AI video generator: animate a photo or turn text into a clip

Seedance by ByteDance — several versions, up to the flagship 2.5 — Kling 3.0 Turbo, Hailuo 02 and H3 by MiniMax, Wan 3.0 by Alibaba, Gemini Omni Flash by Google, Grok Imagine 1.0 and 1.5 by xAI, and Vidu Q4 by Shengshu, which works from a photo only. They all sit in one list under the Engine button, and each model shows its price, length ceiling, sound and wait right there. You do not need an account with the model makers: the run starts here and is paid in this site’s tokens.

In price, maximum length, whether they produce sound, and how long you wait — all four are written on the engine itself. The spread is wide: the simplest one returns a five-second clip in about half a minute, while the flagship takes around three minutes for the same length and costs many times more. The cheapest way to compare their look is a short run at low resolution.

The price comes from the engine, the resolution and the length at once, and they multiply: thirty seconds at 1080p on the flagship costs hundreds of times more than five seconds at 480p on the simplest model. That is why the number is recalculated as you tap, rather than shown at the end.

Yes. The work runs on the server and the finished clip lands in My generations, where it stays for 30 days. While a job is running, a note about it shows on any page of the site, so you can come back from another device.

The tokens go back to your balance automatically — we charge only for a delivered file. A run that is canceled halfway or rejected by the content filter costs nothing.

A machine-readable note about synthetic origin is written into the MP4 metadata — the EU AI Act requires it and platforms expect to find it. It does not touch the picture or the sound, and nothing is printed over the image.

Yes, up to three minutes. No model draws that much at once — one generation tops out around half a minute — so the video is built from scenes: pick «Long video» in the length menu, describe your idea, and we split it into scenes you can edit before you start. Each scene starts on the last frame of the one before, so the motion carries on from the same point, and we join the finished scenes into one MP4. The price is the sum of the scenes; if the engine fails a scene, you get the video made of the finished ones and are charged only for those. Long videos are made by the three engines whose seam between scenes we measured: Wan 3.0, MiniMax H3 and Gemini Omni Flash. Each scene goes to the engine as one continuous shot, and we check every finished scene frame by frame: if a scene opens with a cut, we regenerate it once at our own expense.