Custom H3 I2V/REF2V ComfyUI workflow: 20s video in 22 minutes on a 3090, controlling 5 characters
EQLMonk · reddit · 2026-11-07
A Reddit user shares their first custom ComfyUI workflow for H3-based I2V/REF2V video generation (workflow posted to pastebin), attempting to control 6 people in one video — currently managing 5 with multiple attempts. On a 3090 with 64GB RAM, a 20-second video takes about 22 minutes. They used Claude to help with the tedious parts and are asking the community for feedback on the workflow.
More from Multimodal
- Kimi K3.1 Pro turns a photo into a 3D Three.js scene, with open-sourced template — demian_ai · 2026-11-07
- TischEins adds 8,800+ Midjourney SREF style codes in November, free to browse — tisch_eins · 2026-11-07
- ComfyUI-TabCapture: live tab capture node for video, stills and audio — Sad_Berry_4621 · 2026-11-07
- 20-minute deep dive on AI world models: Genie 3, Cosmos, and the synthetic data revolution — bilawalsidhu · 2026-11-07
- Cinematic AI-Generated Character Portraits Show Photo-Real Film-Quality Results — gen_ericai · 2026-11-07
- Dreamina hybrid reference workflow: 7 images + 1 video, stunning 3D characters in 360s — gen_ericai · 2026-11-07