> ## Documentation Index
> Fetch the complete documentation index at: https://docs.rainbrand.com/llms.txt
> Use this file to discover all available pages before exploring further.
# Create a campaign image set
> Use an MCP-connected agent and Rainbrand to explore three campaign directions, select one, and resize it for social and web.
This example starts with `rabbit.tech` and ends with three campaign directions, followed by a selected direction in 4:5, 1:1, and 16:9.
An MCP-connected agent ran the workflow. Rainbrand held rabbit's brand and active Brand Skill, supplied Brand Library references, and created the images. The agent defined the task, compared the results, selected a direction, and asked Rainbrand for each size.
## Explore three directions
The agent asked Rainbrand for three distinct 4:5 interpretations of the same campaign idea. Each used rabbit's Brand Skill and r1 product references from the Brand Library.
The agent selected direction 01 because it had the clearest type hierarchy, the most accurate product geometry, and enough open space to work across all three sizes.
## One direction, three sizes
In this example, Rainbrand resized the selected poster for each aspect ratio, adjusting the composition rather than cropping one canvas. To repeat the workflow with the current generation tools, choose a model that accepts image references and supports the sizes you need.
## Run it with your agent
This feature is in beta. During beta, endpoint schemas and behavior may change
or be removed.
Setup for Codex, Claude, ChatGPT, Cursor, or any MCP-compatible agent.
Once connected, give your agent this prompt:
```text theme={"dark"}
Use Rainbrand through MCP to create a campaign image set for [brand website].
Reuse a matching ready brand or create one in Rainbrand and wait until it is ready.
Inspect the complete active Brand Skill before making creative decisions. Search
the Brand Library for product images that fit the task and use only relevant
references.
Choose an image model and read its input schema. Include relevant brand guidance
and references in its supported input fields. Propose one source-grounded
campaign idea. Generate three distinct 4:5 directions: one
product-led, one lifestyle-led, and one travel-led. Compare their brand fit,
product fidelity, copy, and composition. Select the strongest result. If the
model supports image references and the required sizes, use that result as a
reference to generate 1:1 and 16:9 versions, one output per call.
Check every completed image for product fidelity and text accuracy. Do not
publish anything externally.
```
## How it worked
Rainbrand created rabbit's brand from `rabbit.tech`. Once it was ready, the
connected agent read its Brand Skill.
The agent asked Rainbrand for r1 product images from the Brand Library, then
chose the references that showed the device clearly. It set one campaign
idea around `agents on the go.` for all three directions.
Rainbrand generated a product poster, a street scene, and a train-window
travel scene. Keeping all three at 4:5 made the creative direction easier to
compare.
The agent compared every result, selected the product-led poster, and asked
Rainbrand to resize it to 1:1 and 16:9. It then checked the finished set for
product fidelity and text accuracy.
See the [API quickstart](/api) or [MCP generation tools](/mcp/tools#model-discovery-and-generation) to generate images with a selected model. See [Credits and limits](/api/usage-limits) for rate limits and credits.
This independent example uses rabbit; it is not an official rabbit campaign.
This documentation is built and hosted on [Mintlify](https://mintlify.com), a developer documentation platform.