> ## Documentation Index
> Fetch the complete documentation index at: https://docs.zerogen.com/llms.txt
> Use this file to discover all available pages before exploring further.

# Frames and references

> Guide a generation with start and end frames and reference media.

**References** guide what a generation looks like: a character, a costume, a
location or a style. **Frames** fix the exact first or last frame of a video.

## Add references

Choose **Add** in the **References** tray. The picker shows everything in the
project.

![The reference picker, with the project's folders on the left, filters for uploads, generations, derived assets, favorites, type, resolution, creator and labels, an upload area, and a grid of images.](https://mintcdn.com/zero-gen/m4I7fPs1rAXZSMjU/images/generate/reference-picker.png?fit=max\&auto=format\&n=m4I7fPs1rAXZSMjU\&q=85\&s=c592f6178f3d140f40c50de91be02181)

* Browse **All assets** or open a folder.
* Filter by uploads, generations or derived assets, favorites, type,
  resolution, who created it, or labels.
* Drag a file onto **Drag or upload media** to upload and use it.
* Select several, then choose **Done**.

If your organization has [LucidLink](https://docs.zerogen.com/integrations/lucidlink.md), the picker also
offers **External sources**.

## Use as a reference or frame

To use something you've already made, open it in the viewer, or hover over it,
and choose **Use as…** › **Use as reference**, **Use as start frame** or
**Use as end frame**. Only the options the current model accepts appear.

## How many references a model takes

The count above the tray, such as **2/16**, shows how many you've added out of
the model's limit. Each model has its own limits. For example, GPT Image 2.5
takes 16 images, and Seedance 2.5 takes up to 50: 30 images, 10 videos and 10
audio clips.

Video and audio references show their length. Models also limit how long each
clip, and all clips together, can be.

## Mark a real person

Image and video references have a person button on their tile in the tray,
**Mark as containing a real person**. Turn it on for footage or photos of a
real, recognizable person. ZeroGen remembers the setting on the asset.

Seedance refuses a recognizable person who isn't marked. When Seedance says
which reference contains a person, ZeroGen marks it and submits the generation
again for you, once. The new generation shows a **Resubmitted** label.

## Edit part of an image with a mask

Image models that edit selected areas show **Mask** in the tray. Add a mask the
exact size of your first reference image. If the mask has transparency, the
transparent areas are edited; otherwise, the white areas are.

## Set start and end frames

In the video generator, choose **Start** under **Frames** to set the exact
first frame. On models that support it, choose **End** for the last frame. You
need a start frame before you can set an end frame.

![The video generator with empty Start and End frames above the References tray.](https://mintcdn.com/zero-gen/m4I7fPs1rAXZSMjU/images/generate/video-generator.png?fit=max\&auto=format\&n=m4I7fPs1rAXZSMjU\&q=85\&s=e623d7e52c3a3fb5789f529f892c4f12)

## Pin images to times with keyframes

Models that take keyframes, such as FLUX.3, add **Keyframe** to the **Frames**
row. Choose it, pick an image, then type the time in seconds where that image
should appear.

Each time must land on a frame of the video, within its duration. Under each
keyframe, ZeroGen shows the frame number, or **Between frames** or **Past the
end** until you change the time.


This documentation is built and hosted on [Mintlify](https://mintlify.com), a developer documentation platform.