Reference mode

Reference mode builds the video from a set of images instead of one. Upload the product from a few angles, a packaging shot, a colour or texture swatch, a location — and the model composes a scene that stays faithful to all of them.

When to use it

• The product is hard to read from a single photo (multiple parts, a set, a bundle, a range of colours).
• You want a specific style or setting — give it a scene reference and a style reference alongside the product.
• Consistency matters across a longer take: more references mean fewer hallucinated details.

If one photo already says everything, Single Image is cheaper to set up; for a two-frame before/after use Transition.

How many images

• Version 2.0 — up to 9 reference images.
• Version 2.5 — up to 30 reference images.

At least one image is required. The counter on the panel shows how many slots remain for the tier you have selected. See Quality tiers for the rest of the differences between the two engines.

Real faces on Version 2.5

Version 2.5 refuses reference images that contain real human faces. The Generate page warns you before you submit; if your set includes a person, switch to Version 2.0 (which allows faces) or crop the person out. Either way you still need permission from anyone shown — see the Acceptable Use Policy.

Aspect ratio on Version 2.5

On Version 2.5 the output ratio follows your source images automatically (there is no ratio picker). On Version 2.0 you choose the ratio as usual.

Tips

• Lead with the hero product image; add supporting references after it.
• Keep references consistent — one lighting style, one colourway — unless the variation is the point.
• Name what each reference is for in the prompt ("use the second image for the background"). Write with AI works in Reference mode too and reads every image you attached.
• Uploads accept JPG, PNG, WebP, AVIF and HEIC up to 25 MB each — see Image uploads & formats.
• Browse Reference-mode renders on the Examples page.