MiniMax H3 Image to Video AI

Frame 001 · yours

002003004005

Generated from what you described

From there MiniMax H3 image to video AI moves the camera and the subject the way you described it, for any whole number of seconds from 4 to 15, at 24 fps in 768P or 2K. JPG, PNG, WEBP and HEIC all work, so a picture straight off your phone is fine.

Your photo is frame one

001

Your photo is frame one

Any whole second

4–15 s

Any whole second

Frame rate

24 fps

Frame rate

Resolution

768P / 2K

Resolution

• No style transfer
• No init strength slider
• No re-describing your photo

• Start here

  I2VA
  

  One photo as the first frame. Your video opens on your image and moves forward from it. This is the one you want for making a product turn on its stand, a portrait start speaking, or a landscape breathe.

  How to write it

  Describe what happens next, not what is already in the picture — the model can already see the picture. A prompt that just re-lists the contents of your photo wastes the only instruction you get.

  Try this
• Rare

  End here

  L2VA
  

  One photo as the last frame. Give H3 the ending instead and it works out a plausible opening, then converges onto your image on the final frame. This is how you build a reveal: the lid lifts and lands exactly on the product shot you already have.

  How to write it

  Write it backwards from the ending: say what state things must have been in for your image to be the final frame, then let the motion resolve into it.

  Try this

• Start → End

  FL2VA
  

  Two photos, a start and an end. H3 fills in the path between them. Before and after, a transformation, a change of scene. The closer your two frames are in subject, framing and light, the cleaner that path comes out.

  How to write it

  If they are wildly different, say what happens in between in so many words, or H3 will invent a transition you did not ask for.

  Try this

More worked setups in prompt ideas — every one ships with its full structured prompt, and the prompt generator will write all three official fields from one sentence.

Same input photo · two prompts

The input

The coat changes weave and the hairline moves. Nothing in the prompt said they should not.

Run this one

Same camera move, same length, same seed conditions. The clause is the only change.

Run this one

The only difference between them

The subject — red wool coat, dark fringe, silver ring on the left hand — keeps the same face, hair, clothing and colours throughout; only the camera and the described motion change.

Appearance lock

Type what must not change. The clause below is what gets appended to your prompt.

Goes on the end of your prompt

The subject keeps the same face, hair, clothing and colours throughout; only the camera and the described motion change.

Add to prompt

If you need the same person across several separate clips rather than within one, a first frame is the wrong tool for the job. That is an identity problem, not a composition problem, and reference to video is built for it.

iPhone photos work. HEIC and HEIF go straight in, which is worth saying out loud because most tools quietly make you convert first.

Drop a file in and you will see its dimensions, ratio and size checked against those limits before you spend anything — a rejected upload should cost you a second, not a generation. Over 30 MB, export at lower quality rather than shrinking the pixels. Outside the ratio range, crop rather than pad — letterbox bars become part of the frame and get animated too. The full guide covers the API's own first_frame and last_frame fields if you are sending these yourself.

Pricing · 2K on every paid plan · No watermark