Vidu Q4 Preview

Vidu Q4 Preview is Vidu's new flagship video model preview, recorded in its official update log on October 7, 2026. Its documented API supports image-to-video and reference-to-video, sound generation, 3–16-second clips and output up to 4K. The request model ID is viduq4-preview.

The useful question is whether its reference and camera controls can produce the shot you need at an acceptable cost. Vidu positions the update around character performance, motion, visual effects and cinematic camera language. Those are reasons to design a trial; they do not establish a measured success rate.

Choose the input mode before writing the prompt

The image-to-video specification and reference-to-video specification define different starting points:

Use image-to-video when the opening composition is already decided. Use reference-to-video when several assets must guide the result. Assign each reference a clear role, such as character, clothing, product or location, so conflicting images do not obscure the brief.

Two boundaries matter. The model map says 1–16 seconds, while the endpoint schemas and price table say 3–16; use the endpoint range when constructing a request. Q4 Preview's checked pages do not document text-only generation, a dedicated first-and-last-frame mode or video-reference input. Other Vidu tools do not automatically become Q4 Preview inputs. Reference-to-video also says subject-library invocation is not supported for now.

Start with a ten-second shot you can judge

A useful first trial isolates a few requirements. For example, prepare a clear image of a person in a blue coat holding a red notebook. Use assets you have permission to upload, including any voice reference. Choose a supported ten-second setting and describe this sequence:

1. 0–3 seconds: medium shot by a station window; the person notices something outside and pauses.
2. 3–7 seconds: the camera moves closer as the person turns; keep the face, blue coat and red notebook unchanged.
3. 7–10 seconds: hold on the expression as surprise becomes a small smile; no extra people or on-screen text.
4. Sound: quiet station ambience and clothing movement; no music or dialogue for this first pass.

This is a proposed evaluation prompt, not a clip we generated. Watch identity and object shape across the turn, the continuity of the camera movement and whether the emotional change happens at the requested moment. Add a short spoken line in a separate trial to evaluate lip sync without changing the visual brief.

Review every successful take against the same criteria: identity, action, camera, audio and delivery quality. Keep the settings and rejected takes in the record. A beautiful opening frame cannot compensate for a changed face halfway through a clip intended for a character scene.

Official API pricing and the cost of a usable take

Vidu's API pricing page lists RMB 0.03125 per credit. Its Q4 Preview table applies the same rates to image-to-video and reference-to-video:

Calculation: clip cost = duration × credits per second × RMB 0.03125. These are published API credit costs, not a web subscription, introductory offer or OmniaKey quote. Keep the source currency; check your account's applicable pricing and any separately selected services before a batch.

Vidu says failed generations, including moderation failures, consume no credits. A completed clip you reject creatively is a different case. For a hypothetical batch of ten successful ten-second 1080p clips, the base generation cost is RMB 75. If four pass your review, the cost is RMB 18.75 per accepted clip. Four accepted takes is an assumption, not a measured Q4 success rate; editing time is additional.

This is why our recommendation is to settle the motion and composition at a lower resolution first, then evaluate the delivery resolution. Recheck the final result: an earlier accepted draft does not guarantee the next generation will preserve every detail.

A minimal request to Vidu's own API

The following example follows Vidu's image-to-video documentation and requires a Vidu-issued API key. Replace the image URL with an accessible asset. The English prompt is kept identical across this guide's language versions.

curl --request POST 'https://api.vidu.com/ent/v2/img2video' \
  --header "Authorization: Token ${VIDU_API_KEY}" \
  --header 'Content-Type: application/json' \
  --data '{
    "model": "viduq4-preview",
    "images": ["https://example.com/reference.jpg"],
    "prompt": "Slow camera push-in; the subject turns toward the window.",
    "is_rec": false,
    "duration": 5,
    "resolution": "720p",
    "audio": true
  }'

Set is_rec: false when testing your own prompt. The documented default is true, which asks the system to recommend and use a prompt. Leaving that default on can change what your experiment is evaluating.

Save the returned task_id and query GET /ent/v2/tasks/{id}/creations until state becomes success or failed, with a polling deadline. The I2V page documents Authorization: Token for this query; the R2V page currently shows Authorization: Bearer in its query example. Confirm that discrepancy with Vidu before generalizing an integration across both modes. This example has been checked against the documentation, not executed against a paid endpoint.

The documented creation URLs expire after 24 hours. Download successful output for your own storage instead of treating the returned URL as a permanent asset address.

Is Q4 Preview a good fit for your workflow?

It is a reasonable candidate to evaluate when a short scene depends on a recognizable subject, several visual references or coordinated sound and camera movement. Treat those as selection criteria, not proven quality advantages. If your requirement is text-only input or an exact ending frame, start by finding a model with that explicit API mode.

For a broader shortlist, use our video API model comparison. For an OmniaKey integration, check the available video models and their current prices. Verify the exact model ID and provider contract there; the Vidu request above uses Vidu's own endpoint and credentials.

Frequently asked questions

Is Vidu Q4 Preview the final Q4 release?

The official update and API documentation identify this version as Preview. Use its current identifier and limits without assuming they describe a future final release.

Does 4K support mean every result is production-ready?

No. 4K is a documented output option. It does not certify face consistency, accurate text, physical motion or synchronized dialogue. Evaluate those requirements on the finished clip.

Is there a universal free or promotional API rate?

The verified API source uses the credit schedule above. Web plans and temporary offers have their own conditions. Check the actual offer, resolution and account eligibility before comparing prices.