Vidu Settings Explained: Modes, Resolution, Duration
Vidu’s key settings are the generation mode (text, image or reference to video), the model version, resolution, duration and aspect ratio, plus the reference or first-frame images you attach. Choose the mode first, test at the lowest resolution and shortest duration, then raise them only for the clip you keep, because resolution and duration are what drive credit cost.
Vidu (vidu.com, by Shengshu) has a compact settings panel, which is a relief after some video tools, but each control changes both the result and the credit cost. This guide explains what every setting does, which to decide first, and a testing order that stops you spending high-resolution credits on clips that were never going to work. Labels in the current interface may differ slightly from the descriptions here, and the options available depend on the model version selected.
Mode and model version: the first decisions
Vidu offers three ways to generate, and the mode you pick determines what every other setting means.
- Text-to-video. Prompt only. Most freedom, least control. Good for atmosphere plates, abstract motion and exploring an idea before you have any stills.
- Image-to-video. An uploaded still becomes the first frame, and the prompt describes motion. Use it when you have an approved image and want that exact shot to move. Some versions also accept an end frame so the clip travels between two stills.
- Reference-to-video. Uploaded images act as ingredients whose identity is preserved while the prompt builds a new scene. This is Vidu’s distinctive mode and the right one for the same character or product across several shots; the reference-to-video guide covers it in depth.
Decide the mode from the job, not from habit. A storyboard with fixed frames wants image-to-video; a presenter in five locations wants references.
Several model versions are available in the interface and the list changes with each release. Newer versions typically follow prompts more closely, hold subject identity better and move more naturally, but they may cost more per clip and offer different resolution or duration options. Check which version is selected before generating. For a set of clips that must match, keep the same version throughout; switching mid-project changes colour and motion character in ways that are hard to hide in the edit.
Resolution, duration and aspect ratio
These three decide cost and fit.
- Resolution. A choice of output sizes; the exact options depend on the version. Higher resolution adds detail and cost but does not fix a bad prompt. Test low, finish high.
- Duration. Clip length in seconds, usually a choice between a short and a longer option. Short clips hold identity and composition better; longer ones have more room to drift and cost more. Most edited sequences use clips of a few seconds anyway.
- Aspect ratio. Landscape, portrait and square options. Set it to the delivery format before generating. Cropping a landscape clip to portrait throws away most of the frame and often the subject with it.
If a setting you expect is missing, it is usually because the selected model version does not support it; try another version before assuming the feature is gone.
Inputs, prompt and prompt enhancement
Depending on the mode, the upload slots mean different things. In image-to-video the slot is a first frame (and sometimes an end frame); in reference-to-video it is one or more subject references. Upload sharp, evenly lit images with no text overlays. For first frames, match the aspect ratio you have chosen so nothing is cropped. For references, one subject per image on a plain background works best.
The prompt box is the setting people forget is a setting. In text-to-video it must carry subject, action, camera, setting and light. In image-to-video it should only describe motion and camera. In reference-to-video it should point at each reference and build the scene. Some versions offer automatic prompt enhancement, which rewrites your text before generation; it can help a first attempt but reduces control, so switch it off once you know what you want. Options for a negative prompt or a seed, where present, are worth using: a fixed seed lets you change one word and see its effect, and a negative prompt is the place for "text, watermark, blur".
Settings at a glance
| Setting | What it does | Start with |
|---|---|---|
| Mode | Text, image or reference to video | Match the job: fixed frame vs recurring subject |
| Model version | Which Vidu model generates | Latest version; keep it fixed across a project |
| Resolution | Output size and detail | Lowest for tests, highest for the keeper |
| Duration | Clip length | Shortest option until the motion is right |
| Aspect ratio | Frame shape | The delivery format |
| First frame / end frame | Fixed start (and optional end) in image-to-video | Start frame only |
| Reference images | Subjects kept consistent in reference-to-video | One clean image per subject |
| Prompt enhancement | Platform rewrites the prompt, where offered | Off for control |
| Negative prompt / seed | Exclusions and repeatability, where offered | "text, watermark, blur"; fixed seed when comparing |
Credit costs scale with resolution, duration and version. Check the platform’s current pricing page rather than older posts.
A credit-saving testing order
- Lock mode and aspect ratio first. They cannot be changed afterwards without regenerating.
- Test at the lowest resolution and shortest duration. Judge motion, identity and composition only.
- Change one thing per generation. Prompt wording, then camera, then version. Fixed seed if available.
- Raise resolution only for the approved clip. Then, if the shot truly needs it, the longer duration.
- Write down the recipe. Mode, version, resolution, duration, ratio, references and prompt for every clip you keep. Settings are not embedded in the download, so the note is your only record.
Try it
Mode: image-to-video. Model: latest. Resolution: lowest. Duration: shortest. Aspect ratio: 16:9.
First frame: a lighthouse on a rocky headland at dusk, waves below.
Prompt: Waves roll in and break against the rocks, spray catching the last light. The lighthouse beam sweeps slowly across frame. Static wide shot, calm natural motion.
Negative prompt: text, watermark, blur.
Once the motion is right, re-run at the highest resolution with nothing else changed. Then try the same first frame with the longer duration and compare how much the waves and beam drift; that tells you whether the extra seconds are worth the credits for this shot.