VESFlow accepted at NeurIPS 2026: training-free velocity editing makes few-step flow matching safe
mohitban47 · x · 2026-10-09
VESFlow, a training-free safety method for flow matching models, has been accepted to NeurIPS 2026. Existing safety approaches fail in the few-step regime: trajectory-level guidance needs many steps to accumulate small corrections, while embedding-level editing is ineffective because toxic concepts spread across sentence-level embeddings like T5. VESFlow instead edits the velocity field itself, improving generation safety even with only a few sampling steps.
More from Multimodal
- Speridlabs Releases Iris-3B: Pixel-Space Generative Model That Can Replace DINOv2 — Apprehensive_Sky892 · 2026-10-09
- GPT-6 Luna Shows Surprising Skill at Spotting AI-Generated Images — Angaisb_ · 2026-10-09
- KAIST's LongTake: long-horizon teacher forcing keeps 30-60s video generation dynamic — kaist-ai · 2026-10-09
- AI image of the day: GPT-6 Image 2.5 recreates 1940s Chinese village newsreel — DeryaTR_ · 2026-10-09
- Video character swap on 8GB VRAM: 640x480 render in 10:33 on RTX 4070 — big-boss_97 · 2026-10-09
- Google opens SynthID Detector globally; Vidu Q4 Preview prices video gen from $0.013/sec — 创业邦 · 2026-10-09