StepRefine: Step-Aware Decoding Cuts Hallucinations in Multimodal Diffusion LMs
Cornell0917 · hf · 2026-11-16
Researchers from Monash, Cornell, UC Santa Cruz, Northeastern, and Minnesota propose StepRefine, a plug-and-play decoding strategy that mitigates hallucinations in multimodal diffusion language models (MDLMs).
Key observations: mask uncertainty follows an overall downward trend across iterative denoising, but a few denoising steps show localized entropy spikes closely associated with hallucinations. The authors argue that committing to discrete tokens discards distributional information at masked positions, limiting continued interaction, and identify anomalous steps by measuring each step's entropy deviation from its neighboring temporal window.
StepRefine switches between discrete denoising and latent refinement based on entropy spikes and distributional stability, reducing hallucinations without retraining the model.
More from Multimodal
- Relay VSR vs Starlight: 4x video upscaling, nearly 6x faster on old tapes — reeight · 2026-11-16
- Kijai ships vae_approx preview fix for MiniMax H3 image model — VRGoggles · 2026-11-16
- Fable 5.2 learns to run: 30-second 3D skeleton animation rendered in agent-focused Vork engine — FinanceYF5 · 2026-11-16
- Music video made with Claude Code driving Blender: physics-solver dolphin animation — maximstark_ · 2026-11-16
- Musk shows Grok Bot auto-editing 23-hour Starship livestream into highlight reel — elonmusk · 2026-11-16
- OpenFill's Baby adds AI object selection and MiniMax H3 inpainting to ComfyUI — shukanimator · 2026-11-16