MiniMax H3 Image · Text to Image
Text to Image AI turns a description into a still image. Explore single-image settings, introductory free access and the limits that apply.
MiniMax H3 Image is connected through ComfyUI. Sign in to generate an image.
Sign in to see your private images.
Use this workflow when the studio is available.
One text-to-image output; 1K/2K area levels; 1:1, 16:9, 9:16, 4:3, 3:4 ratios; JPEG, PNG and WebP. Prompts are limited to 2,000 characters, with prompt enhancement disabled in this release.
1K/2K describe target pixel area, not a fixed long edge. Ratio changes the output shape; the actual width and height are shown with the result. For a square, the official area targets are 1024×1024 and 2048×2048; other ratios use different dimensions.
Free during the introductory period. Image generation costs 0 credits.
High demand may lead to longer queue times. Thank you for your patience.
The official identifier is MiniMax-H3-Image. It is a still-image model, distinct from MiniMax H3 video generation. FreeVideo AI runs a quantized H3 Image model through its connected ComfyUI service. It is an independent application, not the official MiniMax website. The table above describes this interface, not every capability in the official model documentation.
Model weights and API services have applicable terms. Do not treat downloadable weights as proof of hosted-service permission, or this page as a guarantee that an output is commercially licensed.
Official model cardOfficial parameter documentationModel weights license
The prompts below are writing exercises, not verified generation examples or guaranteed results.
Subject + setting + composition + lighting + visual style + essential constraints.
Practice one subject, a simple background and a clear light source.
A ceramic teapot on a pale wooden table beside a window, soft morning light, close-up still life, muted green and cream colors, no text.
Change only “soft morning light” to “warm side lighting at sunset”; check shadows and material appearance.
Practice a clear scene, one visual style and a limited palette.
A small cabin beside a mountain lake at dawn, wide composition, mist above the water, watercolor illustration, soft blue and peach palette.
Keep the scene and change only the medium to pencil drawing. Check whether composition and object placement remain consistent.
Practice describing materials and a background without inventing a product endorsement.
An unbranded amber glass bottle on a matte beige surface, centered composition, soft side lighting, plain studio background, product photograph, no lettering.
Change only the surface color. Inspect reflections, bottle shape and any unwanted lettering; do not assume the constraints succeeded.
Local ComfyUI · MiniMax H3 Image reference-guided generation. A product study in materials, lighting and advertising composition, not a text-only result. Reference editing is not available in the public studio.
Original prompt：Edit <Picture 1>. Recreate this luxury skincare product advertisement as closely as possible. Preserve the same single clear cylindrical serum bottle, pearly ivory inner container, thick transparent base, polished gold neck and trim, white pump beneath a transparent cap. Preserve the bottle tilted with its top leaning left, hovering just above the glossy black circular pedestal and its reflection. Preserve the suspended clear water droplets, delicate mist, dark charcoal background, blurred green leaves at left and warm beige bokeh at upper right. Match the composition, camera angle, scale, lighting and realistic glass, gold and water materials. Keep the full bottle and cap visible. Preserve the label wording and typography from the reference: LUMINOSE, Hydrating Serum, 30ml. No new objects, no additional advertising text, no watermark. Landscape 4:3 luxury product photograph.
MiniMax-H3-Image · Requested ratio 4:3 · 2K · 2304×1728 · PNG
Generated on：
Observed limitations：Generated with the locally installed quantized H3 Image model in ComfyUI, using one supplied reference. The composition closely follows that reference; droplets, reflections and fine label lettering differ. Fine background texture is visible. Native 2304×1728 output, without an external upscaler or retouching. This is a reference-guided showcase, not evidence that the public text-only studio reproduces a specific product.
This release does not provide image-to-image, image editing, batches, multiple references, custom sizes, 4K, masked editing or transparent backgrounds. Results can contain incorrect text, distorted details or an unexpected composition. Inspect each image rather than assuming that a prompt was fulfilled.
The first-release design saves task records and images to the signed-in account, with permission checks for preview and download. Private results are not automatically added to public examples. Prompts are sent to our connected ComfyUI generation service; avoid sensitive personal information.
The workflow is AI Image → Image to Video → AI Video. Manually download a result, check the video tool’s upload requirements and upload it to the separate Image to Video tool. This is a manual handoff, not a new one-click feature; video credits still apply.
There is no single best choice for every task. Compare actual outputs, trial restrictions, supported sizes, privacy and usage terms. FreeVideo AI has no published image-quality benchmark; its image tool is available under introductory trial conditions.
A text-to-image tool turns a written description into a still image. Describe one subject and scene, choose composition and lighting, then select supported output settings. Use the studio when available and inspect the downloaded result.
Use subject + setting + composition + lighting + visual style + essential constraints. Start with one clear image, change one detail at a time and check the output rather than assuming the request was followed. The practice prompts on this page have not been verified by generation.
Free during the introductory period. Image generation costs 0 credits. Sign-in, one active task per account, submission pacing and queue-capacity limits apply. There is no promise of unlimited access or daily credit refreshes. Existing video credit costs do not change.
The first release is limited to one image, 1K or 2K area levels, ratios 1:1, 16:9, 9:16, 4:3 and 3:4, and JPEG, PNG or WebP. An area level is not a fixed long-edge size; the result shows its actual pixel dimensions. Custom dimensions and 4K are not exposed here.
MiniMax-H3-Image is the official model name for a still-image generation model in the MiniMax H3 family. It is distinct from the video workflow on this site. Official model capabilities are broader than this first-release website interface and do not establish that every feature is offered here.
When the studio is available, sign in, enter a prompt, select supported settings and submit one task. Wait for the result before creating another. This interface fixes the model and requests one image; it does not ask visitors for an upstream API key.
The first release only supports text-to-image and prompts up to 2,000 characters. It does not offer image-to-image, image editing, batch generation, multiple references, masked editing or transparent backgrounds. Text, details and composition can be incorrect, and requests may queue or fail. No guaranteed quality, completion time or commercial license is promised.
High demand may lead to longer queue times. Thank you for your patience. Tasks are processed through a queue. Submit once and follow the saved status; a full queue may require trying later. No fixed wait time is promised.
Yes. Your account is used to save private history and check permission to preview and download results. Personal images are not automatically published as examples.
No. Text-to-image produces a still image; Image to Video animates an input image. You can use a manually downloaded image as video input after checking the video tool’s upload requirements. No automatic transfer or one-click connection is provided here.
After an interrupted request, the upstream service may still be generating. The task is not automatically resubmitted. Keep the record and wait for reconciliation rather than repeatedly creating the same request.
FreeVideo AI is an independent service, not the official MiniMax website. Model capabilities and features available here may differ. Official model documentation
FreeVideo AI