H3 Image — P-H3, 8 steps

H3's eye for light and skin as a still-image model.

P-H3-Image (Pruna AI) distilled MiniMax H3 into an 8-step image generator — no CFG, up to 8 MP, text that renders. It runs as a 2 GB LoRA on the Flow checkpoint, with its own text encoder and VAE. Optional prompt upsampler included.

Model: MiniMax H3
VRAM: 24–48 GB
Level: Beginner

Download workflow — free

Drag the .json onto ComfyUI. Runs best on the Gmanski RunPod template.

You get

H3 Image (P-H3) workflow
Optional prompt-upsampler nodes (bypassed)

You need

ComfyUI 0.44+
24 GB VRAM for 4 MP, ~21 GB more headroom for 8 MP

Model packs · Text to image

minimax-h3-flowminimax-h3-pimage

On the template: gmanski-packs install minimax-h3-flow minimax-h3-pimage

Pairs well with.

MiniMax H3new

H3 Image Edit — up to 8 references

Edit a picture with other pictures, in 8 steps.

Intermediate · 24–48 GB$19

MiniMax H3

Character Portrait — one photo

A new photo of a person from a single reference picture. No training.

Beginner · 48 GB+$24

MOSS-Imagenew

MOSS-Image

The open model that topped Civitai this month. Apache 2.0.

Beginner · 16 GB+Free