Video
Vidu Q3 Drama
Video
Available on Ofrai — chat, images, and video in one place with shared credits.
Example prompts
Choose an example to load its prompt and recommended settings into the Vidu Q3 Drama creator above.
How it works
Create with Vidu Q3 Drama in three steps
- 01
Choose your inputs
Keep Vidu Q3 Drama selected, then add a prompt and any supported reference assets.
- 02
Set the output
Choose the aspect ratio, resolution, and available Video options before generating.
- 03
Generate and refine
Review the result, adjust the prompt or settings, and generate another version from the same place.
Same honesty as the ad page, and here it matters even more, because narrative work is exactly where an undocumented model name would do the most damage. The documentation documents three Q3 identifiers and no drama-specific one. What it does document is two capabilities that narrative briefs live on, and they sit on different models: audio-video sync and storyboard video generation on Pro, and best-in-family multi-camera consistency with smart shot switching and multi-reference image generation on the base `viduq3`.
So a narrative job has a real answer here — it just is not a fourth model. It is a decision between which of the two documented behaviours the sequence actually needs.
No drama model; two narrative capabilities on two models
The documentation gives you no drama entry, so the useful structure is this: which narrative requirement are you actually expressing? If it is planning shots and keeping generated sound synchronised with them, the reference puts storyboard video generation and audio-video sync on Pro, and Pro is priced at roughly twice turbo for that tier.
If it is holding one character or one object recognisable across several distinct angles, the reference puts multi-camera consistency, smart shot switching and multi-reference image generation on the base model — and the base model is also the only Q3 identifier that accepts more than two reference images, up to seven, and the only one where that parameter is mandatory. Recognising that your requirement falls on one side or the other is most of the work of picking a model for a sequence.
Aimed vaguely, it is the easiest way to render a scene twice and get it wrong both times.
A sequence is a shot list before it is a render
Narrative work fails more often from structure than from image quality, and the parameters make the structural decision concrete. Duration is 1 to 16 seconds on Pro and turbo, defaulting to 5, while the base model starts at 3 seconds — which means a two-second insert, a reaction beat, a cutaway that exists only to change the rhythm, is expressible on the first pair and not on the second. Decide the cut points before rendering anything, hold each shot to a length the family can actually produce, and assemble afterwards.
A single continuous generation asked to cover several beats tends to deliver a take that is fine for the first half and drifts after it, which is the classic symptom of asking a clip model to do sequence work it was not built for.
Continuity across cuts is a reference-image problem
The mechanism the family documents for keeping a subject consistent is reference input, and the count you can pass depends on the model. On Pro and turbo, the parameter is optional and the count is the control: none gives text to video, one image anchors the first frame, two bound the clip at both ends. On the base `viduq3` the parameter is required and accepts up to seven images, which is the documented path for a cast or a product that has to be reconciled across several setups.
That is the honest answer to 'how do I keep my character from changing between shots': on the base model, feed the references; on Pro, anchor the first frame and design around it. Nothing in the reference promises identity preservation from text alone, and a narrative pipeline built on that assumption will discover the problem in the edit rather than in the render.
Generated audio: cast the lines, dub when you must
Audio defaults to true across all Q3 models, and when enabled the system generates speech and background sound effects according to the prompt. For narrative work that is a genuine feature and a genuine trap. It is genuine because a dialogue scene can be requested in words, and the prompt ceiling of 5000 characters is ample for subject, action, scene, camera, style and the lines to be heard.
It is a trap because the speech is derived from your prompt, not from a recorded performance, so what you get is generated dialogue rather than cast dialogue — do not plan a sequence around lip-sync precision that the reference only claims for Pro's audio-video sync, and check that claim against your own frames before committing. The parameter takes false as well, which is the route when the scene will be dubbed or scored in post.
Resolution and pacing decisions that belong together
Three resolutions are available — 540p, 720p and 1080p, with 720p the default — and for narrative work the choice interacts with editing rhythm rather than sitting apart from it. A scene assembled from short shots is judged on whether the cuts land, and 720p is enough to judge that; texture and facial detail are questions you answer once, on the final render, not on every attempt.
Aspect ratio takes 16:9, 9:16 and 1:1, with the family noted as supporting arbitrary ratios, and passing a reference image often lets that image decide the ratio, which in a sequence means one early request can quietly set the frame for everything downstream. Set the ratio explicitly on the first shot of a sequence and keep it identical across the rest.
Running a multi-shot night on an asynchronous endpoint
Submission is asynchronous and returns a task identifier, which is what makes a sequence feasible at all: shots are independent requests, so they can be dispatched as a set and collected as they return rather than rendered in strict sequence. Two habits make that workable. First, key every identifier to its beat — 'shot 3, wide, 6s' — so a render that fails is diagnosable without opening the prompt, which matters when each prompt is up to 5000 characters.
Second, keep a per-shot record of which model, resolution and duration produced an approved take, because the natural instinct at the edit stage is to re-render one shot slightly differently and lose the reason the rest of the scene was approved. Retakes are where a sequence quietly becomes inconsistent.
The one-sentence rule for narrative work
If one identity has to hold across many angles, use the base `viduq3` with multiple references; if a single storyboarded shot needs generated sound synchronised to picture, use Pro; if you are cutting twenty rough versions of the same beat, stay on Turbo and escalate only the keeper.
That is the whole decision, and it is worth making once at the top of the project rather than per shot, because the two paths have different floors — a one-second beat exists on the first and third options and not the second — and a sequence planned without that in mind discovers the gap at the assembly stage, which is the expensive place to discover it.
FAQ
- Is there a drama-specific Vidu Q3 model?
- Not in the documentation, which documents three Q3 identifiers: `viduq3-pro`, `viduq3-turbo` and `viduq3`. The two capabilities a narrative brief needs are documented elsewhere — storyboard video generation and audio-video sync on Pro, multi-camera consistency with multi-reference images on the base model — so the choice is between those documented behaviours, not between a fourth tier.
- How do I keep a character consistent across several shots?
- Use the base `viduq3`, which is the documented path for multi-reference input and accepts up to seven images, with at least one required. Pro and turbo accept one image as the first frame or two as a first and last frame pair, and stop there.
- Can I render a two-second insert in a sequence?
- On Pro and turbo, yes — both accept 1 to 16 seconds, defaulting to 5. The base `viduq3` model shares the 16-second ceiling but floors at 3 seconds, so a one or two second beat is not expressible there and a sequence designed around such beats must be planned against that.
- Should narrative scenes use the generated audio?
- It is available by default on every Q3 model and derived from the prompt, which is useful for blocking dialogue scenes. Treat it as generated rather than cast dialogue, verify the audio-video sync claim on Pro against your own frames, and set the audio parameter to false when the scene will be dubbed or scored in post.
FAQ
- Can I use Vidu Q3 Drama directly on this page?
- Yes. The creator above is the working Ofrai studio with Vidu Q3 Drama selected by default.
- Where are my finished results saved?
- Signed-in generations are kept in your Ofrai library so you can download or revisit them.
- Which prompts work best with Vidu Q3 Drama?
- Describe the subject, setting, visual style, lighting, and camera or motion clearly. The examples above provide practical starting points.
- Can I edit the example prompts before generating?
- Yes. Load any example, then change its wording, aspect ratio, references, or other supported settings in the creator.
Related models
Compare other Video models that can handle similar creative work to Vidu Q3 Drama.
Related guides
Tools
Commercial use depends on the selected model and model terms.





