Semi-transparent Canny trick reduces reference bleeding in Ref2V MM-H3 video generation
Vortexneonlight · reddit · 2026-11-02
A user shares a technique for controlling reference-image "bleeding" in Ref2V MM-H3 video generation: raw pose references tend to leak their environment, clothing, and colors into the video, while pure Canny edges cause H3 to ignore or hallucinate elements on complex poses.
The fix: overlay the original reference at low opacity — a semi-transparent Canny — so the model both reads it as edge structure and understands the elements inside. Benefits: less bleeding, more output variety when changing the prompt, and better handling of complex/unusual poses H3 doesn't know.
The post includes a full working recipe: 6 steps, the Dmad-Larry dareties mix LoRA, MP 0.7, and a complete prompt structured as [Visual Style] / [Asset Mapping] / [Chronological Actions] with multiple reference images and per-shot action timing.
Related event: Semi-transparent Canny trick tackles reference bleeding in video generation(2 posts)→
More from Multimodal
- FrameForge Reuses Video Model Priors for Unified, Adaptive Conditional Image Generation — FrameXAI · 2026-11-02
- Tencent's Wizard World Model Enables Real-Time Agentic Control of Any Game World — tencent · 2026-11-02
- Open-source ComfyUI node pack renders 2–3 minute 1080P videos with MiniMax H3 in one run — ReferenceFluffy6159 · 2026-11-02
- MiniMax H3 video gen on 8GB RX 6600: 5s clip in 10 minutes with ComfyUI tweaks — posojekh · 2026-11-02
- Generative AI Course Lecture 8: How ComfyUI Custom Nodes Work, From Source to App Mode — CeFurkan · 2026-11-02
- The 'Rian Johnson app' generates strikingly cinematic AI photos — threepointone · 2026-11-02