FREE starter on Whop (lite workflow + exclusive cheat sheet): https://whop.com/imagelab/free-starter-h3-image-workflows
FULL pack ($39): https://whop.com/imagelab/h3-image-workflows-minimax-h3-pruna-p-h3-16-mp-multi-reference-edit-comfy — use code H3IMG30 for 30% off ($39 → $27.30)
Requirements — read before downloading: ComfyUI 0.45.0+. The 3 model files total about 49 GB on disk (21 GB diffusion model + 27 GB text encoder + 1.2 GB VAE). ComfyUI streams the models from your system RAM, so you need 24 GB or more of RAM (32 GB recommended) — VRAM matters less, 12 GB cards work. The first run loads about 49 GB and the progress bar can sit at a low percentage for several minutes: this is normal, wait, do not cancel. On 8 GB cards, use the NVFP4 text encoder (15.7 GB) instead of the INT8 one.
## FREE vs FULL
| | FREE (this download) | FULL ($39) |
|---|---|---|
| Workflows | 1 — H3 text-to-image | 4 — T2I + multi-reference edit + P-H3 upsampler + 2-step FastDraft |
| Multi-reference editing | — | Up to 16 reference images per edit |
| Prompt upsampler | — | P-H3 Qwen3.5-9B upsampler workflow |
| Prompt library | — | 10 formulas + edit recipes |
| Guides | This README | Full EN quickstart + Arabic guide |
## What this is
MiniMax-H3-Image is the #1 open-weights image model on the AA Editing leaderboard (Elo 1,138): 16 MP output, 6-step sampling, up to 16 reference images per edit. This free lite workflow gives you the core text-to-image graph — 6 steps, CFG 1, 2048×2048 — ready to run in ComfyUI 0.45.0+.
## Files you need (not included)
From Comfy-Org/MiniMax-H3-Image: minimax_h3_image_pruned_int8_convrot.safetensors (21 GB) → diffusion_models, qwen3vl_32b_minimax_h3_int8_convrot.safetensors (27.1 GB) → text_encoders, minimax_h3_image_vae_fp16.safetensors (1.17 GB) → vae. 24 GB+ VRAM recommended.
## License warning
MiniMax H3 Community License — excludes US / EU / UK / South Korea without a separate licence from MiniMax; products must show "Powered by MiniMax H3". Read the model card before commercial use.
Full pack on Whop: https://whop.com/imagelab/h3-image-workflows-minimax-h3-pruna-p-h3-16-mp-multi-reference-edit-comfy
