hyperflow-h3-flash video model updated: sampling in just 4, 6, or 8 steps
Apprehensive_Sky892 · reddit · 2026-11-07
VideoRebirthAI released an updated version of hyperflow-h3-flash on Hugging Face, supporting 4, 6, or 8 sampling steps for fast few-step video generation. Weights are open at huggingface.co/VideoRebirthAI/hyperflow-h3-flash.
Useful for local workflows (e.g., ComfyUI) where speed matters, with step count as a speed/quality tradeoff.
More from Multimodal
- Veteran C++ dev ships native ComfyUI desktop client, bypassing browser lag — elgeminy · 2026-11-07
- RevealLayer ComfyUI node brings occlusion-aware image layer decomposition — TBG______ · 2026-11-07
- Dev distills Anthropic's Renaissance engraving art style into a reusable image-generation Skill — yihui_indie · 2026-11-07
- 2-step pDMD LoRA for p-h3-image released, fix blurry fast-generation outputs — Illustrious_Row_9971 · 2026-11-07
- Dreamy pastel Midjourney explorations: candy factories and coastal towns in pink clouds — hewarsaber · 2026-11-07
- Midjourney 8.2 test: stunning Munch-inspired expressionist artwork — LudovicCreator · 2026-11-07