Scumble

Product Introduction

1. Definition: Scumble is a free, open-source, AI-native desktop image editor and inpainting tool built with Electron, TypeScript, and Rust. It functions as a specialized layer-based compositing application that integrates directly with generative AI models via API or local computation.
2. Core Value Proposition: Scumble exists to bridge the gap between generative AI models and professional, non-destructive image editing workflows. Its core proposition is delivering AI-generated content as editable, color-matched layers within a traditional digital art and photo editing environment, eliminating the need for manual compositing after generation.

Main Features

1. AI-Generated Layers with Real-Time Color Matching: Every AI operation (inpainting, outpainting, generation) returns its result as a new, separate layer. A proprietary color-matching algorithm automatically blends the new content with the surrounding image. The "Match" slider allows for non-destructive adjustment of this blending at any time, and layers support masks, blend modes, and filters.
2. Multi-Modal Spatial Control with Box Prompts: Beyond simple text prompts, Scumble implements spatial control through drawn bounding boxes. Users can draw multiple boxes on a canvas and assign descriptive prompts to each (e.g., "new," "keep," "remove," "text"). Supported models like FLUX 3 Image and Ideogram 4 interpret these boxes to precisely control object placement and composition within the generated image.
3. Flexible AI Backend Integration (ComfyUI & Direct APIs): Scumble is backend-agnostic. It can connect to a local or remote ComfyUI instance (requiring the ComfyUI-InpaintCanvas node pack) for custom workflow execution. Alternatively, it can operate using only API keys from providers like Black Forest Labs (FLUX), OpenAI (GPT-4 Image), Google, fal.ai, and Replicate, requiring no local AI infrastructure. It also features a dedicated "Comfy Cloud" mode.
4. Model Context Protocol (MCP) Server for AI Agent Control: The entire application is exposed as an MCP server with over 105 tools. This allows AI agents (e.g., in Claude Desktop, Cursor) to directly control Scumble—opening files, making selections, generating content, and exporting—enabling automated, scripted image editing workflows.
5. Professional-Grade Image Engine with Large File Support: Built on a custom tile-based rendering engine with compiled Rust kernels, Scumble handles very high-resolution images (up to 65,535 pixels per side) efficiently. It supports PSD export with layers and masks, OpenRaster format, and includes GPU-accelerated filter layers (46 film looks, grain, curves, LUTs).

Problems Solved

1. Pain Point: The disconnect between AI image generation and practical, iterative editing. Standard AI tools output flat images, forcing users into other software for compositing, color correction, and non-destructive adjustments.
2. Target Audience: Digital artists, concept artists, photo editors, and marketing content creators who need to rapidly prototype, edit, and composite images using AI. It also targets technical users and AI enthusiasts who already use ComfyUI and want a more painterly, layer-based interface for its capabilities.
3. Use Cases: Rapid concept art creation from textual descriptions with spatial control; professional photo editing and object removal/inpainting with perfect layer-based workflows; batch or agent-driven image generation and modification; creating marketing materials by combining AI-generated elements with existing assets.

Unique Advantages

1. Strengths & Limitations (Pros & Cons):

   • Pros: Truly non-destructive, layer-based AI editing; unparalleled flexibility in backend choice (local GPU/API/Cloud); advanced spatial control via box prompts; professional file format support (PSD); unique MCP integration for automation; free and open-source (GPL-3.0).
   • Cons: Currently Windows-focused (Linux builds are experimental, macOS is planned); version 0.1.x indicates early-stage software with potential bugs; requires user configuration for AI backends (ComfyUI setup or API key management); some advertised API provider integrations are untested.

2. Key Alternatives & Differentiation:

   • Adobe Photoshop (Generative Fill): Photoshop is a closed-source, subscription-based industry standard. Its AI features are baked into the workflow but are vendor-locked to Adobe's Firefly models. Scumble differentiates by being free, open-source, and model-agnostic, allowing use of cutting-edge models like FLUX 3 and supporting local execution.
   • ComfyUI Standalone: ComfyUI is a powerful, node-based visual programming interface for stable diffusion workflows. It is highly flexible but has a steep learning curve and lacks built-in, traditional image editing tools. Scumble provides a more intuitive, layer-oriented GUI on top of ComfyUI workflows while also offering a direct API path.
   • Luna.ai / Krita AI Plugins: These are often single-model or limited integrations within host applications. Scumble is a standalone, dedicated editor that consolidates multiple AI backends (ComfyUI + dozens of APIs) into a unified, professional editing environment with full layer support.

Frequently Asked Questions (FAQ)

1. Is Scumble free and what does open-source mean? Yes, Scumble is completely free to download and use. Being open-source (licensed under GPL-3.0) means its source code is publicly available on GitHub, allowing for community inspection, modification, and distribution.
2. Do I need ComfyUI to run Scumble? No, ComfyUI is optional. You can run Scumble using only API keys from supported providers like Black Forest Labs for FLUX or OpenAI for GPT-4 Image. You need ComfyUI only if you want to run custom workflows or models on your own hardware.
3. What image formats does Scumble support for import and export? Scumble supports standard raster formats (PNG, JPG) and, critically, professional formats like PSD (export with layers/masks) and OpenRaster. It can also re-import its own PSD/OpenRaster exports with layers intact.
4. What are the system requirements for Scumble? The primary requirement is Windows 10 or 11 (64-bit). For local AI processing via ComfyUI, you need a compatible NVIDIA GPU with sufficient VRAM. For API-only use, a standard modern computer with an internet connection is sufficient. Experimental Linux builds are available.
5. How does Scumble's "box prompt" feature work? The Boxes tool (X) lets you draw regions on the canvas. You can label each box as containing something new, something to keep, text, or a reference image. When using compatible models (e.g., FLUX 3 Image), these boxes and their labels are sent with the prompt, giving the AI precise spatial instructions for generation.