MiniMax H3 Image AI Image Generator
MiniMax-H3-Image is MiniMax’s open image model in the H3 family. One checkpoint does text-to-image and image editing: about 16 megapixels at every listed aspect ratio, from 4096×4096 (1:1) to 6272×2688 (21:9), with no separate upscaler. The checkpoint is distilled to 6 sampling steps and does not use classifier-free guidance. It accepts up to 16 reference images. Generate and edit with MiniMax H3 Image on ImgToVid.
01 — Capabilities
What MiniMax publishes for MiniMax H3 Image
These are the specs MiniMax publishes for MiniMax H3 Image on ImgToVid.
01
About 16 MP, without an upscaler
MiniMax lists native outputs around 16 megapixels, including 4096×4096 and a 21:9 frame at 6272×2688.
02
6 steps, no classifier-free guidance
The released checkpoint is distilled to 6 sampling steps (6 NFEs) and does not use CFG.
03
Text-to-image and editing in one model
The same checkpoint generates from a prompt or edits one or more reference images. MiniMax lists support for up to 16 references.
04
H3 transformer, audio branch removed
The diffusion transformer is the H3 design without audio layers: 33.1B parameters, with the MiniMax-H3 text encoder left unchanged.
02 — Prompt to output
Prompts that match the stills above
Each prompt describes the picture on its card. “Use this prompt” opens that text in Create with MiniMax H3 Image.
01 · PortraitPromptClose photographic portrait of a young man with curly hair wearing a grey knit sweater, soft window light, he glances aside and smiles warmly, shallow depth of field.
02 · ProductPromptA clean white leather sneaker on a pale grey seamless studio backdrop, soft diffused light, three-quarter view, crisp stitching, minimal commercial photograph.
03 · StreetPromptPhotographic view down a cobblestone lane in an old European village, climbing roses on stone walls, café tables in morning sun, a few people strolling in the distance.
04 · CharacterPromptPhotograph of a young woman with shoulder-length brown hair in an olive t-shirt, golden late-afternoon light, she turns her head and looks into the distance, breeze in her hair.
03 — Where it fits
Where MiniMax positions H3 Image
Use it for high-detail stills and edits that follow a clear instruction.
Large stills across wide and square frames
MiniMax lists about 16 MP from 1:1 through 21:9, generated directly rather than upscaled afterward.
Edits that keep several references in play
One checkpoint covers generation and editing, with up to 16 reference images.
A lighter sibling, P-H3-Image
Pruna’s P-H3-Image is a separate model on ImgToVid, aimed at photographic stills around 8 MP.
Generate on ImgToVid
Write the still here and generate with MiniMax H3 Image in Create. Membership and credits are checked there.
04 — Verdict
MiniMax H3 Image strengths and trade-offs
Strengths
- About 16 MP native output, from 4096×4096 (1:1) to 6272×2688 (21:9), with no upscaler step
- Distilled to 6 sampling steps and no classifier-free guidance
- Text-to-image and image editing share one checkpoint, with up to 16 reference images
- 33.1B H3 diffusion transformer; the MiniMax-H3 text encoder is unchanged
Trade-offs
- 16 MP files are heavier to store and review than typical 1–4 MP stills
- Self-hosting the open weights is a large download: the model card lists a 68.65 GB BF16 DiT, and the text encoder loads from MiniMax-H3
- Vague edit instructions leave more of the original frame unchanged than a single, explicit change
05 — Side by side
MiniMax H3 Image vs P-H3-Image vs Seedream 5.0 Pro
H3 Image and P-H3-Image figures are from their public model cards. Seedream 5.0 Pro is the tier available on ImgToVid.
06 — How it works
Start a still on ImgToVid
Step 1
Describe one still
Name the subject, the light, and the frame. For an edit, say what to change and what to keep.
Step 2
Use the generator on this page
It opens Create with MiniMax H3 Image. The prompt and the settings go with you.
Step 3
Generate in Create
Membership and credits are checked there before MiniMax H3 Image runs.
07 — FAQ
MiniMax H3 Image FAQ
What is MiniMax H3 Image?
MiniMax-H3-Image is MiniMax’s open image model for text-to-image and image editing. It is a stills checkpoint in the H3 family, not the MiniMax H3 video models.
How large are the images?
The model card lists about 16 megapixels at the published aspect ratios, from 4096×4096 (1:1) to 6272×2688 (21:9), without a separate upscaler.
How many reference images can an edit use?
MiniMax lists up to 16 reference images. Generation and editing use the same checkpoint.
How is P-H3-Image different?
P-H3-Image is Pruna AI’s separate open model based on MiniMax-H3. Its card recommends 8 MP, 8 steps, and 1 to 8 input images. It is not the MiniMax H3 Image checkpoint.
Can I generate MiniMax H3 Image on ImgToVid?
Yes. Write a prompt on this page and generate with MiniMax H3 Image in Create.
Can I use the results commercially?
Commercial use of anything you generate on ImgToVid depends on your plan. Check Pricing and our Terms before you publish, and make sure prompts and uploads do not infringe someone else’s rights.
08 — Explore more
Other AI image models
Generate with MiniMax H3 Image on ImgToVid
Write a prompt, add reference images when the edit needs them, and generate in Create.
Start with MiniMax H3 Image