Edit a picture in plain words or make a new one from scratch. For an edit, you can give P-H3 up to seven more pictures and tell it what to take from each one.

How to use it

1. Drop the workflow into ComfyUI. Load the picture you want to change and write your edit.

2. For a reference, connect an image loader to picture 2 on Write the prompt. Add more in order through picture 8. Name the picture in your request, for example “Give the person in picture 1 the jacket from picture 2.”

3. Press Run. Drag the comparison slider to check what changed. Auto prompt shows the expanded request in Show Text.

4. To generate from words, turn Edit a picture off and choose a size in Latent Size. Turn Upscale the result on if you want SeedVR2 to double the width and height.

What you need

Update ComfyUI and AusBoss nodes first. This version ran on ComfyUI 0.44.0 with AusBoss nodes 2.20.5. Pick one matching model and text encoder pair. The workflow opens on Low VRAM.

• Low VRAM · p-h3-image_w4a8.safetensors → models/diffusion_models · 13.4 GB

• Low VRAM · p-h3-image_text_encoder_w4a8.safetensors → models/text_encoders · 3.6 GB

• High VRAM · p-h3-image_int8_convrot.safetensors → models/diffusion_models · 21.2 GB

• High VRAM · p-h3-image_text_encoder_int8_convrot.safetensors → models/text_encoders · 5.1 GB

• Both · p-h3-image_vae_int8_convrot.safetensors → models/vae · 3.1 GB

• Auto prompt · p-h3-image_prompt_upsampler_bf16.safetensors → models/text_encoders · 19.3 GB

• 2× upscale · seedvr2_3b_int8_convrot.safetensors → models/diffusion_models · 3.46 GB

• 2× upscale · seedvr2_ema_vae_fp16.safetensors → models/vae · 501 MB

Tips

• With Auto prompt off, say what should stay the same as well as what you want to change.

• Check faces, lighting and the background after a multi-picture edit. The model can change more than you asked for.

• Edits start at 4 megapixels. The Resize row on the picture loader sets that size.

Powered by MiniMax H3. P-H3-Image is by Pruna AI and uses the MiniMax H3 Community License. Optional upscaling uses SeedVR2 under Apache 2.0.

v1.0 (2026-11-06) · First release.

Report an issue · Discord · YouTube @ausbossdev · X @Zanzibased · GitHub