AI Video Generator
An AI video generator is a tool that creates moving footage from a prompt, a photo, or a clip you already have. Use the editor on the left: pick the job at the top, pick a model, and the price for that combination shows on the button before anything is charged. Six jobs are covered — text to video, image to video, editing a clip, replacing a character, copying motion, and lip sync — across eight models from Seedance, Wan, MiniMax, Grok and Google.
Image to video on Wan 2.2 is running today at 12 credits per second in 480p. The other routes show their real controls and their real price but cannot be submitted yet; they are marked so you are not left clicking a button that does nothing.
Six things this editor does
- Text to video
Describe a scene and get a clip.
- Image to video
Animate a still you upload.
- Edit video
Change an existing clip with a prompt.
- Replace character
Swap who appears in the shot.
- Motion sync
Drive a still with motion from a video.
- Lip sync
Match mouth movement to audio.
Models and what they cost
| Model | Good for | Per clip |
|---|---|---|
| Seedance 2.5 | Highest fidelity, highest price | 320 |
| Seedance 2 | Cinematic look, half the cost of 2.5 | 160 |
| Wan 3.0 | Good value for everyday clips | 45 |
| MiniMax H3 | Dynamic multi-shot video with native audio | 116 |
| Grok Video 1.5 | Cheapest way to test an idea | 20 |
| Wan 2.7 | The apimodels catalog has Wan 2.2 and Wan 3.0 but no 2.7. | Soon |
| Wan 2.2 | Image to video, 5 or 8 seconds | from 60 |
| VEO 3.1 | Three variants in the catalog at $0.07, $0.206 and $1.00 — pick one before listing a price. | Soon |
Prices are per generated clip, charged when the job starts and refunded automatically if it fails. Wan 2.2 is billed per second instead — 12 credits a second at 480p, 24 at 720p.
How to generate a video
- 1. Pick the job. The row of tabs at the top of the editor decides what the tool asks you for. Image to video wants a picture; lip sync wants a clip and an audio file.
- 2. Pick a model. Each one lists its price next to its name. Cheap models are for finding out whether the idea works; expensive ones are for the take you keep.
- 3. Describe the motion, not the picture. The model can already see your image. It needs to know what should move — “the camera pushes in slowly as she turns” beats a description of the scene.
- 4. Set length, shape and size. Start at the shortest length and the smallest size. It costs less and tells you within about two minutes whether the motion is what you wanted.
- 5. Create. The button shows the exact cost before you press it. You can leave the page while it runs; the finished clip waits in your history.
When not to use this
- You need more than ten seconds. Every model here tops out around ten. Generating four clips and cutting them together in an editor gives a better result than asking any of them for a long take.
- You need the same character to look identical across shots. These models re-imagine the subject on every run. Faces drift between clips, and no prompt fixes that.
- You need readable text in frame. Words rendered by a video model come out garbled more often than not. Add captions afterwards.
- You want to improve footage you already have. Generation invents new frames; it does not clean up real ones. Use the video upscaler or video restoration for that.
- The subject is a real person who has not agreed. Not a quality limit — a line we hold.
Questions people ask
What is an AI video generator?
An AI video generator is a model that produces video frames from a description, a still image, or an existing clip. This editor covers six of those jobs in one place: text to video, image to video, editing a clip you already have, replacing the character in a shot, copying motion onto a still, and matching lips to audio.
Which model should I pick?
Start with the cheapest one that can do the job, then move up only if the result is not good enough. Grok Video 1.5 at 20 credits is for testing whether an idea works at all; Wan 3.0 at 45 is the everyday choice; Seedance 2.5 at 320 is worth it when the clip is going somewhere people will look closely. The price sits next to every model name so the comparison is in front of you before you spend anything.
How much does one clip cost?
Between 10 and 320 credits depending on the model, with image to video on Wan 2.2 priced per second instead: 12 credits a second at 480p and 24 at 720p, so a 5-second 480p clip is 60 credits. Credits are charged when the job starts and refunded automatically if it fails.
How long can the video be?
Five to ten seconds, depending on the model. Longer is not better: faces and fine detail drift the further the model gets from the first frame, so a 5-second clip usually holds together better than a 10-second one of the same scene. Generate short, then extend in an editor if you need more.
Is there a watermark?
No. The output is a plain MP4 with nothing burned into it. Most models produce silent video — MiniMax H3 is the one here that generates its own audio track.
What image works best for image to video?
A sharp, well-lit picture between 360 and 2000 pixels on each side and under 10 MB. The editor checks all of that in your browser before uploading, so a file that is too big or too small is rejected immediately rather than after a wait. Soft or heavily compressed images give the model less to work with and the motion comes out mushy.
Can I generate video of a real person?
Only someone who agreed to it. Replace character and lip sync in particular make it easy to put words and movement onto a face — do not use them on a person who has not said yes, and do not use them to make anyone appear to say something they did not. This is the one limit here that is not about output quality.
Why do some models say "Soon"?
Because we do not have a price we can stand behind yet. Wan 2.7 is not carried by our upstream catalog at all, and VEO 3.1 exists there in three variants whose prices differ by more than ten times. Listing a guess and charging something else would be worse than saying it is not ready.
Related tools
Prices and model list last checked 20 September 2026.