ImageVideoAI

Text to image · Edits · 2K and 4K

MiniMax H3 Image: Text to Image and Edits at 2K or 4K

MiniMax H3 Image is the still-image model from the team behind the MiniMax H3 video model. On this page it is the only model on the panel. Write a prompt to create a new image, or attach up to three references to edit them.

Reference images · Optional, up to 3

0/3

JPG, PNG or WebP · up to 10 MB · Add up to 3 photos to edit or combine. Leave empty to create from text.

0/4000

Model

Sign up free to create images, with 30 free credits · MiniMax H3 Image · 2K · Auto

How a run works on this page

  1. Write the request, long if you need it

    The prompt box takes up to 4,000 characters. The other image models on ImageVideoAI stop at 2,000. That extra room is useful when one image has to carry a lot: the subject, the setting, the light, the lens, and the exact words on a sign or label. Put the most important instruction first and keep each detail in its own sentence.

  2. Add references only when you are editing

    With no images attached, the prompt creates a new picture from scratch. Attach one, two or three images and the prompt becomes an edit instruction, such as "Keep the bottle and the label. Replace the table with grey slate." Once a reference is in place, the ratio picker hides and the output takes the shape of the first image you added.

  3. Pick a size and check the quote

    Choose 2K or 4K. The quote bar shows the credit cost before anything runs, and 4K costs about twice as much as 2K. You need to be signed in. Guests who press the button see the sign-up prompt. A new account gets 30 credits that stay valid for 30 days, plus 6 credits each day that reset and do not pile up. If a run fails, the credits come back.

What this one model gives you

Two sizes, and what 4K means in practice

There are two output sizes, 2K and 4K. Free accounts run at 2K and the file carries a watermark. 4K is for paid plans and credit packs. It takes longer to finish, and the PNG you download can be tens of megabytes, so expect a slower download on a phone connection. For drafts and quick checks, 2K is usually enough. Move to 4K once the prompt is settled.

Ratios, references and the 4,000-character prompt

Without references you can pick Auto, 1:1, 2:3, 3:2, 3:4, 4:3, 9:16, 16:9 or 21:9. With references, there is no picker; the first image sets the frame. The reference grid holds three images at most. The long prompt limit is the clearest practical difference from the other image models here, and it suits briefs that read like a shot list.

What MiniMax says about the model

MiniMax released H3 Image on November 10, 2026, and describes it as an open-weight model built on the architecture of its H3 video model, sharing that model's text encoder. According to MiniMax, it renders at high native resolution, keeps text in images tied to its meaning, follows detailed instructions closely, and aims for a realistic look. These are the vendor's claims, not results published by ImageVideoAI.

What it does not do

There is no negative prompt field, so say what you want rather than listing what to avoid. There are no masked edits; an edit applies to the whole image through your wording. It does not output a transparent background. For a different style of edit, Nano Banana accepts more reference photos, and Seedream is another option on the site.

To compare every image model side by side, open the AI image generator. To start from an existing photo with edits in mind, use the image to image page.

From a still to a MiniMax clip

ImageVideoAI is mainly an image to video AI site, and this model fits that path. Under a finished image, the result panel offers Animate this image. It saves the still as an upload and opens the video side with that frame as the first frame. Free runs carry the watermarked file over. Paid runs carry the clean one.

The natural next step is the MiniMax H3 video model, which lives on the Hailuo AI video generator page. Motion, duration and resolution are set there, with a separate quote.

The same handoff also lands on the animate a photo tool if you would rather start there.

Plans and credit packs are listed on the pricing page.

Watermark, license, storage, and content rules

Without paid access, downloads carry a small watermark and the license is for personal use. A plan or credit pack removes the watermark and adds a commercial license to images made while paid access is active. Those files keep their license after the plan ends. Files are kept for 7 days on a free account and 180 days with paid access.

We do not use your uploads or results to train models. They go to the model provider only to produce your image. Adult content is refused, and you cannot use another person's photo for an intimate edit without their clear consent. If a file or a charge looks wrong, email [email protected]. Unused purchased credits can be refunded within seven days of the charge, as set out in the refund policy.

MiniMax H3 Image questions

What is MiniMax H3 Image?

It is a text to image model from MiniMax, released on November 10, 2026. MiniMax says it shares the architecture and text encoder of its H3 video model and ships with open weights. On ImageVideoAI it creates images from a prompt, edits up to three reference images, and outputs 2K or 4K files.

Is this a MiniMax image generator I can use online without installing anything?

Yes. The model runs on our servers, so you only need a browser and a signed-in account. You do not need a GPU or a copy of the weights. Write a prompt, add references if you are editing, pick 2K or 4K, and read the quote before you run it. Free accounts get 2K with a watermark.

How long can a MiniMax H3 prompt be?

Up to 4,000 characters, which is twice the 2,000-character limit on the other image models on this site. Use the space for concrete details: subject, setting, light, framing, and any words that must appear in the image. A long prompt is not better by default. Cut lines that repeat or contradict each other.

Can MiniMax AI image results include readable text?

MiniMax says the model places text that follows the meaning of the prompt, such as a sign, a label or a short headline. Put the exact words in quotes and say where they go. Short phrases are easier than paragraphs. Check spelling in the result before you use it, and run the prompt again if a letter is wrong.

Why did the ratio picker disappear?

You attached a reference image. When references are present, the output follows the shape of the first image, so the ratio choice hides. Remove the references to get the full list back: Auto, 1:1, 2:3, 3:2, 3:4, 4:3, 9:16, 16:9 and 21:9. To change the frame of an edit, crop the first reference before uploading it.

What does a free account get with this model?

Sign in first; guests see the sign-up prompt instead of a result. A free account runs this model at 2K, and the download has a watermark. 4K needs a paid plan or a credit pack. New accounts get 30 credits for 30 days, plus 6 daily credits that do not carry over. Failed runs return the credits.

Can I turn the image into a video?

Yes. Press Animate this image under the result. The still becomes the first frame on the video side, where you can pick the MiniMax H3 video model or another one and describe the motion. The video has its own credit price, shown before it runs. The image charge is separate and already settled.

  • Image to Video AI

    Turn any photo into a short AI video. Upload an image, describe the motion and pick a model, from Kling and Veo to Grok Imagine and Vidu. Start free with 30 credits.

  • AI Image Generator

    Type a prompt and get a still. This AI image generator opens on Seedream 5 Flash. Text to image AI, optional reference photos, and a credit price shown first.

  • Seedream 5.5

    Run Seedream 5.5 in the browser, with Seedream 5 Flash and Seedream 5 Lite on the same panel. Text prompts or photo edits at 2K. The quote shows the credits.

  • Nano Banana

    Use Nano Banana in the browser. This page locks the generator to Nano Banana 2 and Nano Banana 2.1 for text prompts and photo edits. The quote shows the credits.

Write the next image

The panel holds one model. Put the full brief in the prompt, attach references only for edits, start at 2K, and check the quote before you spend credits.

Run MiniMax H3 Image again