MiniMax H3 - Simple Image Edit — v1.0 — v2.0

This MiniMax H3 v2.0 workflow edits an image from a text prompt: provide an image, describe what to change, add, or remove and what to preserve, and receive an edited image. You can also use up to nine reference images.

Version 2.0 uses an experimental VAE tailored for image generation.

The graph includes steps to preview and save the edited image.

Version 2.0 accepts up to nine reference images.

Have these model files available: `minimax_h3_audio_vae_fp32.safetensors`, `minimax_h3_ref2va_pruned_int8_convrot.safetensors`, `minimax_h3_video_vae_fp16.safetensors`, and `qwen3vl_32b_minimax_h3_nvfp4_awq.safetensors`. The workflow also requires the `rgthree-comfy` node pack.

Prompting tip: describe what to change, add, or remove, then say what to preserve.

Suggestion · not verified

Native image sizes listed for H3 are 608 × 352, 736 × 416, 864 × 480, 960 × 544, 1056 × 608, 1152 × 640, 1216 × 672, 1280 × 736, 1344 × 768, 1376 × 768, 1504 × 832, 1664 × 928, 824 × 1024, and 1920 × 1088.

Images at other sizes are automatically rescaled, and results may differ in quality.