TextSculpt-Bench - Visual Quality: leaderboard

Visual quality (0-100) of text edits, the mean of GPT-5.2 yes/no judgments on placement, style consistency and physical plausibility on TextSculpt-Bench (2026), 800 scene-text editing tasks (adding, removing, replacing and mixed edits of text in photos and posters). Higher is better.

Metric: Visual quality (0-100): mean of three yes/no GPT-5.2 judgments (edit location, style consistency, physical plausibility), over 800 TextSculpt-Bench text-editing tasks (200 each of addition, removal, replacement and hybrid editing) on Pexels photos and generated posters, default model settings; higher is better. Source: arxiv.org. Saturation forecast: Rough model projection: around 2026. 9 models tracked. Updated 7 Oct 2026.

Top models

#ModelScoreOverall rank
1Seedream 4.576
2FireRed-Image-Edit-1.070
3Gemini 2.5 Flash Image69
4Qwen-Image-Edit-251163
5Seedream 4.057
6LongCat-Image-Edit54
7Step1X-Edit45
8OmniGen228#4579
9Bagel25

Interactive version: theaggregate.ai/benchmark?slug=textsculpt-bench-visual-quality · How It Works · Data refreshed daily, snapshot 2026-11-21.