This page explains how to operate the MiniMax H3 generator: choose a tab, prepare the required files, upload and tag references, select output settings, submit the job, and review the result. It does not repeat the full prompt-writing specification. When you need field syntax or reference-label rules, open the MiniMax H3 Prompt Guide.
Start with the correct generator tab
| Generator tab | Choose it when you start with | What the interface expects |
|---|---|---|
| Text to Video | A written idea and no required source media | Prompt only |
| First & Last Frame | Two exact endpoint images | One first-frame image and one last-frame image |
| Reference | Images, videos, or audio that should guide the result | At least one image or video, plus optional reference audio |
Quick start from opening the generator to download
Open the generator and select MiniMax H3
Use the generator on the homepage, sign in when asked, then confirm that MiniMax H3 is selected in the Model control before preparing the job.
Choose one of the three input tabs
Pick Text to Video, First & Last Frame, or Reference. The visible upload controls change with the selected tab, so make this choice before writing or attaching files.
Upload only the files the job needs
Wait until every thumbnail finishes uploading. In Reference, the current interface accepts up to 9 images, 3 videos, and 3 audio clips, with no more than 12 files in total.
Write the instruction and connect references
Describe the desired result. In Reference, type @ in the prompt and select Image, Video, or Audio tags so the instruction points to the correct uploaded file.
Set ratio, resolution, and duration
Choose the output ratio, select 2K or 768P, then set a duration from 4 to 15 seconds. Reference mode also offers Adaptive ratio.
Check credits, generate, and review
The button shows the required credits before submission. Generate the clip, play the result with sound, inspect the important moments, then download or revise one variable and try again.
How to use each generator tab
Text to Video tab
Create from a written idea
Use this tab when no uploaded asset must control identity, composition, motion, or sound. It is the fastest way to test a scene concept because the form contains only the prompt and output settings.
Select Text to Video
Choose the Text tab after selecting MiniMax H3. No upload area is required for this workflow.
Describe one readable video beat
State the subject, environment, visible action, camera behavior, lighting, and important sound. Keep the requested action achievable within 4-15 seconds.
Choose output settings
Select the aspect ratio for the destination, choose 2K or 768P, then use the duration slider. A shorter first test is easier to evaluate and costs fewer output credits.
Generate and inspect the first failure
Check the opening composition first, then action, camera, ending, and sound. If something fails, change that instruction before rewriting the entire prompt.
Practical tip
Use Text to Video for exploration. Move to Reference when a specific face, product, visual style, motion clip, or audio characteristic must carry into the result.
First & Last Frame tab
Connect two exact endpoint images
This tab is for a controlled transition. Both upload slots are required by the current interface. The first image defines the opening frame and the second defines the final frame.
Upload the first frame
Choose the opening image and wait for its preview to appear. Use an image with a clear subject, readable composition, and enough room for the intended movement.
Upload the last frame
Choose the exact ending image. Keep the subject, camera side, and environment compatible unless the transformation itself is the point of the video.
Describe only the bridge
Explain what moves and changes between the two frames. Do not ask for an unrelated third scene, and avoid adding cuts unless the transition truly requires one.
Preview the arrival at the end frame
After rendering, check whether motion begins naturally from the first image and whether pose, object state, lighting, and framing settle cleanly into the second image.
Practical tip
If you have only one image and want it used as a general appearance reference, use Reference instead of leaving one frame slot empty.
Reference tab
Generate from mixed reference media
Use Reference when uploaded media should guide the target video without acting only as two exact endpoints. The interface provides separate Image, Video, and Audio upload controls and assigns each file a numbered tag.
Add reference images
Upload up to 9 images. Supported image formats shown by the current form are JPEG, PNG, WebP, BMP, TIFF, and GIF, with each file under 10 MB.
Add up to 3 reference videos
Upload MP4 or MOV files, each under 50 MB. Use short clips with one clear motion, camera, edit, or effect role instead of several near-duplicate examples.
Add audio only when it has a job
Upload up to 3 MP3 or WAV files under 10 MB each. Audio cannot be submitted alone; the Reference job must also contain at least one image or video.
Insert the numbered @ tags
Click in the prompt, type @, and select the uploaded Image, Video, or Audio. Confirm that every tag points to the intended thumbnail before generating.
Use Adaptive ratio when source framing varies
Reference mode exposes Adaptive alongside the standard aspect ratios. Choose a fixed destination ratio when delivery requirements are already known.
Practical tip
A strong Reference job gives each file one clear responsibility. Remove redundant media before adding more instructions.
Three outputs to study before your first job
These videos demonstrate three different output goals. Use the notes to choose a generator tab and plan your inputs. They are outcome studies, not claims that one prompt or one setting guarantees the same result.
Recommended start: Reference
UGC product advertisement
Use this workflow when a recognizable presenter, product, packaging, or brand look matters more than free exploration.
- Upload the clearest presenter or product images first.
- Add one short motion reference only when a gesture or camera move is difficult to describe.
- Use @ tags to assign appearance and motion to the correct files.
Review: Check face and product consistency, hand interaction, label readability, and whether the final product moment receives enough screen time.
Recommended start: Text to Video
Cinematic scene
Use this workflow to explore a self-contained scene when no exact character or frame must be preserved.
- Choose one main subject and one primary action.
- Use one camera idea instead of stacking several competing moves.
- Set the environment, lighting, and sound so the scene feels intentional from the first frame.
Review: Check whether the opening establishes the subject quickly, the camera remains readable, and the final state resolves within the selected duration.
Recommended start: First & Last Frame
Dynamic poster transition
Use this workflow when the opening artwork and final layout are already designed and the main task is controlling the transition between them.
- Export both endpoint images at the same aspect ratio.
- Keep key typography and subject placement compatible across the pair.
- Describe the transition path, timing emphasis, and final hold instead of redescribing both images.
Review: Check the first-frame match, transition continuity, text stability, and whether the video lands on the intended final artwork without a late jump.
Final checks before and after generation
This checklist is about operating the generator successfully. Use the Prompt Guide only when the problem is the wording or structure of the prompt itself.
Before upload
- The selected tab matches the starting material and control you actually need.
- Every file is supported, within the displayed size limits, and easy to recognize from its thumbnail.
- Reference clips are short and each demonstrates one useful behavior.
- First and last frames use compatible dimensions and composition.
Before Generate
- All uploads have finished and no thumbnail is still showing progress.
- Every @ tag points to the intended reference file.
- Model, ratio, resolution, and 4-15 second duration are correct.
- The credit total on the Generate button matches the job you intend to submit.
After rendering
- Play the complete result with sound before downloading.
- Check the opening, main action, ending, reference consistency, and audio timing in that order.
- Save the successful setup and change only one failed variable for the next attempt.
- Download the result only after confirming the correct version and aspect ratio.
Using the MiniMax H3 generator
Which tabs are currently available in the generator?
The current MiniMax H3 interface provides Reference, First & Last Frame, and Text to Video. Reference accepts mixed media, First & Last Frame requires both endpoint images, and Text to Video requires only a prompt.
Can I upload only one image in First & Last Frame?
No. The current form validates both upload slots before submission. If you have one image that should guide appearance or composition, use the Reference tab and insert its @Image tag in the instruction.
How many reference files can I upload?
The current Reference interface accepts up to 9 images, 3 videos, and 3 audio clips, with a maximum of 12 mixed files. Audio requires at least one image or video in the same job.
Which file formats does the Reference tab accept?
Images accept JPEG, PNG, WebP, BMP, TIFF, and GIF. Videos accept MP4 and MOV. Audio accepts MP3 and WAV. The form currently limits images and audio to 10 MB each and videos to 50 MB each.
Why does typing @ matter in Reference?
Each uploaded file receives an Image, Video, or Audio number. Selecting that tag after typing @ connects the instruction to the correct thumbnail and prevents ambiguous phrases such as use the second clip.
Which resolution and duration can I choose?
MiniMax H3 currently exposes 2K and 768P output options. The duration slider runs from 4 to 15 seconds. Available ratios depend on the selected tab, and Reference includes Adaptive.
When can I download the video?
After generation completes, play the full result, confirm the correct version and aspect ratio, then use the Download control on the result panel.
How many credits will the job use?
The generator calculates the current total from the selected output settings and reference inputs. Review the credit number displayed on the Generate button and the breakdown below it before submitting. Check the Pricing Page for the current credit packages.
Open the generator and build your first job
Choose a tab, upload only the references you need, check the settings and credit total, then generate and review the result.