رجوع
Claude Sonnet 5.5ChatGPT GoGeminiGrokDeepSeekFacial analysisVisagismConsistency

Use Case: Comparing Visagism and Phenotype Facial Analyses Across Five AI Tools

96% مفيد4% غير مفيد
0 تعليقات

1. Summary

I uploaded profile photos of my face to five AI tools and asked each one the same questions about facial features, visagism impressions, and phenotype. I cataloged the answers by topic to see where the tools agree, where they diverge, and how each handles questions that invite speculative conclusions.

Result in one sentence: the AIs are consistent when describing what is visible, vary widely when interpreting it, and are unreliable when asked to infer origin from a face.

2. Context

Visagism is the practice of relating facial features to the image a person conveys, and it is often used to guide choices about hair, beard, glasses, and personal style. Because each AI has different training and behavior, I wanted to know how reproducible this kind of analysis is across tools and how much I should trust it.

3. Objective

Evaluate how the tools:

  • Describe visible features: face shape, eyes, eyebrows, nose, lips, and cheekbones.
  • Interpret the visual impressions the face conveys from a visagism perspective.
  • Handle uncertain or speculative questions, such as phenotype.

4. Method

ItemDetail
ToolsClaude Sonnet 5.5, ChatGPT Go, Gemini, Grok, DeepSeek
Input3 photos (neutral expression facing forward, profile view, smiling expression)
QuestionsWhat is the shape of this face? What is the shape of the lips? What is the shape of the eyes? What is the shape of the eyebrows? What is the shape of the nose? Where are the cheekbones on this face? What phenotypes does this face appear to have? From a visagism perspective, what impressions does this face convey?
RecordAnswers cataloged by topic in the note [[Rosto]]
AnalysisSide-by-side comparison to identify consensus, divergence, and abstentions

5. Results

5.1 Descriptive features

FeatureConsensus across the AIsDivergence
Face shapeOval (5 of 5)DeepSeek added some diamond-shaped traits
EyesAlmond-shaped (5 of 5)Variations: slight tendency toward round, horizontal, deep-set, slightly hooded
EyebrowsSoftly arched (Gemini, Grok, DeepSeek); straight with a slight arch (Claude, ChatGPT)Whether a slight asymmetry was noticed
NoseStraight, slightly rounded tip (5 of 5)Only the size: medium or moderate
LipsMedium, fuller lower lip, soft cupid's bowGemini saw them as "full"; DeepSeek saw a well-defined cupid's bow
CheekbonesBelow the eyes, soft (Claude, Gemini, Grok)Largest divergence: ChatGPT could not locate them; DeepSeek saw them as high and prominent

5.2 Visagism impressions

Despite different wording, recurring themes appeared:

  • Softness of the features contrasting with the seriousness and depth of the gaze.
  • Approachability and discretion, without superficiality.
  • Sensitivity, empathy, and a calm presence.
ToolMain reading
Claude Sonnet 5.5Mixed messages: a gentle face, but not immediately open
ChatGPT GoSoftness, approachability, and discretion
GeminiWelcoming, empathetic, and deep, with a quiet firmness
GrokSensitive, observant, more "deep" than "light" and more "authentic" than "constructed"
DeepSeekTwo archetypes: upper T-zone (the Sage: reserved, intelligent, magnetic) and lower T-zone (the seductress)

Highlight: DeepSeek gave the most elaborate and structured analysis, splitting the face into zones. It is also the one that departs furthest from the others, which shows how much interpretation depends on the framework used.

5.3 Phenotype: the most inconsistent point

ToolBehavior
Claude Sonnet 5.5Did not answer and steered away from the topic
ChatGPT GoAttempted it, then concluded it was inconclusive
GeminiAnswered directly: a mix of Indigenous and Iberian/Mediterranean ancestry
GrokGave a generic answer ("pardo, mixed-race") without going deeper
DeepSeekMost detailed: a Mediterranean profile with several possible origins

The answers ranged from refusal to detailed hypotheses and did not converge. This is the most relevant finding of the experiment: the more specific the question, the less consistent (and less defensible) the answers.

6. Critical Analysis

What worked

  • Describing visible features: high agreement between independent tools, which suggests good reliability at this layer.
  • Visagism impressions: useful as a starting point for reflecting on personal image, since the themes repeated.

Limitations shared across the AIs

  • Visagism is a subjective interpretation. The impressions should not be treated as conclusions about personality.
  • Phenotype, ethnicity, or ancestry cannot be reliably determined from a photo. Answers naming specific origins are speculation presented with confidence.
  • Results depend on photo quality, angle, lighting, and the prompt used.
  • The sample is small (one face, five tools) and has no validation from a professional visagist.

On tool behavior

  • Declining to answer or calling the analysis inconclusive, as Claude and ChatGPT did, is more honest than presenting an origin as fact. Direct answers feel more useful, but they convey a certainty the photo cannot support.

7. Practical Applications

  • Use the consensus on shape (oval face, straight nose, almond eyes) as a reference when choosing haircuts, beard styles, and glasses.
  • Use the recurring impressions (softness, depth, discretion) to think about what I want to communicate in professional photos and online profiles.
  • Treat divergence as a warning sign: where the AIs disagree, the information is weak.

8. Conclusion

Comparing several AIs turns a subjective analysis into something more robust: what shows up in all of them tends to be reliable, and what shows up in only one deserves skepticism. For describing a face and reflecting on the image it conveys, the method works well. For inferring origin or ancestry, it does not, and the comparison helps show why.

تعليقات0

سجّل الدخول للتعليق
  • لا توجد تعليقات بعد.

مقالات ذات صلة