK or UNSURE with a short evidence phrase.

Phase 3 — Set-level checks. Duplicate expressions, poses and scenes, delivered count versus expected, mixed sizes.

Phase 4 — Rework list. Only the failing images, one line each, plus a "keep as is" list and the questions a human still has to answer.

Sample output

Real, unedited output from running this Skill on a sample problem:

Mode D — 8 images — PASS 0 / REWORK 4 / UNSURE 4 — distinct 7

## Brief and set
- Mode: D (described). This run has no image viewing tool, so only what the manifest states was judged. Nothing below was seen.
- Brief (quoted): "8 images of one fictional character, 'Mina', a 26-year-old cafe owner, semi-realistic webtoon style, vertical portraits... 8 different scenes with different expressions, same face and hair (shoulder-length chestnut hair, round glasses) in every image, no text or logos, adult look only." Expected size: 1024x1536.
- Set: 8 files, mina_01.png to mina_08.png, numbering complete, no extras, all PNG. Sizes stated: 1024x1536 for all except mina_07 (1024x1024).
- Anchor image for consistency: mina_01.png (the only image whose description states the full face/hair/glasses set; the requester says it "looks fine"). The anchor is itself unverified.

## Per-image results
| File | C1 hands | C2 face | C3 mark/text/crop | C4 same char | C5 brief | C6 safety | Verdict | Evidence |
|---|---|---|---|---|---|---|---|---|
| mina_01.png | UNSURE | UNSURE | UNSURE | n/a (anchor) | PASS | UNSURE | UNSURE | C1: "holding a coffee cup in both hands", finger count not described. C2/C3: face and corner marks not described (size 1024x1536 matches). |
| mina_02.png | UNSURE | UNSURE | UNSURE | UNSURE | PASS | UNSURE | UNSURE | C1: requester note "right hand, waving, seems to have six fingers - not sure", unverified. C4: chestnut hair and round glasses stated, hair length not stated. |
| mina_03.png | UNSURE | UNSURE | UNSURE | PASS | PASS | UNSURE | UNSURE | C1: "book hides them", hands not visible or not described. C4: "chestnut and shoulder-length", "glasses on". |
| mina_04.png | UNSURE | UNSURE | UNSURE | UNSURE | PASS | UNSURE | UNSURE | C1: cup in both hands, fingers not described. C4: hair length not stated. Duplicate of mina_01 (see set-level). |
| mina_05.png | UNSURE | UNSURE | UNSURE | REWORK | PASS | UNSURE | REWORK | C4: "dark brown, long, past the shoulders" and "NO glasses" vs. brief chestnut shoulder-length with round glasses; requester: "looks like a different person?". |
| mina_06.png | UNSURE | UNSURE | REWORK | UNSURE | PASS | UNSURE | REWORK | C3: "small sparkle-shaped mark... bottom-right corner", requester thinks it is the generator logo (the identification is theirs). C4: nothing stated about hair or glasses. |
| mina_07.png | UNSURE | UNSURE | REWORK | UNSURE | PASS | REWORK | REWORK | C3: 1024x1024, brief 1024x1536. C6: requester says "face looks very young, maybe teenager" while the brief requires adults only; flagged conservatively, not an age claim by me. C4: hair length not stated. |
| mina_08.png | UNSURE | UNSURE | REWORK | UNSURE | PASS | UNSURE | REWORK | C3: sign reads "CAFFE ANGLEO" with melted letters, brief says no text. C6: requester sees a possible resemblance to a famous actress, suspect flag, unverified, no name given. C4: nothing stated about hair or glasses. |

Cells that say UNSURE without a note in the Evidence column mean the manifest does not describe that aspect (for C6: apparent age and lookalike were not raised or described).

## Set-level findings
- Duplicate group 1: mina_01.png and mina_04.png. Both are "smiling warmly at the viewer, holding a coffee cup in both hands" (same expression and pose, both at a cafe; only the wall colour differs, as the requester notes). Keep mina_01 (the anchor); redo mina_04 with a different expression, pose and place. Count: 8 files, 7 distinct.
- Other expressions as described are distinct: laughing (02), calm neutral (03), surprised (05), mild smile looking sideways (06), crying (07), sad (08). Camera distance and outfit are only partly described, so no further groups could be tested ("not described").
- Count and order: 8 of 8 delivered, no gaps or extras.
- Uniformity: mina_07 is square while the other seven are portrait. Art style consistency cannot be judged from text.

## Rework list
1. mina_05.png | regenerate with chestnut shoulder-length hair and round glasses | C4
2. mina_06.png | remove or regenerate without the corner mark (e.g. leave room at the bottom and crop it, or regenerate) | C3
3. mina_07.png | regenerate at 1024x1536 and with clearly mature features (the brief says adult look only) | C3, C6
4. mina_08.png | regenerate without readable or melted sign text, or leave the sign blank | C3
5. mina_04.png | redo with a different expression, pose and place (duplicate of mina_01) | set-level

Keep as is: none confirmed. No image earned a PASS from description alone.

Needs a human look (one question each):
- mina_01, 03, 04, 05, 06, 07, 08: count the visible fingers and check eyes, teeth and glasses. Can you view them at full size, or describe hands and face?
- mina_02: how many fingers does the right waving hand show when zoomed in?
- mina_08: is the resemblance to a real person real? This is a suspect flag only; a reverse image search by you is the way to check. I will not guess a name.
- mina_07: does the face look clearly adult to you? If not, it is already on the rework list.
- mina_02, 04, 06, 07, 08: what are the hair length and glasses (C4)? The manifest does not say.

Tell me which rework items to turn into request text.

## Limits of this check
- Mode D: no image was viewed. Hand, finger, face-integrity and watermark checks could only use the manifest and the requester's notes. Requester notes are treated as claims, not verified facts.
- The four REWORK verdicts rest on explicit statements in the manifest; the four UNSURE verdicts are not passes and are not failures.
- Not checked: art style consistency, hidden metadata, legal or platform compliance, and any real-person identification.

Use cases

• Checking a freelancer's AI image delivery before you pay
• Catching extra fingers, melted signs and generator watermarks before publishing
• Verifying that a character looks like the same person across a whole set
• Finding repeated expressions or poses in a batch you are about to upload
• Turning a messy review into a short list of images to send back

FAQ

Can it see my images? Only if the agent you run it in has an image viewing tool. If it cannot, it says so, judges only what you describe, and marks everything else UNSURE instead of guessing. It will never make up findings for an image it has not seen.

Does it edit, crop or upload my files? No. It is read-only and keeps nothing after the conversation. No API key or hosting is needed.

Does it tell me who a face looks like? No. A possible real-person resemblance is only an unverified suspect flag for you to check; it never names anyone.

Does it guarantee legal or platform compliance? No. It is a visual delivery check, not legal, copyright or age certification.

What image types? Any set of images you can view or describe, from one folder or a list.