How do I use reference images, videos and audio?
Reference slots appear under the prompt whenever the selected model accepts input media; drop a file in, or pick one from your Library, and mention it in the prompt with the @Image chip the Workspace inserts.
The slots
A model that accepts input media shows one or more slots under the prompt: an image slot, an end-frame slot, a video slot or an audio slot, depending on the row. Each slot says what it accepts and how many files. Drag a file in, or open the Library picker to reuse something you already generated or uploaded.
Referring to a slot in the prompt
When you add an image the Workspace inserts a chip such as @Image 1. Write the prompt around it — "the woman in @Image 1 walks through the market in @Image 2" — and the Workspace translates the chip into whatever handle the model expects.
What each kind of model wants
- Image-to-video — one image that becomes the first frame. Describe motion only; the picture already fixes the look.
- Image edit — the picture to change, plus an instruction; say what must stay the same.
- Reference-to-video — several images and short clips that lock a character, a place or a style; the model page lists the limits (Vidu Q3 up to 7 images, Gemini Omni 1.1 up to 10 images and 3 clips, and so on).
- Lip-sync and talking-head — an image or a video plus an audio track; the price follows the length of the audio or video you supply.
Limits the Workspace enforces
Reference clips have a per-model maximum length; the Workspace probes the file and refuses one that is too long before any credits are spent. A model that needs a reference will not run without one, and says so.
Frequently asked
Which models take references?
The model picker and every model page say so: image-to-video rows take a first frame, edit rows take the picture to change, reference-to-video rows take several images and short clips.
Why do I see 'This model needs at least one reference image'?
The model cannot run on text alone. Add an image to the slot — the message names how many the model accepts, for example up to 10.
Is there a limit on reference video length?
Yes, per model. The Workspace measures the clip before you generate and tells you the model's maximum in seconds.