Fizgig H3 Still node unlocks MiniMax H3 ref2img in ComfyUI, but one image takes ~20 min
Neggy5 · reddit · 2026-09-27
An artist who hand-draws in Procreate tested the ComfyUI-Fizgig-H3-Still custom node, which decodes still images with the MiniMax H3 VAE, and confirmed that ref2v (reference-to-image) works: combining a hand-drawn OC design, a chow chow photo, and a Melbourne street photo produced style-consistent chibi illustrations, including a two-character group shot. Main drawback is speed: 3-4 minutes for t2i, 20 minutes per ref2v image. Full prompts with <Picture N> references are included.
More from Multimodal
- TeleOCR Trends on Hugging Face: A Qwen2.5-VL-Based Chinese Document OCR Model — XingChen-AGI · 2026-09-28
- Higgsfield ships 11 production skills that leave Claude with editable project files — xiaohu · 2026-09-28
- Opus 5.5 makes a music video, drawing love from AI circle — repligate · 2026-09-28
- Under $1 per 66-second reel: open-model pipeline with Qwen3-TTS and MiniMax H3 — victor_explore · 2026-09-28
- Music generation is '110% solved', says researcher as AI song quality stuns — teortaxesTex · 2026-09-28
- 282 viral Claude Opus 5.5 videos with exact prompts, all in one open-source repo — dotey · 2026-09-28