# Edit Images with Intent-Based RunComfy Routing

Different image edits require different models and input formats. This skill selects a RunComfy route and prepares model-specific prompts and CLI requests.

## Install

```bash
npx skillstore add prime-skills/image-edit
```

## Metadata

- Status: approved
- Slug: prime-skills-image-edit
- Skillstore revision: r1
- Version status: missing
- Tree hash: 43fb8f2a8e7e7e0dbde1ff070de103d01b361dcad54fa3d367f2c4f85b223428
- Author: prime-skills
- GitHub username: prime-skills
- License: MIT
- Repository: https://github.com/prime-skills/runcomfy-agent-skills/tree/fca19ae084c2ae4667f5c8affbb99da5103d5832/image-edit
- Ref: 26908a3ee86cf306c97e871bf2c30610e8f763ec
- Supported tools: Claude, Codex, Claude Code
- Audit status: complete
- Agent install advisory: allowed
- Manual install advisory: allowed
- Artifact signature: available
- Audit attestation: unavailable
- Human verification: not\_verified
- Risk factors: external\_commands, network, filesystem
- Quality score: 77
- Quality tier: bronze
- Public page: https://skillstore.pages.dev/skills/prime-skills-image-edit
- Manifest: https://skillstore.pages.dev/api/skills/prime-skills-image-edit/manifest

## Capabilities

- Route requests to Nano Banana Edit, GPT Image 2 Edit, Flux Kontext Pro, or Z-Image Turbo Inpaint based on intent.
- Prepare Nano Banana requests with 1 to 20 input image URLs and consistent aspect ratio and resolution settings.
- Prepare multilingual text replacement and numbered reference-image composition requests for GPT Image 2 Edit.
- Prepare single-reference local edits with preservation instructions for Flux Kontext Pro.
- Configure grayscale-mask inpainting with strength and control settings for region removal or replacement.
- Provide RunComfy CLI examples, output-directory options, and documented exit-code explanations.

## Use Cases

- Standardize Product Image Backgrounds: Prepare consistent background edits for a product series while specifying preserved subjects, aspect ratio, and resolution.
- Localize Text in Campaign Images: Request exact headline replacements in another language while preserving the photograph, brand mark, and layout.
- Retouch Selected Photo Regions: Prepare masked removal or replacement requests with explicit preservation boundaries and suitable strength settings.

## Prompt Templates

### Beginner: Replace a Background

```
Edit the image at [image URL]. Replace only the background with light gray studio paper. Preserve the subject, pose, and clothing.
```

### Intermediate: Edit a Product Series

```
Use Nano Banana Edit for [image URLs]. Apply a consistent white background and preserve each product. Set aspect ratio 1:1 and resolution 1K.
```

### Advanced: Localize an Image Headline

```
Use GPT Image 2 Edit on [poster URL]. Replace only the headline with "[exact text]" using [script]. Preserve the photograph, logo, and layout.
```

### Expert: Configure Masked Replacement

```
Use Z-Image Turbo Inpaint with [image URL] and [mask URL]. Replace the white-masked region with [replacement]. Preserve unmasked details. Set strength [value].
```

## Limitations

- Requires the external RunComfy CLI, an authenticated account, and network access; this skill does not implement image editing locally.
- Input images and masks require server-accessible URLs; remote processing is unsuitable for assets without permission for external sharing.
- Each call uses one model; reference counts, sizes, and mask requirements vary by route.
- The supplied file contains instructions, not tested results or CLI source; model availability, security protections, and output quality are not independently verified.

## Best Practices

- State what must remain unchanged before describing the edit, and split complex changes into smaller passes.
- Verify permission for remote processing, protect credentials, and use approved image URLs before running the CLI.
- Match the route to reference counts and mask availability; inspect generated images for text errors and unintended changes.

## Anti Patterns

- Sending confidential images or credentials through unapproved public URLs or placing tokens inside prompts.
- Using a single-reference route for multiple images or requesting inpainting without a grayscale mask.
- Combining many unrelated changes in one request and assuming subject identity or text accuracy is guaranteed.

## Security Audit

- Audited at: 2026-11-01T14:45:42.028\+00:00
- Summary: All 84 static detections are false positives involving Markdown formatting, placeholder URLs, documentation links, or documented owner-only credential storage. No evidence found of malicious instructions, unauthorized data collection, or prompt injection in SKILL.md. Remote image processing is explicit; the external CLI implementation and its stated security protections were not verified.

## Stats

- Views: 0
- Downloads: 0
- Favorites: 0
- Popularity score: 0