firefly

---
name: firefly
description: Generate or edit images using Adobe Firefly via Chrome browser automation. Use when the user says "firefly", "adobe firefly", "generate with firefly", "edit image with firefly", "firefly image", "use firefly to", or wants to use Adobe Firefly for image generation, generative fill, text effects, style transfer, or any image processing. Takes a prompt or task description as the argument.
user-invocable: true
allowed-tools:
  - mcp__claude-in-chrome__tabs_context_mcp
  - mcp__claude-in-chrome__tabs_create_mcp
  - mcp__claude-in-chrome__navigate
  - mcp__claude-in-chrome__read_page
  - mcp__claude-in-chrome__find
  - mcp__claude-in-chrome__form_input
  - mcp__claude-in-chrome__computer
  - mcp__claude-in-chrome__get_page_text
  - mcp__claude-in-chrome__javascript_tool
  - mcp__claude-in-chrome__upload_image
  - Bash
  - Read
---

# Firefly — Image Generation & Editing via Adobe Firefly in Chrome

Arguments passed: `$ARGUMENTS`

## Argument Handling

- `$ARGUMENTS` is the task description (e.g., "generate a fantasy landscape", "remove the background from ~/Desktop/photo.png", "add text effects to the word HELLO")
- If no arguments provided, ASK the user what they want to create or edit

## Supported Capabilities

Adobe Firefly offers several tools. Determine which to use from the user's request:

| User wants | Firefly tool | URL |
|---|---|---|
| Generate an image from text | Text to Image | `https://firefly.adobe.com/generate/images` |
| Fill/remove parts of an image | Generative Fill | `https://firefly.adobe.com/generate/inpaint` |
| Apply text effects/styles | Text Effects | `https://firefly.adobe.com/generate/font-styles` |
| Recolor a vector/SVG | Generative Recolor | `https://firefly.adobe.com/generate/recolor` |
| Generate from a reference image | Style Reference | `https://firefly.adobe.com/generate/images` (with reference upload) |

If unclear which tool to use, default to **Text to Image**.

## Workflow

### Step 1: Get Browser Context

Call `mcp__claude-in-chrome__tabs_context_mcp` to see current tabs.

- If Firefly is already open in a tab, reuse it
- Otherwise, create a new tab

### Step 2: Navigate to Firefly

Navigate to the appropriate Firefly URL based on the task (see table above).

- Wait for the page to load
- If the user is not signed in, tell them: "You need to be signed into your Adobe account in Chrome. Please sign in and try again."

### Step 3: Upload Source Image (if applicable)

If the task involves editing an existing image (generative fill, recolor, background removal):

1. Check if the user provided a file path in the arguments
2. If yes, upload the image using the upload mechanism on the page
3. If no file path but the task requires one, ASK the user for the image path

### Step 4: Enter the Prompt and Configure

1. Read the page to find the prompt input
2. Type the prompt/description
3. If the user specified style preferences (photorealistic, illustration, 3D, etc.), look for style options and select them
4. If the user specified aspect ratio, look for ratio controls and adjust
5. Submit / click Generate

### Step 5: Wait for Results

1. Wait for Firefly to process — this can take 10-30 seconds
2. Read the page periodically to check if results have appeared
3. Firefly typically generates multiple variations — note how many appeared

### Step 6: Download the Image

1. If multiple variations were generated, select the best one (or the first if no preference stated)
2. Look for a download button and click it
3. Check `~/Downloads/` for the newly downloaded file
4. Report the downloaded file path to the user:
   ```
   Image downloaded: ~/Downloads/<filename>
   ```

## Rules

- **Don't trigger alerts or dialogs** in the browser
- **If the page structure is unexpected**, describe what you see and ask the user for guidance rather than clicking blindly
- **If Firefly shows an error** (content policy, credit limit, etc.), report it clearly to the user
- **Firefly uses credits** — if a credit warning appears, inform the user before proceeding
- **Keep it simple** — navigate, configure, generate, download. Don't over-automate.