Bloom
The world model
Bloom is the system that owns brand. When you call bloom_generate_image or bloom_edit_image, the brand's visual DNA (palette, typography, tone, aesthetic, logo treatment) is already attached as image references, and a generation system prompt that knows the brand is applied on top, enforcing quality stances like restraint, whitespace, and no tech clichés.
So your prompt's job is narrow: describe the subject, the composition, and the medium. That's where your prompt adds information; everything else is already handled.
Where to look
Read the rule for the task you're on:
| Task |
Rule |
| Writing the prompt text |
rules/prompting.md |
| Writing the headline copy on the image |
rules/copy.md |
| Setting up a run (references, model, batch) |
rules/workflow.md |
| Choosing an aspect ratio |
rules/channels.md |
References
- For tool semantics (parameters, async behavior, reference images, workspaces, credits), the tool descriptions on
bloom_generate_image, bloom_edit_image, bloom_list_images, etc. are the source of truth. Read those rather than restating them here.
- Bloom has no memory between calls. If your prompt references an image (with a pronoun, a definite article, or "this"), that image has to be attached. The MCP server's
instructions field covers the architectural details.
1---2name: bloom3description: Creating, editing, resizing, or adapting brand images, ads, and social/marketing creative with Bloom: Instagram, LinkedIn, Facebook, X, TikTok posts, stories, carousels, thumbnails, banners, billboards, hero images, OG cards, or product shots using a brand's mascot, logo, or palette. Covers generating from a prompt, removing backgrounds, vectorizing logos, changing aspect ratio, and writing in-image headline copy. Trigger whenever the user mentions Bloom or asks to create/edit/resize a brand picture, post, or ad, or before any bloom_generate_image / bloom_edit_image / bloom_resize_image call. If a request produces or changes a brand picture, use this skill first.4license: MIT5---67# Bloom89## The world model1011Bloom is the system that owns brand. When you call `bloom_generate_image` or `bloom_edit_image`, the brand's visual DNA (palette, typography, tone, aesthetic, logo treatment) is already attached as image references, and a generation system prompt that knows the brand is applied on top, enforcing quality stances like restraint, whitespace, and no tech clichés.1213So your prompt's job is narrow: describe the **subject**, the **composition**, and the **medium**. That's where your prompt adds information; everything else is already handled.1415## Where to look1617Read the rule for the task you're on:1819| Task | Rule |20|---|---|21| Writing the prompt text | [`rules/prompting.md`](rules/prompting.md) |22| Writing the headline copy on the image | [`rules/copy.md`](rules/copy.md) |23| Setting up a run (references, model, batch) | [`rules/workflow.md`](rules/workflow.md) |24| Choosing an aspect ratio | [`rules/channels.md`](rules/channels.md) |2526## References2728- For tool semantics (parameters, async behavior, reference images, workspaces, credits), the tool descriptions on `bloom_generate_image`, `bloom_edit_image`, `bloom_list_images`, etc. are the source of truth. Read those rather than restating them here.29- Bloom has no memory between calls. If your prompt references an image (with a pronoun, a definite article, or "this"), that image has to be attached. The MCP server's `instructions` field covers the architectural details.
tomasward1