Guide AI video with reference assets

Reference-to-video can use multiple assets to communicate appearance, composition, motion, or style. The references guide a new generation; they do not splice the uploaded files together or guarantee that every visible detail will be copied.

Choose assets with a clear purpose and remove contradictory examples. A small, coherent reference set is easier to direct than a large group that shows different subjects, lighting, or visual identities.

Supported asset types and limits vary substantially. Wan 2.6 R2V currently accepts a combined set of up to five reference images or videos, including at most three videos. Other routes may accept images only or expose different limits.

Seedance 2 reference workflows can also expose supported video and audio reference slots, while Veo Fast and Kling Omni use their own image-reference rules. The studio shows the applicable count, format, duration, and size validation before generation.

• Use image references for appearance and composition
• Use supported video references for motion context
• Use reference audio only when the selected workflow exposes that slot

The prompt should identify the action, camera behavior, setting, and the role of the supplied references. Avoid assuming that upload order alone communicates which visual details matter most.

Some reference workflows allow an optional prompt and others require one. Even when text is optional, a concise direction can make the intended relationship between assets easier to evaluate.

Upload only material you own or are authorized to use, especially when a source contains a recognizable person, protected character, logo, voice, or copyrighted footage. Consent and usage rights remain important when AI transforms the material.

Check identity, product details, text, continuity, and unintended similarity before publishing. Reference guidance can improve consistency, but it is not a promise of exact reproduction.