Pico

Pico is an open source color correction tool that uses Claude as its backend to handle the tedious technical work of standardizing footage while leaving creative decisions in the hands of the colorist. It is built specifically for video professionals who work with log footage and need to convert it to a standard color space before they can start grading. The tool's guiding principle is explicit: the creative act should not be automated away. Instead of trying to replace the colorist, Pico focuses on removing friction from the parts of the job that are purely technical and repeatable.

The core problem it addresses is the repetitive standardization work that precedes any creative grade. If you shoot in S-Log3, for example, you need to bring that footage into Rec.709 before you can meaningfully evaluate and adjust your colors. The log image looks flat and washed out on purpose. It captures the most dynamic range from the sensor, but it is not meant to be viewed as is. Converting it to a viewable color space is not creative work. It is technical, repeatable, and exactly the kind of task that benefits from automation. Every colorist does this step, and it always follows predictable rules based on the camera profile and target color space. Pico handles this so you can spend your time on the decisions that actually matter.

Under the hood Pico operates as what its creator calls a Claude harness. It sends footage parameters to Claude, receives correction instructions, and applies them to your clips. The API call does the thinking about what LUT or transform to apply, and Pico handles the implementation. This architecture means the tool can leverage the reasoning capabilities of a large language model without requiring you to write prompts or interact with an AI chat interface. You feed it footage, it gives you corrected footage. The integration is designed to be invisible, handling the technical details without getting in your way.

The distinction the tool makes is between correction and grading, and this distinction matters. Correction is about getting footage to a neutral, accurate starting point. It is objective work, bringing the image to where it should be based on what the camera captured. Grading is about pushing it somewhere interesting. It is subjective work, choosing a look, a mood, a feeling. Pico does the first part. You do the second. The creator makes this explicit because they believe many AI tools in creative fields get the boundary wrong, trying to automate artistic judgment rather than technical labor.

The philosophy behind Pico shapes every design choice. Many AI powered creative tools try to do more, to automate the entire process and generate finished work. Some even market themselves as replacements for human artists. Pico explicitly rejects that approach. It treats the artist as the decision maker and positions itself as support staff. This is a deliberate stance, not a limitation of the technology. The creator built it this way because they believe creative control should stay with the person, not the algorithm. If you are a colorist who feels protective over your creative process, this tool is designed with that mindset in mind.

Being open source means you can inspect exactly what it does, modify it to fit your workflow, and run it on your own hardware. The repository is available on GitHub under the name agentic-color-grader. If you want to understand how it communicates with Claude, add support for different log formats, or adjust how it handles edge cases in your footage, the code is there for you to read and change. This transparency also means you do not have to trust a black box with your footage. You can see the entire pipeline and verify exactly what it does before running it on client work.

The documentation does not list supported formats beyond S-Log3 to Rec.709, which suggests that may be the primary or only supported conversion at this stage. The tool is clearly early in its development, built by a single creator with a specific vision. That means you should expect a focused feature set rather than broad codec support, and you may need to contribute or wait if your workflow involves formats it does not yet handle. The tradeoff is that what it does, it does with clear intent and without bloat.

Pricing is not mentioned anywhere on the site, which combined with the open source license suggests it is free to use. You will need your own Claude API access to run it, which is a separate cost depending on how much footage you process. The tool itself does not charge you anything. This makes it especially appealing if you already have API credits or want to control your own infrastructure rather than paying a subscription to a hosted service.