thomaszx85531
New member
Vidu Q4 is an AI video model that works from images rather than from text alone. According to the product page, the Preview release offers two input modes: image-to-video, where one image plus a prompt becomes a clip, and reference-to-video, where several reference images and voice references are combined into a single video.
The specifications listed on the page are specific enough to plan around. Image-to-video clips run from 3 to 16 seconds and keep the aspect ratio of the uploaded image. Reference-to-video accepts 1 to 15 images for characters, scenes and style, plus up to 3 voice references, and can hold a multi-shot scene for up to 16 seconds. Output resolutions are 540p, 720p, 1080p, 2K and 4K.
Audio is generated with the picture instead of being added afterwards. The page states that sound and picture are produced together: if a line of dialogue is written into the prompt, the clip comes back with that line lip-synced, along with sound effects. Voice references are used to clone a character voice with the emotion of the performance.
A practical way to use either mode is to keep the first pass short. Prepare the still or the reference set, write the prompt so it states who is in frame and what they do, pick the resolution and ratio, then generate a short clip and review the motion before spending credits on a longer take. With reference-to-video, the useful check is whether the same character still reads as the same person across shots.
There are two limits worth knowing before starting. The page says text-to-video is not supported yet, so an image is always required, and the model is currently labelled Preview. The API model identifier is viduq4-preview, and the pricing table lists credits per second for each resolution, so cost scales with clip length and output size rather than with the number of attempts alone.
For teams already producing stills, that combination is a reasonable fit: the input is an asset they already have, the prompt stays short, and the audio step no longer needs a separate pass. The trade-off is that the workflow is image-first, which means the quality of the source frame matters more than the length of the prompt.
The specifications listed on the page are specific enough to plan around. Image-to-video clips run from 3 to 16 seconds and keep the aspect ratio of the uploaded image. Reference-to-video accepts 1 to 15 images for characters, scenes and style, plus up to 3 voice references, and can hold a multi-shot scene for up to 16 seconds. Output resolutions are 540p, 720p, 1080p, 2K and 4K.
Audio is generated with the picture instead of being added afterwards. The page states that sound and picture are produced together: if a line of dialogue is written into the prompt, the clip comes back with that line lip-synced, along with sound effects. Voice references are used to clone a character voice with the emotion of the performance.
A practical way to use either mode is to keep the first pass short. Prepare the still or the reference set, write the prompt so it states who is in frame and what they do, pick the resolution and ratio, then generate a short clip and review the motion before spending credits on a longer take. With reference-to-video, the useful check is whether the same character still reads as the same person across shots.
There are two limits worth knowing before starting. The page says text-to-video is not supported yet, so an image is always required, and the model is currently labelled Preview. The API model identifier is viduq4-preview, and the pricing table lists credits per second for each resolution, so cost scales with clip length and output size rather than with the number of attempts alone.
For teams already producing stills, that combination is a reasonable fit: the input is an asset they already have, the prompt stays short, and the audio step no longer needs a separate pass. The trade-off is that the workflow is image-first, which means the quality of the source frame matters more than the length of the prompt.