AIM-Bench (Affective Image Manipulation) - Perceptual Quality: leaderboard
AIM-Bench affective image editing scored by the VIEScore perceptual quality rating of naturalness and freedom from artifacts, on a 0 to 10 scale, higher is better.
Metric: VIEScore PQ (0-10): naturalness of the edited image, structural fidelity and absence of distortions, 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. Updated 7 Oct 2026.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Seedream 4.0 | 6.97 | |
| 2 | FLUX-Kontext-Pro | 6.77 | |
| 3 | FLUX-Kontext-Max | 6.77 | |
| 4 | Qwen-Image-Edit-2509 | 6.4 | |
| 5 | UniWorld-V2 | 6.22 | |
| 6 | DreamOmni2 | 6.11 | |
| 7 | Qwen-Image-Edit-Plus (AIM-Bench checkpoint unspecified) | 6.11 | |
| 8 | SeedEdit 3.0 | 6.02 | |
| 9 | OmniGen2 | 5.97 | #4579 |
| 10 | Step1X-Edit (AIM-Bench checkpoint unspecified) | 5.89 | |
| 11 | FLUX.1-Kontext-dev | 5.66 | |
| 12 | InstructPix2Pix | 4.62 | |
| 13 | BAGEL-7B-MoT | 3.66 | #4619 |
Interactive version: theaggregate.ai/benchmark?slug=aim-bench-affective-image-manipulation-perceptual-quality · How It Works · Data refreshed daily, snapshot 2026-11-21.