Image to video
Upload a source image or first/last frames, review the opening composition, then unlock the full animation when it is ready.
Output controls
Set the model, duration and frame before previewing and rendering.
Hailuo 3 · 768P or native 2K full video with synchronized audio and multimodal reference control.
First frame (optional)
Last frame
The output ratio follows the uploaded frame automatically.
Verified H3 outputs
Start from a real MiniMax example
Six MiniMax-published prompt and output pairs. Reference templates state exactly how many source images you still need to add.
Visual review uses 0 credits · Full render uses 25 credits after purchase
Describe the scene before generating.
How MiniMax H3 image to video works
Image to video starts from a frame you already have. Upload a single source image and H3 animates forward from it, or supply both a first and a last frame and the model generates the motion that connects them. That second mode is the one worth understanding — it turns H3 from a generator into an in-betweener, which is how you get predictable, art-directed motion instead of a surprise.
Choosing between one frame and two
- 1
One frame: open-ended motion
Upload a single image and describe what should happen. H3 decides where the shot ends. Best when you want the model to surprise you, or when only the starting composition matters.
- 2
Two frames: controlled transition
Supply a first and last frame and H3 renders the path between them. Use this when the end state matters — a product rotating to a specific angle, a character arriving at a mark, a logo resolving.
- 3
Write the prompt for the motion, not the image
The frames already describe what things look like. Spend the prompt on how they move: "camera pushes in as the petals open", not another description of the flower.
Getting clean results from a source image
Most disappointing image-to-video results trace back to the input rather than the prompt. Four things matter:
- Match the aspect ratio of your image to the output ratio. A 1:1 source rendered at 16:9 forces H3 to invent the edges.
- Use a clean, well-lit source. H3 amplifies compression artifacts and motion blur already present in the frame.
- When using two frames, keep the subject recognizably the same in both. Wildly different framing gives the model no coherent path to interpolate.
- JPEG or PNG, up to 30 MB per image, two images maximum for the first/last frame pair.
A worked example
For a product shot with a first and last frame set, this prompt directs the connecting motion:
The bottle rotates slowly clockwise on a matte black surface, camera holds locked-off, studio key light sweeps across the label as it turns, soft room tone with a faint glass resonance.
The frames fix the start and end angles, so the prompt only has to describe speed, camera behaviour and lighting movement. That division of labour is what makes two-frame generation repeatable.
Frequently asked questions
- Can I set both a first and a last frame?
- Yes. Supply both and MiniMax H3 generates the motion connecting them. Supply only a first frame and the model chooses its own end state based on your prompt.
- What image formats and sizes are supported?
- JPEG and PNG, up to 30 MB per image, with a maximum of two images when using the first/last frame pair.
- Why does my animation look different from the source image?
- Usually an aspect ratio mismatch. If your source is 1:1 and you render at 16:9, the model has to invent the side edges. Match the ratios and keep the source clean and well-lit.