Vidu: Q2 Text-to-Video

Vidu Q2 creates video from a written scene through the text-to-video profile selected by viduq2. Its required inputs are the exact model identifier and a prompt describing the scene. Use it for a short product concept, an establishing view or a character action whose appearance and movement can be specified in words. This configured method is text-to-video; its description does not assign a separate image-reference or video-editing mode to the selector.

Build the brief around a visible sequence. Identify the subject, location, action, shot size, camera movement and final composition. For a product concept, describe an object on a tabletop, light moving across its material and a slow camera approach that ends on a recognizable detail. For a location scene, establish the environment before moving toward the focal point. Keep the action appropriate to the chosen clip length so the scene has room for its intended ending.

The prompt limit is 2,000 characters. Put the essential subject and action first, then add the lighting, material, environment and camera details that change the desired result. Specify the visual style in the prompt. The Q2-specific guide says the separate style and movement_amplitude parameters do not take effect for this model, so the general or anime preset and the motion-amplitude options are not substitutes for a clear scene description.

The model-specific duration setting accepts whole seconds from one to ten. The Q2-specific resolution choices are 540p, 720p and 1080p, with 720p as the documented default. Choose duration and resolution together according to the action and destination. A short reveal and a longer movement need different timing; increasing the requested duration should serve the visible sequence rather than simply stretching an unchanged brief. Inspect the returned clip's actual dimensions and length when preparing the final placement.

The documented aspect_ratio choices are 16:9, 9:16, 3:4, 4:3 and 1:1, with 16:9 as the default. The guide explicitly includes 3:4 and 4:3 for the Q2 family. Pick the frame before generation and describe a composition suited to it: keep a principal subject readable in a portrait view, or use the width of a landscape frame for the surrounding setting. Leave the closing view clear enough for the intended page, presentation or later caption treatment.

The seed field controls variation. Omitting it or supplying 0 selects a random value; a nonzero value selects the documented fixed-seed option for repeatable generation. Preserve the returned seed with the prompt and settings when comparing versions. The shared operation also documents audio and off_peak controls specifically for Q3 models. Those Q3 options are not capabilities assigned to this Q2 profile; plan this request around its documented text input and video output.

Generation is asynchronous. Submission returns task_id and an initial state, with created used for a newly created task. Retain that task identity while following processing. The documented result query uses state=success for a successful result and returns a creations array. Each entry has its video URL in creations[].url; save every returned video rather than assuming the task always contains exactly one creation.

The result documentation also reports the actual video's duration, fps and resolution width and height. Keep those reported values with the completed clip instead of treating an example's frame rate or dimensions as a fixed Q2 specification. The task-creation response includes a credits value alongside the request configuration. Retain that reported task metadata and the completed identity with the result, so each video remains connected to its generation settings.

Review the entire sequence against the original brief: subject appearance, object shape, movement direction, camera continuity and the intended final view. For a revision, identify the concrete detail that missed the brief and preserve the successful parts of the scene. Keep the prompt, exact selector, settings, task identity and completed videos together so a later comparison is based on the actual artifacts.