MiniMax H3 Max Image-to-Video on fal: Price, Limits and Examples (October 2026)
The cheapest listed image-to-video on fal at $0.0325 per second (768P base). What the Max line accepts, what the price hides at 1080P, where it suits animating a photo and where it does not, and what we have not tested.
MiniMax H3 Max is the second generation of MiniMax's H3 video family, added to fal in late August 2026. Its image-to-video and text-to-video endpoints list at $0.0325 per second, which makes them the cheapest mainstream image-to-video models on the platform, and fal now names them its default pick for both jobs. The number needs reading carefully, because it is a base rate, and the model has a quality ladder above the base.
What the price actually covers
The $0.0325 is quoted for 768P output. The endpoint also offers 480P and 1080P, and 1080P runs a "latent refinement" pass that costs more. fal does not publish the multiplier, so the 1080P price is unknown to us and probably several times higher. If you plan to post at 1080P, price a test clip before you multiply anything.
| Length | At the 768P base rate |
|---|---|
| 5 seconds | $0.16 |
| 10 seconds | $0.33 |
| 15 seconds (the maximum) | $0.49 |
Compare that with Kling 3.0 Pro image-to-video and HappyHorse 1.1 image-to-video at $0.14 per second (720p for HappyHorse), and with the older first-generation H3 at $0.26. On paper the Max line is a quarter of the price of the models most people reach for. On paper.
What it accepts
- Duration: from just under one second (0.92) up to 15 seconds.
- Resolution: 480P, 768P or 1080P.
- Frames: a first frame and an optional last frame, so you can describe where the motion should end up.
- Audio: a
target_audio_urlinput of at least two seconds, which lets the clip follow a supplied track. - Reference-to-video sibling: a separate endpoint takes up to twelve files in any mix of photos, video clips and audio, at $0.05 per second base.
There is also a Turbo tier at half the price; we compare it in H3 Max Turbo vs H3 Max.
What it is good for
A lot of useful video contains no recognisable person, and for that the cost advantage is real:
- Product motion: a bottle that rotates under changing light, a mug with steam, packaging that floats. Five seconds is sixteen cents.
- Scenery and establishing shots: a street at dusk, waves, clouds, a slow push-in on a photo of a room.
- Animating illustrations and cartoon characters, where there is no face to keep consistent.
- Cover and thumbnail motion: a still that moves slightly is more likely to stop a scroll than the still alone.
In each case the workflow is the same: make the still first, where a retry costs cents (see Nano Banana 2.1), approve it, and only then animate it. Deciding the composition in images and the motion in video is the cheapest habit in this whole category.
Where to be careful
Faces. MiniMax's moderation, error code 1026, rejects named public figures and protected characters, and fal's recommender gives no identity guarantee for any model. We have not put an ordinary person's photo through H3 Max, so we cannot tell you whether the face stays recognisable over a 10-second clip. For a talking portrait we still use HappyHorse or Kling, and for a talking portrait driven by a recording we look at H3 Max Lip Sync and OmniHuman 2.0.
Short clips win. Every model drifts as the clip lengthens. Five to eight seconds, cut together, beats one 15-second take, and it costs the same.
Dates. The late-August release and the prices are from fal's listing as read on 18 October 2026. fal's pricing API does not state what an endpoint meters, so confirm with one paid run before you build a budget on the number.
Where it fits in our pipeline
H3 Max image-to-video is a candidate for two jobs we run: animating a photo when no audio is supplied, and the face-less scenes of a Reel that must be regenerated rather than kept. Both are cost-sensitive, so a model at a quarter of the price is worth testing; neither is switched on until a test says the result is as good. The same caution applies to FLUX.3 video, which is our current choice for face-less scenes.
Common questions
Why is it so much cheaper than Kling or HappyHorse?
The listed $0.0325 per second is a base rate for 768P output, while the other two are quoted at 720p. The higher-resolution multipliers for H3 Max are not published, so the gap at 1080P is unknown. It is also a newer, less proven model.
Can I use it for a talking portrait?
It can animate a face, and it accepts an audio track of at least two seconds, but we have not tested how well it holds an ordinary person's face. For a talking portrait with a known record, use a talking-avatar model.
What is the maximum length?
Fifteen seconds per clip, with a minimum just under one second. Five to eight seconds, cut together, drifts less than one long take.
Does it refuse real people?
MiniMax's moderation rejects named public figures and protected characters (error code 1026). Whether it accepts ordinary photos reliably is untested by us.
How to do this without code: Animating a still is the Animate photo tool, and H3 Max is a candidate we are testing for it. In AutoHustle this is the Animate photo tool: one photo, an optional sentence and a length slider (about 2 credits per second on the default model, quoted before you press Generate). See How to animate a photo and Animate a photo.
Related reading: Kling Native 4K on fal: $0.42 per Second, and When 4K Is Worth Paying For (October 2026) · Wan 3.0 Prime on fal: 30-Second Clips, Reference Videos and the Real Price by Resolution (October 2026)
For developers: the fal endpoints
| Model | fal endpoint |
|---|---|
| MiniMax H3 Max image-to-video | minimax/h3-max/image-to-video |
| MiniMax H3 Max text-to-video | minimax/h3-max/text-to-video |
| MiniMax H3 Max reference-to-video | minimax/h3-max/reference-to-video |
| MiniMax H3 Max Turbo image-to-video | minimax/h3-max-turbo/image-to-video |
Prices are fal's listed base rates for 768P, read on 18 October 2026. Duration 0.92–15 s; resolution 480P / 768P / 1080P; optional last-frame image and a target_audio_url (2 s or longer).