MiniMax H3
H3: first and last frames, multimodal references and editing, from a model MiniMax publishes as open-weight.
ProMiniMaxTeam pick
In short
Evaluate when the shot has a clear start or end and several references have different roles. Test action timing and subject continuity on one representative clip before committing a longer sequence. Keep the source material organised so another production route can be evaluated with the same brief.
What it can do
- Text/image-to-video
- First and last frames
- Multimodal references
- Video editing on documented H3 workflow
Limits to plan for
- H3 and H3 Max have different output constraints.
- Do not transpose older Hailuo API examples to the current model.
MiniMax's current video documentation describes H3 and H3 Max, with generation, frame guidance and multimodal references; editing is documented in the H3 workflow. The two models have different output constraints, so a family name alone is insufficient for production planning.
Design a representative shot
Choose an action with a clear beginning or ending, then assign a role to each reference. Test a meaningful turn, an object interaction or another moment that is likely to expose continuity problems in the real brief. Use this one clip to decide what needs further testing before planning a sequence.
Watch how the subject changes when references compete. A frame that protects the opening composition may not preserve the subject through movement. Compare the clip against the storyboard and identify the usable section, including any frames needed around the cut.
Keep the workflow reproducible
Organise references, action notes and approvals so the same brief can be assessed through another route if required. Inspect every moment where objects touch, the subject turns or important details become hidden and reappear.
Confirm account and regional conditions for the documented MiniMax API route. Older Hailuo examples should not be assumed to match H3. For a reference-led sequence, explore 4AI's AI-video production service with the intended action and delivery requirements.
Sources checked 27 October 2026: MiniMax video generation.
Before you commit
Inspect every moment where references conflict, objects touch or the subject turns.
Compare it with
Other models in the same category, nearest tier first.
ProLuma Ray
LumaRay 3.2
Ray 3.2 for shots with set start, end or middle frames, plus editing, reframing and extension.
Read the profile
Team pickProGemini Omni
GoogleGemini Omni Flash (stable model ID gemini-omni-1.1-flash)
Make and edit short clips in conversation, starting from text, images or video.
Read the profile
ProWan
AlibabaWan 3.0 (wan3.0-video) and its speed-optimized version (wan3.0-video-prime) on Alibaba Cloud Model Studio
Alibaba's Wan 3.0: text to video, first and last frames, multimodal references and audio controls.
Read the profile
Where we use it
Read next
- Choosing an AI video model for a campaign: generation, editing and deliveryA production-led method for choosing an AI video workflow shot by shot, checking references, motion, sound and the work needed to deliver campaign-ready masters, with routes checked on 27 October 2026.
- Directing AI video so it does not look generatedThe models are good enough. What separates a usable commercial from an obvious AI clip is almost entirely what happens before and after generation.
- Can you use AI video to market property in Dubai?What AI video genuinely does for a Dubai property listing, what still needs a camera, published AED prices for every route, and the permit, QR code and portal rules that decide what you can publish.
Choosing tools for a real brief?
Tell us what you need to make, launch or improve. We scope the work first, then choose the tools around it.
No pitch decks until we understand the problem.
hello@4ai.ae