> ## Documentation Index
> Fetch the complete documentation index at: https://captions.ai/llms.txt
> Use this file to discover all available pages before exploring further.
# Chat-Based Editing
> Edit Captions projects with natural-language prompts. Attach images or videos, chain requests, and iterate with follow-ups to change music, fonts, and more.
We’re rolling out improvements to Chat on iOS. You may notice changes to your experience over the next few weeks.
## Overview
Chat-based editing lets you make edits to your project by describing what you want in plain language. For example, the prompt "Add chill background music" will automatically add a chill background track to your project.
On Web, Chat can handle nearly anything you’d do manually in the editor — including motion graphics for vector elements, kinetic captions, product callouts, smooth camera moves, tracked effects, beat-synced graphics, and clean transitions.
## How it works
After opening a project, tap **Chat to edit** to open the chat.
Describe the changes you want. A prompt can be a sequence of actions — for example, replacing images and adding music. You can also attach additional media assets by tapping the **paperclip** icon next to the chat prompt.
After you send the prompt, Captions starts planning the edits and shares updates as it updates the project.
## Example prompts
* [Add chill background music](/help/docs/audio/music)
* Remove/replace background with the X picture
* Change the font to comic sans
* [Generate a b-roll about duck soaps and add it to the first shot](/help/docs/visual/media)
* [Edit captions](/help/docs/captions/add-captions)
## Attach media to your prompt
* Attach images (e.g., background replacements).
* Attach videos (e.g., b-roll to insert).
* Combine attachments with text in a single prompt to give the editor more context.
Prompts are flexible. You can chain multiple requests in one message, or iterate with follow-ups like "make it softer" or "use a different font".
## How it works
From the left toolbar, select **Chat**.
Type the edits you want Captions to make. You can describe a single change or a sequence of actions in one message.
Click elements on the timeline to attach them to your prompt. For example, click a piece of text if you want to add a transition to that specific element. You can select multiple elements to reference in your prompt by holding down the Shift key.
Use the **Add annotations** button to mark parts of the video or timeline. You can add prompts to those annotations, then either send them for editing right away or add them to a queue to process together later.
Click the **plus** icon to add more assets to your prompt:
* Import media from your device
* Add assets from your library
* Generate new media — image, video, sound, voiceover, or music
Choose how much effort Chat should put into the edit based on how complex the task is:
* **Low** — Simple, straightforward edits
* **Medium** — Regular edits
* **High** — Complex tasks (for example, advanced motion design)
* **Ultra** — The most demanding tasks
Click the send button. Captions starts planning the edits. You can keep working on other parts of the project while Chat processes your request.
## Design and animate custom elements
You can also use Chat to design and animate custom elements, including vector graphics. For example, ask Chat to create two vector elements that act like arrows with a wobbling, liquid feel.
## Example prompts
* [Add chill background music](/help/docs/audio/music)
* Remove/replace background with the X picture
* Change the font to comic sans
* [Generate a b-roll about duck soaps and add it to the first shot](/help/docs/visual/media)
* [Edit captions](/help/docs/captions/add-captions)
* Create two vector arrows with a wobbling, liquid feel
Prompts are flexible. You can chain multiple requests in one message, annotate specific timeline moments, or iterate with follow-ups like "make it softer" or "use a different font".
This documentation is built and hosted on [Mintlify](https://mintlify.com), a developer documentation platform.