AIM-Bench (Affective Image Manipulation) - Emotion Accuracy: leaderboard
Metric: Emotion accuracy (%): Gemini-2.5-Pro classifies each edited image into the 8 Mikels emotions, and an edit counts as correct when the predicted emotion is the target emotion, over the 800 AIM-Bench affective image editing items (EmoSet source images, a target Mikels emotion with target valence-arousal-dominance coordinates and an editing instruction; 8 emotion categories, 5 editing types), each model run once with its official default settings; higher is better. Source: arxiv.org. Saturation forecast: Rough model projection: around 2026. 13 models tracked.
Top models
| # | Model | Score |
|---|---|---|
| 1 | Step1X-Edit (AIM-Bench checkpoint unspecified) | 68.84 |
| 2 | Qwen-Image-Edit-2509 | 62.34 |
| 3 | Seedream 4.0 | 61.55 |
| 4 | UniWorld-V2 | 60.38 |
| 5 | Qwen-Image-Edit-Plus (AIM-Bench checkpoint unspecified) | 58.39 |
| 6 | Flux-kontext-max | 57.01 |
| 7 | BAGEL-7B-MoT | 56 |
| 8 | Flux-kontext-pro | 54.77 |
| 9 | OmniGen2 | 54.37 |
| 10 | SeedEdit 3.0 | 52.73 |
Interactive version: theaggregate.ai/benchmark?slug=aim-bench-affective-image-manipulation-emotion-accuracy · How It Works · Data refreshed daily, snapshot 2026-10-07.