Vidu Q4 AI Video Generator

Vidu is a video model family from China. Q4 is its fourth quality generation, and Kie.ai, the API provider AgentReel uses, serves it as Vidu Q4 Preview (viduq4-preview). Kie.ai describes it as built on Q3 with more expressive character performance, multi-reference control, native audio and output up to 4K. Because it is a preview, features may change.

On AgentReel it always needs at least one image: there is no text-only mode. One image runs the image-to-video endpoint, and two to 15 run reference-to-video.

Sources: the Kie.ai model page and the Kie.ai pricing page, read on 2 November 2026.

Use one image when the picture is already the shot and only needs motion: a portrait that turns, a landscape where clouds drift. The output follows that picture's frame, so crop it first.

Use references when the shot does not exist yet. Upload clean, well-lit images of each element, ideally on plain backgrounds, then write the scene: who is where, what happens, how the camera moves. In this mode you pick the ratio, so a vertical 9:16 ad can come from square product photos.

Describe action and camera, not the pictures. "The woman from the portraits walks into the bakery, picks up the green box and smiles at the camera, slow push-in, warm morning light, soft shop chatter" gives the model a sequence, a move and a sound bed. Prompts can be up to 20,000 characters.

Resolution changes the price more than anything else here: 4K costs several times the 540p rate per second. Test the idea at 540p or 720p, check the motion and the references, then render once at the size you need. Length scales the price linearly, so a short test is cheap.

On AgentReel a clip costs 9 to 250 credits: the per-second rate of the resolution times the duration. The number of reference images does not change the price. The table on this page lists every combination, and a failed clip is refunded automatically.

Credits come from a plan or credit pack; free check-in credits do not pay for AI video. See the credits and pricing guide.

Kling 4.0 takes text alone or a first and last frame, at 720p or 1080p. Veo 3.1 also reaches 4K, with clips of 8 seconds. Seedance 2.5 runs text-to-video with sound. Choose this model when you need many references in one shot, the longest single clip, or 4K with audio. The model overview compares them all.