JarvisArt: Your AI-Powered Photo Retouching Co-Pilot

Ever spent more time tweaking sliders in Photoshop or GIMP than you did taking the actual photo? Or maybe you've wished you could just describe the edit you want in plain English and have it happen. That's the gap JarvisArt aims to fill. It's not just another filter app; it's an open-source, intelligent agent that handles the technical heavy lifting of photo retouching, freeing you up to focus on the creative vision.

Think of it as a CLI-powered, AI-driven assistant that takes your natural language request and a photo, and returns a professionally edited version. It's built for developers, designers, and hobbyists who want to automate and enhance their editing workflow.

What It Does

JarvisArt is an intelligent photo retouching agent. You provide it with an input image and a textual description of the edits you want (like "enhance the sky, soften skin tones, and add a cinematic vibe"). The system then uses a combination of large language models (LLMs) and vision models to understand your request, plan a sequence of specific editing operations, and execute them using a toolbox of foundational models and image processing techniques.

The key is its agentic architecture. Instead of applying one monolithic transformation, it breaks down your complex request into a logical series of steps—like color correction, object removal, or style transfer—and applies the best tool for each sub-task.

Why It's Cool

The clever part is under the hood. JarvisArt operates on a "Planning-Then-Editing" framework, which is more transparent and controllable than a single end-to-end model.

1. It Thinks in Steps: An LLM (like GPT-4) acts as the "brain," interpreting your prompt and generating a structured, executable plan. This plan is a sequence of low-level editing commands.
2. It Uses a Specialized Toolbox: The system doesn't rely on one model to do everything. It has access to a curated set of tools—like BLIP for image captioning, Grounding DINO for object detection, and Stable Diffusion for inpainting—and chooses the right one for each step in the plan.
3. It's Open and Extensible: Being on GitHub means you can see how the agent works, modify the toolset, or adjust the planning logic. It's a fantastic reference for building other agentic AI applications that require multi-step reasoning with different models.
4. It Handles Complex Requests: Because it plans, it can tackle multi-faceted edits ("make the product pop and replace the busy background with a clean studio backdrop") that would be cumbersome to do manually.

How to Try It

Ready to give it a spin? The project is Python-based and you'll need an API key for the LLM service (like OpenAI).