Image models can't count circles: Verifiable Visual Rewards boost instruction following
StellaLisy · x · 2026-09-29
Stella Lisy's team shows image generation models still fail at precise instruction following (object counts, spatial relations) and introduces Verifiable Visual Rewards (VVR): prompts and deterministic verifiers derived programmatically from geometric scenes, replacing unreliable reward models.
Key points:
- Releases VVRBench (10,000 tasks, 32 constraint types) and harder VVRBench-Challenge (720 tasks); best model GPT-Image-2.5-Sunburst solves only 21% of the Challenge set
- Using VVR scores as RL rewards (RLVVR) lifts Stable Diffusion 3.5 Medium from 2.8% to 28.3% on VVRBench, with easy-to-hard generalization
- Mixing VVR into existing post-training objectives further improves overall quality and human preference
More from Multimodal
- Stunning AI-generated motion-blur portraits with a film-photography feel — azed_ai · 2026-11-16
- A Bauhaus functionalist prompt template for geometric poster-style AI art — LudovicCreator · 2026-11-16
- Vivix W1 video model: 100 ten-second ads for about $5 at $0.005 per second — Aiden_Tech_Ai · 2026-11-16
- User Shares Coral Reef Creation Labeled 'Qwen 4 Max (xhigh)', an Unannounced Model Name — intellectronica · 2026-11-16
- Reddit megathread crowdsources the best open-source audio models: ASR, TTS, and text-to-music — rm-rf-rm · 2026-11-16
- Animating AI-generated 2D game art with FLUX.2 klein: masked redraws work but remain a hack — emilrueh · 2026-11-16