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

# Krea Agent: brief it once, it runs every tool

> Krea Agent plans multi-step creative jobs from one brief, picks the right image, video and audio model for each step, and saves outputs to shared context.

Krea Agent is a chat interface on top of every tool in Krea. You write a brief. The agent plans the job, chooses which image, video and audio models to use, generates the assets, reviews them, edits them, and saves everything in a context library your team shares.

<Note>
  Krea Agent is open on every plan. Anyone can open it and browse the templates. Sign in to send a message, attach files or pick a model. What each plan includes is on the [Plans, access and credits](/docs/user-guide/agent/plans-and-credits) page.
</Note>

<a href="https://www.krea.ai/agent" className="not-prose group my-6 flex items-center justify-between gap-4 rounded-xl border border-gray-200 dark:border-gray-800 px-5 py-4 no-underline transition-colors hover:border-gray-300 dark:hover:border-gray-700">
  <div>
    <div className="font-semibold text-gray-900 dark:text-gray-100">Open Krea Agent</div>
    <div className="mt-0.5 text-sm text-gray-500 dark:text-gray-400">Start a session at krea.ai/agent.</div>
  </div>

  <span aria-hidden="true" className="text-gray-400 transition-transform group-hover:translate-x-0.5 dark:text-gray-500">→</span>
</a>

## What the agent does

A typical request: "Here is our product photo and our website. Make six static ads for Instagram in our brand colours, then a 15-second vertical video for Reels." The agent reads the website, writes a short brand reference, generates the stills, reviews them, generates the video from the best still, and puts every file in a folder you can open.

Two things set it apart from a chat model with an image button:

* **It runs multi-step workflows.** Research, generation, review, editing and export happen in one session. The agent can run several directions in parallel.
* **It picks the model for each step.** Krea hosts more than 150 image, video, audio and enhancement models. The agent chooses the one that fits: an edit that must respect a reference, a text-heavy layout, a fast exploration, a hero video, an upscale.

## What the agent can do

<AccordionGroup>
  <Accordion title="Generate">
    * Images, one at a time or in batches of up to eight, with any image model in Krea's catalog.
    * Video, from text or from a start image, with models such as Seedance 2.5.
    * Audio: text-to-speech and dialogue with Seed Audio and ElevenLabs voices.
    * Upscales and enhancements, including Topaz.
    * Background removal with Krea's dedicated segmentation tool.
  </Accordion>

  <Accordion title="Reference and style">
    * Train a style from a set of your images, then use it in generations.
    * Create and use moodboards with Krea 2 models.
    * Look at an image you upload and describe or critique it.
  </Accordion>

  <Accordion title="Research">
    * Search the web and search for images.
    * Open a URL and read the page as text. This is how the agent turns a website into a brand brief.
  </Accordion>

  <Accordion title="Edit and assemble">
    * Work in a private computer attached to the session: trim clips, change aspect ratios, add captions, build PDFs and convert formats.
    * Build small web tools on demand, and open them in the side panel.
  </Accordion>

  <Accordion title="Organize and repeat">
    * Save files into the [context library](/docs/user-guide/agent/context-library-and-skills) and organize them in folders.
    * Save a workflow as a [skill](/docs/user-guide/agent/context-library-and-skills#skills) and run it again later.
    * Remember your preferences across sessions.
    * Create routines: recurring runs that start a fresh session on a schedule (Enterprise).
  </Accordion>

  <Accordion title="Connect">
    Link external tools through OAuth: Slack, Notion, Linear, Airtable, Webflow, Pinterest, Meta Ads, Google Drive, Google Docs, Google Sheets, Gmail, Dropbox, OneDrive, GitHub, WhatsApp Business, YouTube, Supabase, TikTok Ads, ManyChat, Frame.io and Figma. Each tool has an allow-or-ask permission you control under **Integrations**.
  </Accordion>
</AccordionGroup>

## How a session works

There are two layers of AI in every session.

1. **The orchestrating model.** A large language model reads your brief, plans the steps, calls the tools and reviews the results. **Auto** picks this model for you. Paid plans can pin a specific model instead: Basic and Pro choose from a short list, Max and up from the full list of GPT, Claude, Gemini and other models.
2. **The generation models.** For each step, the orchestrator picks a generation model from Krea's catalog and writes the prompt for it. Current defaults: Nano Banana Pro for edits and reference-following, ChatGPT Image 2 for fine detail and text-heavy layouts, Krea 2 for fast exploration, Seedance 2.5 for hero video, Topaz for upscaling.

An **effort** setting (Instant, Medium, High, Extra High) controls how much planning and review the orchestrator does. Instant is for quick edits. Extra High is for a full campaign where you want the agent to check its own output and retry. Which efforts you can pick depends on your plan.

Each session has a chat, a context panel, a files panel, a generations panel and a media editor. You can annotate a generated image, mark the region to change, and ask for a revision.

## Start a session

<Steps>
  <Step title="Open Krea Agent">
    Go to [krea.ai/agent](https://www.krea.ai/agent), or open **Agent** from the product switcher in the top-left corner of Krea. Sign in when the composer asks you to.
  </Step>

  <Step title="Write the whole brief in one message">
    State the outcome, the format, the audience and the constraints. Attach references: product photos, a logo, a PDF, a link to your website. The agent reads them before it plans.
  </Step>

  <Step title="Pick an effort">
    Leave **Auto** on. Choose **Instant** for a quick edit, **Medium** for most jobs, **High** or **Extra High** when you want the agent to review and retry on its own.
  </Step>

  <Step title="Review and steer">
    The agent shows its plan and its outputs as it works. Reply to steer, annotate an image to ask for a change, or stop the run. Approve when it asks before an expensive step.
  </Step>

  <Step title="Collect the files">
    Outputs land in the **Files** panel and in your context library. Download them, or move them to **Workspace Context** so your team's agents can use them too.
  </Step>
</Steps>

## Three prompts to start

Copy one of these into a new session.

> Read [https://yourbrand.com](https://yourbrand.com) and build a brand book: logo directions, a colour palette with hex codes, type pairings, and three image-style references. Save it as a PDF in a new folder called Brand.

> Here is one ad (attached). Make variations for the US, Spain and Germany: translate the copy, swap the setting to a local street, and export each in 1:1, 9:16 and 16:9.

> Take the attached product photo. Make a looping 4-second GIF of the product slowly rotating on a plain background, plus a transparent PNG of the product for our website.

## Where Krea Agent works

* **Web.** [krea.ai/agent](https://www.krea.ai/agent) is the full product: chat, context library, media editor, skills, integrations and, on Enterprise, routines and custom agents.
* **Slack.** The [Krea Slack agent](https://www.krea.ai/slack) answers in threads and generates images, video, 3D and edits where your team already talks.
* **ChatGPT.** [Krea for ChatGPT](https://www.krea.ai/plugins/chatgpt) brings Krea's image and video models and your context library into a ChatGPT conversation.
* **Other agents.** [Krea MCP](/docs/developers/mcp) connects Krea's generation models to Claude, Codex, Cursor and any agent that supports MCP. The [Krea CLI](https://www.krea.ai/cli) does the same from a terminal.

## What the agent does not do yet

* **3D generation is not inside the web agent.** It is available through Krea MCP and the Slack agent.
* **Nodes workflows are separate.** The agent does not run Krea Nodes graphs. [Krea Nodes](/docs/user-guide/features/nodes) has its own prompt-to-workflow agent.
* **Sessions are private.** There is no public share link for a session. Share the outputs from the context library instead.
* **Some jobs fail and need a retry.** The agent retries most generation failures on its own.