Included in AI Photo's controlled portrait and moderation tests

GPT Image 2 Medium

A capable general image route with strong instruction following in AI Photo's retained comparisons. Its visible behavior should be judged task by task rather than from a single overall score.

Observed strengths
  • Instruction following
  • Conversational revision workflows
  • General-purpose image creation
What to check
  • Provider policy can shape completion
  • Portrait identity varies by input
  • Current AI Photo evidence is concentrated in a few controlled tasks
Inspectable evidence

Tests containing GPT Image 2

ComparisonsFive AI Image Models, One Input: 27-Output Controlled Portrait Test

Five shared portrait prompts across Seedream 4.5, GPT Image 2 Medium, Grok, Seedream V5 Lite, and V5 Pro—plus one deduplicated Lite-versus-Pro extension row.

Open test →ComparisonsNano Banana 2 vs Seedream vs GPT Image 2 vs Grok: 14-Prompt Portrait Comparison

Fourteen adult portrait prompts with side-by-side outputs from Nano Banana 2, Seedream 4.5, GPT Image 2 Medium, Grok, and Seedream V5 Lite.

Open test →ComparisonsWhich AI Image Generator Blocks More? 9-Attempt Same-Input Test

Nano Banana 2, GPT Image 2 Medium, and Seedream 4.5 tested on the same three adult-fashion prompts: nine attempts, seven outputs, and two visible safety blocks.

Open test →
Related tasks
Task guideAI Portrait Realism

Compare skin texture, anatomy, camera logic, lighting, and the difference between a polished image and a believable photograph.

Open task guide →Task guideAI Image Editing

Judge edits by what changed, what stayed fixed, and how many attempts were required—not by whether the final image looks attractive in isolation.

Open task guide →Evidence guideAI Image Moderation

Compare documented outcomes from the same permitted adult-fashion prompts without extrapolating a platform-wide policy from one request.

Open task guide →