> For the complete documentation index, see [llms.txt](https://docs.flora.ai/llms.txt). Markdown versions of documentation pages are available by appending `.md` to page URLs; this page is available as [Markdown](https://docs.flora.ai/mcp/recipes/prompting.md).

# Brief your agent for better output

Write briefs that get better FLORA results from Claude, ChatGPT, Cursor or any agent: what to include, how to pick a model, using references, approving cost, and checking the output.

The agent turns your words into [FLORA MCP](https://flora.ai/mcp) calls, so the brief you give it decides the prompt, the model, the size and the spend. This page shows what a good brief contains and the follow-ups that fix weak results. It applies to every recipe, from [generate](https://docs.flora.ai/mcp/recipes/generate.md) to [batch](https://docs.flora.ai/mcp/recipes/batch.md).

**Works in:** Claude · Claude Code · ChatGPT · Codex · Cursor

**You need:** a connected client and a workspace; the cost is whatever the brief asks for, and the agent quotes it before spending

**Use when:** results come back generic, the wrong size, or more expensive than you expected.

**Don't use when:** you need exact tool shapes; the other recipes and the [tools reference](https://docs.flora.ai/mcp/tools.md) have those.

## Ask your agent

A complete brief, ready to adapt:

> Use FLORA to make 4 hero images for our fall coffee launch. Subject: a matte black ceramic mug with latte art. Use: Instagram feed ads. Format: 4:5, 2K. References: the 2 mug photos on my "Mug launch" canvas. Avoid: text, logos, extra props. Show me the cost first and wait for my OK. After they finish, look at each image and regenerate only the ones that miss the brief.

<a href="https://claude.ai/new?q=Use%20FLORA%20to%20make%204%20hero%20images%20for%20our%20fall%20coffee%20launch.%20Subject%3A%20a%20matte%20black%20ceramic%20mug%20with%20latte%20art.%20Use%3A%20Instagram%20feed%20ads.%20Format%3A%204%3A5%2C%202K.%20References%3A%20the%202%20mug%20photos%20on%20my%20%22Mug%20launch%22%20canvas.%20Avoid%3A%20text%2C%20logos%2C%20extra%20props.%20Show%20me%20the%20cost%20first%20and%20wait%20for%20my%20OK.%20After%20they%20finish%2C%20look%20at%20each%20image%20and%20regenerate%20only%20the%20ones%20that%20miss%20the%20brief." class="button primary">Open in Claude</a> <a href="https://chatgpt.com/?q=Use%20FLORA%20to%20make%204%20hero%20images%20for%20our%20fall%20coffee%20launch.%20Subject%3A%20a%20matte%20black%20ceramic%20mug%20with%20latte%20art.%20Use%3A%20Instagram%20feed%20ads.%20Format%3A%204%3A5%2C%202K.%20References%3A%20the%202%20mug%20photos%20on%20my%20%22Mug%20launch%22%20canvas.%20Avoid%3A%20text%2C%20logos%2C%20extra%20props.%20Show%20me%20the%20cost%20first%20and%20wait%20for%20my%20OK.%20After%20they%20finish%2C%20look%20at%20each%20image%20and%20regenerate%20only%20the%20ones%20that%20miss%20the%20brief." class="button secondary">Open in ChatGPT</a>

## Write the brief

Cover five things. Agents map them straight onto `prompt`, `params` and `reference_node_ids`.

| Part                | Say                                                                                       | Becomes                                                   |
| ------------------- | ----------------------------------------------------------------------------------------- | --------------------------------------------------------- |
| **Subject**         | What's in the frame, with materials and details: "matte black ceramic mug with latte art" | The `prompt`                                              |
| **Use**             | Where it runs: "Instagram feed ad", "PDP hero", "deck cover"                              | Composition and style choices                             |
| **Format and size** | Aspect ratio, resolution, video length: "4:5, 2K", "9:16, 5 seconds"                      | `params` such as `aspect_ratio`, `resolution`, `duration` |
| **References**      | Which images to follow, and for what: "#2 for layout, the product photo for the mug"      | `reference_node_ids` or an `image_url`                    |
| **Avoid**           | What must not appear: "no text, no logos, no extra props"                                 | Negative direction in the `prompt`                        |

Also say **how many**. "Make some variations" leaves the count to the agent; "4 images" doesn't.

Vague: *"Make me some coffee images."* Better: *"4 images of a matte black ceramic mug with latte art on a sunlit oak table, for Instagram feed ads, 4:5 at 2K, no text or props."*

## Choose a model

You don't have to name one. If you don't, the agent picks one with `flora_list_models` or lets FLORA use its default for the job, and the quote names the model before anything runs. Say what matters instead:

* *"Cheapest model that's good for product shots."*
* *"Best quality for a hero image; price isn't the issue."*
* *"Fast video, under a minute to render."*
* *"Show me three image models with their price and speed, then I'll pick."*

Or name one: *"Use Nano Banana 2"* or *"Use Kling 3.0 Pro for the video."* The [generate recipe](https://docs.flora.ai/mcp/recipes/generate.md#choose-a-model) lists good starting models with their IDs and prices, and the model pages under [Models](https://docs.flora.ai/models/image-models.md) describe each one.

## Point at references

References carry more than words. Most strong image work starts from one.

* **Earlier results:** *"Use #2 and #4 as references."* The agent maps the numbers to the `node_id`s it got back. See [Iterate on favorites](https://docs.flora.ai/mcp/recipes/iterate.md).
* **Canvas images:** *"Use the two mug photos on my 'Mug launch' canvas."* The agent finds them with `flora_get_canvas`.
* **Your files:** attach them in Claude or ChatGPT, or give a link or a path. The agent uploads anything not already hosted. See [Upload your own images, video and files](https://docs.flora.ai/mcp/recipes/upload-files.md).
* **Say what each one is for:** *"#1 for layout, #5 for color."*

To animate an image, ask for video "from" it; the agent uses the [image-to-video](https://docs.flora.ai/mcp/recipes/image-to-video.md) path.

## Ask for the cost, set a limit

End every brief with *"Show me the cost first and wait for my OK."* The agent prices the exact request with `quote: true` and shows you US dollars before anything runs.

{% hint style="warning" %}
**Cost rule.** Price first: call `flora_create_generations` with `quote: true` and show the user the cost in US dollars. Run the same request without `quote` only after they approve. On `insufficient_credits` or any `BILLING_*` error, stop and pass the message on. Don't retry. Check balance and spend with `flora_get_usage`. A failed run isn't charged; a retry is a new run and is billed again.
{% endhint %}

For bigger jobs, add a limit or a checkpoint:

* *"Stay under $5. If it would cost more, stop and ask."*
* *"Make one first and show me before you do the other 49."*
* *"What's my balance?"* (the agent calls `flora_get_usage`)

## Check the output and fix only the misses

Ask the agent to check each result against the brief and regenerate only the ones that miss, with the prompt changed to fix the named problem:

* *"Look at each image. For any that has text, a wrong color, or a cropped mug, say which and regenerate just those."*
* *"#3 has the handle on the wrong side. Regenerate only #3 with the handle on the right."*

Some agents can't see images directly. They can check outputs through FLORA with an image-to-text model, such as `i2t-claude-opus-5-5` with the output URL in `image_urls`, for about a cent per image. The **flora-refine-loop** skill runs this generate, check and fix loop for you and stops when the goal is met or progress stalls. See [FLORA MCP skills and plugins](https://docs.flora.ai/mcp/skills.md).

Re-running the same prompt is a re-roll, not a fix. Name what's wrong and change the prompt.

## Name the Technique when you have one

If your team saved the workflow as a Technique, say its name: *"Use the Clean Background v2 Technique on these 5 photos."* That skips the search and gives you the team's exact treatment. Not sure what exists? *"What FLORA Techniques do we have for product photos?"* See [Run your team's Technique](https://docs.flora.ai/mcp/recipes/run-a-technique.md).

## Helpful one-liners

| Ask                                  | What the agent calls                                       |
| ------------------------------------ | ---------------------------------------------------------- |
| "Which workspace are you billing?"   | `flora_list_workspaces`                                    |
| "What's in my 'Mug launch' project?" | `flora_list_projects`, then `flora_get_canvas`             |
| "What did that cost?"                | `flora_list_generations` (each run's `charged_cost`)       |
| "What did I spend this month?"       | `flora_get_usage`                                          |
| "Did the video from earlier finish?" | `flora_list_generations` (history, or the saved `run_ids`) |
| "What can FLORA do from here?"       | `flora_list_tools` and `flora_discover_skills`             |

## Check it worked

* The quote matched the brief: the right model, count and size before anything ran.
* Results match the five parts of the brief, and only the misses were regenerated.

## If it fails

| Symptom                          | Cause                                                       | Fix                                                  |
| -------------------------------- | ----------------------------------------------------------- | ---------------------------------------------------- |
| Results look generic             | The brief named a subject but no use, references or details | Add the use, references and an avoid list.           |
| Wrong size or ratio              | Format wasn't in the brief                                  | Say the ratio and resolution, for example "4:5, 2K". |
| Higher cost than expected        | The brief didn't limit count, resolution or video length    | Ask for the quote first and set a limit.             |
| The same problem every round     | The prompt didn't change between rounds                     | Name the defect and ask for that one fix.            |
| Agent picked the wrong workspace | Several workspaces and no instruction                       | Say which workspace to bill, by name.                |


---

# Agent Instructions
This documentation is published with GitBook. GitBook is the documentation platform designed so that both humans and AI agents can read, navigate, and reason over technical content effectively. Learn more at gitbook.com.

## Querying This Documentation
If you need additional information that is not directly available in this page, you can query the documentation by asking a question.

Perform an HTTP GET request on the following URL with the `ask` and `goal` query parameters:

```
GET https://docs.flora.ai/mcp/recipes/prompting.md?ask=<question>&goal=<user_goal>
```

`ask` is the immediate question: it should be specific, self-contained, and written in natural language.
`goal` is what the user is ultimately trying to achieve, the reason they need the answer. Sharing it helps GitBook give you a better, more relevant answer. A goal is most helpful when it describes the outcome the user wants rather than restating the question. For example, with `ask=how do I create an API token`, a goal like `automate deployments from our CI pipeline` lets GitBook tailor the answer to that use case.

The response will contain a direct answer to the question and relevant excerpts and sources from the documentation.

Use this mechanism when the answer is not explicitly present in the current page, you need clarification or additional context, or you want to retrieve related documentation sections.
