Minimax H3 local runs full of artifacts: ComfyUI workflow troubleshooting
Dendwdls · reddit · 2026-09-09
A user with 16GB VRAM / 64GB RAM locally generating a 10s Minimax H3 video (woman playing with a cat, then zoom to a phone) reports heavy artifacts, broken faces, and a background that drifts with camera movement, taking 20+ minutes per run. The workflow uses rf2va plus an 8-step LoRA with a character sheet and location image as references; they posted the full workflow link asking whether the workflow or the prompt is at fault.
Related event: Running MiniMax H3 Locally: Artifacts and Heavy Render Times Plague Users(2 posts)→
More from Multimodal
- World Labs launches RTFM world model generating explorable 3D worlds from one image — drfeifei · 2026-09-09
- Painterly samples surface from rumored OpenAI model GPT-Image-2.5 Sunburst — DavidmComfort · 2026-09-09
- Indistinguishable AI stunt video sparks call to ban unlabeled AI content accounts — JourneymanChina · 2026-09-09
- DLSS 5 comes to ComfyUI via open-source custom node ComfyUI-DLSS5 — jib_reddit · 2026-09-09
- Canva launches MCP server: create and manage designs inside chat — modelcontextprotocol · 2026-09-09
- Overworld launches Waypoint 2 Nano: a world model running locally at 60fps with 50ms latency — lcastricato · 2026-09-09