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

# AI agent

> The chat-based AI agent is your creative co-pilot — generating images and videos based on your instructions and canvas context.

The AI agent is a chat-based creative partner that lives in the panel on the right side of the editor. You describe what you want in natural language, and the agent generates images and videos — placing results directly on your canvas.

## How it works

The agent understands three kinds of context:

1. **Your message** — What you type in the chat. Be as descriptive or brief as you want.
2. **Canvas context** — The images and videos currently on your canvas. The agent can see and reference these.
3. **Workspace memory** — Your saved preferences, brand guidelines, and creative patterns from past projects.

When you send a message, the agent combines all three to produce the most relevant output.

## What the agent can do

The agent has access to a full suite of creative tools:

| Category             | Capabilities                                                                                                                |
| -------------------- | --------------------------------------------------------------------------------------------------------------------------- |
| **Image generation** | Create images from text descriptions, edit existing images, remove backgrounds, upscale resolution                          |
| **Video generation** | Create videos from text, animate between keyframes (first and last frame), generate reference video with visual consistency |
| **Media analysis**   | Analyze images and videos to extract style, composition, color, and motion details                                          |

Each of these is covered in detail in the [Creating with AI](/creating/generating-images) guides.

## Talking to the agent

You can be conversational or specific:

* *"Create a sunset over a calm ocean"* — The agent picks reasonable defaults
* *"Generate a 16:9 image of the main character standing in the rain, matching the color palette from the reference photo"* — More specific, references canvas images
* *"Animate this image with a slow zoom out"* — References an existing canvas image
* *"What's the style of this uploaded photo?"* — Asks the vision engine to analyze an image

The agent remembers your conversation history within a project, so you can iterate naturally: *"Make it darker"*, *"Add more fog"*, *"Try a wider angle"*.

## Batch and parallel generation

The agent can handle multiple generation tasks at once. When you request several outputs in a single message, the agent runs them in parallel — you don't have to wait for one to finish before the next one starts.

**Quantity requests:**

* *"Generate 5 different poster concepts for this campaign"*
* *"Create 3 color variations of this logo"*

**Style exploration:**

* *"Try this scene in warm tones, cool tones, and high contrast"*
* *"Generate both a realistic and an illustrated version"*

**Character views:**

* *"Create front, side, and back views of this character"* — The agent generates all three in parallel using the same reference

**Batch edits:**

* *"Remove the background from these 3 images"*
* *"Upscale all the character portraits on the canvas"*

Each parallel result appears on the canvas as a separate item, so you can compare them side by side.

<Tip>
  Batch requests are one of the fastest ways to explore creative options. Instead of generating one image at a time, describe all the variations you want and let the agent handle them simultaneously.
</Tip>

## Using skills

[Skills](/workspace/skills) are reusable prompt templates that you invoke by typing `/` in the chat input. Instead of retyping the same style direction or complex instruction every time, save it as a skill and trigger it with a short command like `/cinematic` or `/product-shot`.

When you type `/`, a popup shows your available skills. Select one to insert it as a visual chip in your message. You can combine multiple skills with your own text in a single message.

Skills are workspace-scoped, so your entire team shares the same set of `/slash` commands. See [Skills](/workspace/skills) for how to create and manage them.

## Conversation memory

The agent remembers everything discussed within a project's chat session. This includes:

* Your creative direction and preferences expressed during the conversation
* What you liked and didn't like about previous generations
* Character names, scene descriptions, and terminology you've established
* Corrections and refinements you've made

This conversation context combines with [workspace memory](/core-concepts/workspace-memory) — your persistent preferences that carry across projects. Together, they mean the agent needs less explanation over time.

## How generations appear

Every generation creates a new item on the canvas. Nothing is overwritten or replaced — you always keep your previous results. This means you can:

* Compare multiple variations side by side
* Go back to an earlier generation at any time
* Mix and match results from different attempts

## Credits

Each generation uses credits from your plan. Different operations cost different amounts — image generation is relatively inexpensive, while video generation uses more credits. See the [pricing page](/workspace/pricing) for a full breakdown.

<Note>
  If a generation fails (due to a model error or timeout), credits are automatically refunded to your account.
</Note>