RefMod + MiniMax H3 ComfyUI workflow: character consistency without LoRA training on 6GB VRAM
cgpixel23 · reddit · 2026-10-15
A Reddit user shared a ComfyUI workflow combining RefMod custom nodes with MiniMax H3 for character-consistent image generation without training a traditional character LoRA.
Workflow: prepare 8 photos of your character → build a RefMod file with the custom nodes → load it into the MiniMax H3 workflow → prompt any scene, outfit, pose or environment while preserving character identity.
Timing at 1.8MP: 6 min at 8 steps, 14 min at 25 steps. Runs on just 6GB VRAM; workflow file, Civitai write-up and a video tutorial are linked.
More from Multimodal
- MiniMax-H3 hailed as favorite open release of the year, runs fully local — AdinaYakup · 2026-10-23
- Michal Malewicz experiments with depth maps in his creative workflow — michalmalewicz · 2026-10-23
- One-Shot Tunable Style Transfer on Flux3 Via Reference Latent + Node — Winter_unmuted · 2026-10-23
- PhysVidBench accepted at EMNLP 2026: do video gen models grasp physical commonsense? — aykuterdemml · 2026-10-23
- EMNLP 2026 paper teaches multimodal LLMs to get cartoons via incongruity-resolution supervision — aykuterdemml · 2026-10-23
- A gorgeous Midjourney style: sref code 820221253 for dreamy lavender watercolor looks — umesh_ai · 2026-10-23