comfyui-workflows/models/minimax/minimax-h3-int8-r2v-community/README.md

MiniMax H3 INT8 R2V Community

Summary

Community MiniMax H3 reference-to-video workflow built around INT8 inference and an explicit acceleration stack.

Model Family

• minimax

Status

• imported-only

Source

• Workflow repo: https://huggingface.co/javawock7618/comfy-MiniMax-H3-workflows
• Workflow file: https://huggingface.co/javawock7618/comfy-MiniMax-H3-workflows/resolve/main/MiniMax_int8_R2V-javano2608.5.json?download=true
• Original publisher: javawock7618
• Date imported: 2026-08-08

Developer notes

• The repo describes this as part of an INT8 FLF & Reference Video collection.
• The author explicitly targets:
  • reduced VRAM usage
  • faster inference
  • improved memory efficiency
• The documented acceleration stack includes:
  • Sol-Attn
  • Spectrum
  • optional Lightx2v LoRA
  • optional MiniMax-H3-Turbo-Lora
  • optional latent upscaler

Our notes

• This is the strongest low-VRAM-oriented MiniMax import we have so far because it is built around INT8 from the start.
• Still not guaranteed easy on a 3070, but much more aligned with your constraints than generic heavyweight MiniMax examples.
• Good candidate once you want to test whether MiniMax reference-driven video is remotely viable locally.

Required custom nodes

• ComfyUI-SolAttn_triton
• ComfyUI-Spectrum-MiniMax-H3
• optional ComfyUI-H3-Motion-Context for related FLF workflows
• optional ComfyUI-MiniMaxH3_LatentUpscaler
• Kijai/MiniMax-H3_comfy for Lightx2v acceleration
• optional MiniMax-H3-Turbo-Lora-ComfyUI

Notes

• Best current MiniMax low-VRAM / INT8 reference-video workflow in the repo.
• More of a carefully chosen experimental path than a guaranteed easy win.