Becomes very hard to verify once the face turns too far, and the report says the bindi disappears entirely, leaving the identity lock extremely weak.
What was measured
Identity preservation
How closely the generated face matches the reference image across scenes.
decisive for this rankingtransformation
This is the core of the task: the tool must keep the same character looking like the reference image across scenes. (3 of 3 judges)
What was given, what came back
Test input: Three-quarter face portrait · image
Input — what we sent
Three-quarter face portrait
Three-quarter face reference image with medium-dark skin, tight curly hair in an updo, bindi, and a floral dress. The partial angle and softer lighting make it a harder reference than the frontal portrait and are meant to stress identity consistency.
Why this input is hard
- · Identity preservation with partial face angle
- · Hair texture and updo retention
- · Skin tone fidelity
- · Reference-image difficulty under softer lighting
Output — unretouched
Also checked on this input — same tool, 4 other criteria
Accessory & detail retention✓ WorkedRetains the bindi and strong brows in the frontal output, while only slightly darkening skin tone relative to the reference.Expression accuracy✓ WorkedDelivers the requested cold, guarded, unsmiling expression with the exact vibe the prompt asked for.Scene compliance✓ WorkedCleansly follows the interrogation-room prompt with the navy shirt, hands flat on the table, and bare room composition all in place.Scene compliance✓ WorkedExecutes the market scene strongly: the crowd feels authentic, the mustard sari and red blouse match the prompt, and the jute bag with vegetables is present.
Provenance
- Observation
- 65f7903c-08c2-4f33-987b-43e5e8b93aa7
- Evidence run
- 1dfb8fa4-f7f0-47dc-911b-7db3af467e9a
- Study
- Generate Consistent AI Characters Across Different Scenes and Poses
- Research task
- 86b96df11
- Tested at
- not recorded
- Source
- first-party
- Evidence state
- verified
- Proof shown
- input + output shown
- Cost / latency
- not captured
- Repeat run
- not captured
- Tester
- not captured
The last three rows are honest blanks, not placeholders — our capture has no field for them yet.
Query this
get_evidence({
tool: "chatgpt"
})MCP · mcp.aidemos.com/api/mcp
Free with attribution.
Same input, same check — 4 other tools
measured on Identity preservation
Gemini✓ WorkedIn the crowded market scene, the character stays highly recognisable, preserving face shape, smile, eyebrows, and overall facial structure.ImagineArt✓ WorkedThe interrogation-room output stays close to the reference in face shape, skin tone, expression, posture, and overall appearance.Leonardo AI◐ MixedWhen the face is turned partly away in a busy scene, the output still feels like the same character but cannot be fully verified.Scenario⚠ StruggledIdentity drifts in a crowded full-body scene, with a narrower jawline and a thinner body making the result read as a similar-looking person rather than the same subject.
This evidence is published in
Real inputs and real outputs, no retouching · every cell queryable via API & MCP · aidemos.com