Wan 3.0 Prime on fal: 30-Second Clips, Reference Videos and the Real Price by Resolution (October 2026)
Wan 3.0 Prime lists a $0.05 base rate that rises to roughly $0.28 per second at 1080p. What its image, text and reference-to-video endpoints accept, how reference-video seconds are billed, and when it beats the alternatives.
Wan 3.0 Prime is Alibaba's August 2026 video generation line, added to fal on 24 August with image-to-video, text-to-video and reference-to-video endpoints. It made headlines for one number: $0.05 per second, the lowest price among credible full-quality video models. That number is a base unit, though, and the interesting facts are what it rises to at the resolutions people actually publish, and the one feature that sets Prime apart: it can take video references and clips up to 30 seconds long.
The price, by resolution
fal lists $0.05 as the base. The price per output second rises with resolution; the figures below come from fal's reference-to-video listing and third-party price trackers, and the image and text endpoints are reported to follow the same ladder.
| Resolution | Price per second | 5 seconds | 10 seconds |
|---|---|---|---|
| 480p | about $0.068 | $0.34 | $0.68 |
| 720p | about $0.14 | $0.70 | $1.40 |
| 1080p (the default) | about $0.28 | $1.40 | $2.80 |
At 720p, Wan 3.0 Prime costs the same as Kling 3.0 Pro and HappyHorse 1.1. At 1080p it costs twice that. The headline $0.05 is only a fair comparison against other models quoted at their base rate. Always compare at the resolution you will publish; our resolution-multiplier warning applies here as much as anywhere.
For reference-to-video there is a second charge: the seconds of any reference video are billed in addition to the output, a fact stated in the schema. A 5-second output at 720p with a 5-second reference clip therefore bills about 10 seconds' worth, roughly $1.40, not the $0.70 the output alone suggests.
What the three endpoints accept
Image-to-video and text-to-video: up to 30 seconds per call, at 480p, 720p or 1080p, with 9:16 available. The 30-second ceiling is the longest single generation among the mainstream models, and it is the reason to look at Prime for storytelling or looping backgrounds rather than for a face.
Reference-to-video:
- up to 10 images,
- up to 5 videos (at least 16 frames per second, 15 seconds in total),
- up to 5 audio clips (15 seconds in total),
- duration from 2 to 30 seconds, or left to follow the references,
- aspect ratio
adaptiveor 9:16, and an optional prompt-expansion switch.
Mixing photo, motion and sound references in one call is the capability that matters. It is a generation from references, not an edit: the model builds a new clip guided by your video's motion and does not guarantee it follows the source frame by frame, lip movement included.
Where it fits
- No-face scenes and backgrounds at 480p to 720p, where the lower rungs of the ladder are cheap.
- Long single generations, up to 30 seconds, for ambient or looping material.
- Experiments with motion references, if you want a new clip guided by a video without the strictness of an editor.
Where it does not fit: a face that must stay recognisable (no one has verified it, and third-party coverage mentions likeness rights), and replacing a person in an existing clip, which needs a video-edit model (Recast, Kling O3 edit, HappyHorse video-edit).
What we know and do not know
fal's pricing data states the $0.05 base for every endpoint and the resolution ladder for reference-to-video; the image and text rows follow the same pattern in third-party listings but we have not confirmed it ourselves. We planned a paid test of the reference-to-video endpoint on our own assets (five seconds of output plus a five-second reference, about $1.40 at 720p) and it has not run, so we make no statement about identity or motion fidelity. The earlier Wan 2.7 line is a different generation; we describe its image-to-video endpoint in Kling vs Wan, where Wan 2.7 stays relevant for a driving audio track.
The sensible way to use it today
Treat the $0.05 as a ceiling on what you might pay, not a price. Generate one clip at 480p to check the idea, a second at 720p to check the motion, and only then decide whether 1080p is justified. If you want one clip with your face in it and no budget arithmetic, it is simpler to let a tool choose the model, which is what the Studio does.
Common questions
Is Wan 3.0 Prime really $0.05 a second?
That is fal's base unit. Third-party trackers report the per-second price rising to about $0.14 at 720p and $0.28 at 1080p, which is the default.
What are the reference limits?
For reference-to-video, up to 10 images, 5 videos (15 seconds in total) and 5 audio clips (15 seconds in total).
Do reference videos cost extra?
Yes. The schema says the seconds of reference video are billed in addition to the output.
Can it replace a person in a video?
It generates a new clip guided by references; it does not edit the source and does not guarantee to follow it frame by frame. For replacing a person, use a video-edit model.
How to do this without code: For a still with motion, the photo tool picks the model and quotes the price. 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: MiniMax H3 Max Image-to-Video on fal: Price, Limits and Examples (October 2026) · LTX 2.5 on fal: Open-Weights 22B Video With Multishot and 4K, and a Price Unit Nobody Explains (October 2026)
For developers: the fal endpoints
| Model | fal endpoint |
|---|---|
| Wan 3.0 Prime image-to-video | alibaba/wan-3.0-prime/image-to-video |
| Wan 3.0 Prime text-to-video | alibaba/wan-3.0-prime/text-to-video |
| Wan 3.0 Prime reference-to-video | alibaba/wan-3.0-prime/reference-to-video |
Base unit $0.05; per-second ladder $0.068 / $0.14 / $0.28 (480p / 720p / 1080p); reference-video seconds billed additionally. Read on 18 October 2026.