Many image models typically generate outputs around 1 to 4 megapixels. While this is sufficient for everyday use, creators working on professional graphics, large-format visuals, product imagery, and detailed compositions often need much higher resolutions.
MiniMax H3 Image aims to address these limitations by combining high-resolution image generation, reference-based editing, and stronger instruction following in a single model.
 |
| Minimax H3 Image Showcase(Ref- official page) |
MiniMax-H3-Image is an open-source image generation and editing model from the MiniMax H3 family. It brings text-to-image generation and multi-reference image editing together in one unified checkpoint. The model can generate images at approximately 16 megapixels across different aspect ratios, reaching resolutions such as 4096 x 4096 for square images and 6272 x 2688 for ultrawide compositions.
These outputs are generated directly without requiring an external upscaler. Its main focus is balancing three capabilities- generation speed, high image quality, and accurate instruction following. It also supports up to 16 reference images, giving users more flexibility when building complex compositions.
One of the main highlights is its native high-resolution generation capability. While many image models operate around 1 to 4 MP, MiniMax-H3-Image targets approximately 16 MP outputs while keeping computational requirements manageable. Its effective working range is around 4 MP to 20 MP, with approximately 16 MP generally recommended for the best results. This makes it suitable for detailed images where textures, fine elements, and overall image clarity matter.
Installation
1. First you need to have ComfyUI installed. If already have just update it from the manager.
2. Download Minimax H3 image model by official Comfy team .
-minimax_h3_image_pruned_int8_convrot.safetensors
-minimax_h3_image_pruned_bf16.safetensors
-minimax_h3_image_pruned_w4a8.safetensors
Minimax H3 FP8 by Unsloth. You can also use the Fp8 variant released by the developer.
Choose any of the three variants. Save this into ComfyUI/models/diffusion_models folder.
Alternative-
You can also use GGUF variants released by the community.
- Minimax h3 image by Abiray
- Minimax H3 by RebelAI
Make sure you are using ComfyUI-GGUF custom node by dev city-96. Save the models into ComfyUI/models/unet folder.
3. Download vae model into ComfyUI/models/vae folder.
-minimax_h3_image_vae_fp16.safetensors
4. Download and place text encoders files in ComfyUI/models/text_encoders folder.
-qwen3.5_9b_w4a8_gptq.safetensors
-qwen3vl_32b_minimax_h3_int8_convrot.safetensors
-qwen3vl_32b_minimax_h3_nvfp4_awq.safetensors
If you want to get the similar quality that of official Minimax h3 image API, its recommended not to use the lightweight text encoder as it will not generate good results.
Workflow
1. Download the Minimax H3 image workflows from our Hugging face repository.
(a) Minimax_H3_Image_T2I.json (Text to image workflow)
(b) Minimax_H3_Image_Edit.json (Editing image workflow)
2. Drag and drop into ComfyUI.
3. Load all the relevant models (h3 image model, vae, text encoders, loras) into their nodes.
4. Put the positive prompts into prompt box.No negative prompting needed.
Set Ksampler settings-
(Text to Image/editing)-
CFG- 1.0
Steps-6
Sampler-Euler
Scheduler-Simple
Resolution supported-
Aspect ratio --- size in pixel(width x height)
(a) 1:1 -- 4096x4096
(b) 4:3 / 3:4 -- 4736x3552 / 3552x4736
(c) 3:2 / 2:3 -- 5024x3360 / 3360x5024
(d) 16:9 / 9:16 -- 5472x3072 / 3072x5472
(e) 21:9 -- 6272x2688
Effective range-4MP to 20MP, best is around 16mp.
For editing you can input up to 16 reference images.
5. Hit run to generate.
Prompt:
Pretty Princess Peach leaving the club at night, wearing black sunglasses, pappaeazzi flash photo, does not want to be photographed in motion blur, blurry, grainy. TMZ logo in corner.
Prompt:
A dark grainy security camera footage photo of Santa Claus wearing night vision googles. He is sneaking through a living room and holding a bottle of Jack Daniels.
