Vidu Q3

Positioning Overview

Vidu Q3's reference-to-video route is designed for short video production that requires controllable visuals and stable subjects. Developers submit 1 to 7 reference images plus a prompt to generate a 3 to 15 second video. Reference images are referenced in the prompt as Image1, Image2 in array order, and are used to constrain characters, objects, or visual style. Output supports 720P and 1080P tiers with ten aspect ratios in total, can enable audio generated synchronously with the visuals, and also supports fixed seeds to reproduce results. It is suitable for teams that need rapid multi-version trial footage, such as ad creatives, product demos, and short character clips.

Core Highlights

• Multi-image references constrain subjects and style — Pass in up to 7 reference images at a time and reference them in order as Image1, Image2 in the prompt to constrain characters, objects, or visual style separately.
• Visuals and audio generated in one pass — When audio is enabled, matching sound is generated synchronously with the visuals; when disabled by default, a silent video is output for later custom dubbing.
• Duration and aspect ratio combine as needed — Duration can be 3 to 15 seconds; the 720P and 1080P tiers each include five aspect ratios: 16:9, 4:3, 1:1, 3:4, and 9:16.
• Fixed seeds for easy reproduction and iteration — Supports specifying a random seed from 0 to 2147483647; the same configuration can reproduce results, making it easy to compare the effects of prompt changes.

Capability Description

Multi-reference-image-driven generation: Accepts 1 to 7 publicly accessible JPG, PNG, or WEBP images, up to 50MB each, with aspect ratios from 1:4 to 4:1, used to lock in characters, object appearance, or overall style.

Prompt controls scene and camera movement: A prompt of up to 5000 characters can describe the scene, actions, camera movement, and sound, and uses the Image1, Image2 indices to point to specific reference images for per-image role assignment.

Optional audio generation: The audio switch controls whether matching audio is generated with the visuals; when off, a silent video is output for integration with your own dubbing or music workflow.

Resolution and aspect ratio selection: The size parameter provides 720P and 1080P tiers, each covering landscape, portrait, square, and two vertical ratios, adapting to the final-video specifications of different placements.

Duration and watermark control: duration takes values from 3 to 15 seconds; watermark optionally adds an AI-generated mark in the lower right corner; billing is based on output seconds.

Use Cases

• Brand and media-buying teams: multiple product images + selling-point copy — Produce short ad clips with specified aspect ratios for social platform placements and e-commerce detail page display assets.
• Character content creators: character design sheets + plot descriptions — Generate story short shots with consistent character appearance, used for short drama clips, anime-style concept validation, and storyboard previews.
• Creative teams: style reference images + proposal descriptions — Rapidly produce multiple trial versions for A/B comparison, use fixed seeds to control variables, and move to formal production after selection.
• Product demo production: product photos + usage scenario descriptions — Turn static product images into demo videos with camera movement for official websites, launch events, and training materials.

Limitations and Selection Advice

A single output is limited to 3 to 15 seconds and cannot directly cover long videos or continuous multi-shot narratives; the final video still requires manual editing and splicing. Reference images must be publicly accessible and no more than 50MB each; images in private storage need to be converted to accessible addresses first. Audio is an optional generation item; when dialogue or dubbing must align precisely with the visuals, silent output followed by separate production is more controllable. Currently only reference-image-driven generation is supported; pure text input is not accepted. For ideas without reference materials, please choose the text-to-video route.