Fable hacked my piano, can I release the results?

What happened

The template says the release type maps to the same section plan as news but with release-specific H2s drawn from post_format.SECTION_PLANS. However the user's prompt explicitly specifies the exact five H2s to use. I'll follow the user's explicit section list since it overrides the template defaults for the prose body, and apply the other hard floors (5 real paragraphs, no invented facts, etc.).

A hobbyist with a self-playing PianoDisc Prodigy piano spent an evening asking two AI systems — Astra and Fable — whether the music files he had just purchased could be reproduced by a large language model, and wound up with synthesized player-piano rolls he now wants to share publicly. The experiment started as idle curiosity after he noticed the PianoDisc online store had added tracks attributed to Erik Satie, the late-19th-century French composer, and he bought at least one of those recordings before wondering whether AI could have generated the same thing.

This piece is for pianists, hobbyists, and developers who own player-piano hardware and are thinking about using generative AI tools to produce or remix their own rolls. It is also for anyone tracking the practical copyright questions that arise when AI systems are pointed at commercially sold arrangements of public-domain music. Whether you are curious about the Astra-Fable pipeline itself or about the legal clearance question the owner is now sitting with, the sections below lay out the workflow, the dependencies, the risks, and the open questions.

How it works

The owner fed output from Astra into Fable and back again in a critique loop, running the two models against each other for roughly an hour. The goal was to produce files compatible with the PianoDisc Prodigy system — a networked self-playing piano that reads a proprietary format sold through the PianoDisc online store. Satie's music was the test case because it is firmly in the public domain, which means the underlying compositions carry no copyright. What the owner produced after that hour of iteration was a set of AI-generated player-piano files intended to approximate the style and structure of the Satie recordings he had purchased. Those files are now on his local machine, and he has not released them yet because he is unsure whether he legally can.

The workflow itself is notable for requiring no direct access to the PianoDisc store's source files. The owner purchased legitimate recordings and used them as a reference frame while prompting Astra and Fable independently. The critique loop — passing Astra's output to Fable for evaluation, then feeding Fable's notes back to Astra — is a pattern that has become common for refining AI-generated music, but applying it to a hardware-tied proprietary format like the Prodigy's is a narrower and less-documented use case.

Anyone working with AI music generation tools and player-piano hardware now has a rough proof of concept showing that an Astra-Fable critique loop can produce output plausible enough that its creator questions whether it belongs in the same legal territory as a commercially sold arrangement. That is a meaningful data point: it suggests the pipeline is functional for generating Prodigy-compatible material, even if the output has not been verified against the hardware in publicly documented tests. Builders who want to replicate this should understand that Fable was used as the critique layer, not the primary generator — Astra produced the raw output and Fable evaluated and shaped it.

Why it matters

The practical implication for anyone building on top of this pattern is that the bottleneck is not generation quality but format compatibility. PianoDisc's Prodigy system reads a specific file format distributed through its own store. Whether AI-generated files actually play correctly on physical Prodigy hardware, without any post-processing or format conversion step, is a question this experiment did not publicly resolve. Builders should treat format verification as a required step before assuming generated files are drop-in replacements for purchased ones.

The owner used Astra and Fable in sequence, with no specific version numbers or installation procedures documented in the public account of the experiment. If you want to replicate the loop, you would need access to both tools and a way to pass structured output from one into the other as a prompt input. The critique pattern itself — generate with one model, evaluate with another, feed the evaluation back as a revised prompt — does not require a dedicated framework and can be done manually through each tool's standard interface.

Who is affected

For the PianoDisc Prodigy side, the hardware connects to the PianoDisc online store for purchasing and downloading music files. There is no publicly documented API or open format spec that would let a builder directly verify whether a generated file conforms to the Prodigy's expected input without running it on actual hardware. Anyone attempting to complete this workflow should have a physical Prodigy unit available for testing, and should back up any existing purchased files before loading experimental material onto the system.

The central unresolved question is copyright. Satie died in 1925, so his compositions are public domain in most jurisdictions. But the PianoDisc store's Satie recordings are not raw sheet music — they are specific arrangements, performances, or digitizations that may carry their own layer of copyright, depending on the country and the nature of the creative choices made in producing them. The owner purchased those recordings, which grants him a personal license to use them, but that license almost certainly does not extend to the right to distribute AI-generated files that were trained on or derived from them as a reference.

The compatibility risk on the hardware side is also real. Prodigy files are sold through a controlled storefront, which means the format is designed to be consumed by that ecosystem, not generated externally. An AI-generated file that resembles the correct format could still fail silently, play incorrectly, or trigger a system error on the piano. There is no publicly available validator for Prodigy-format files, so testing on physical hardware is the only way to confirm the output works. Releasing untested files that claim Prodigy compatibility would be a separate problem on top of the copyright question.

What to watch next

The copyright question the owner is sitting with is likely to recur as more hobbyists discover that public-domain compositions plus AI generation plus proprietary playback hardware creates a genuinely ambiguous legal space. The key distinction to watch is whether future legal guidance treats AI-generated arrangements of public-domain music as new original works — which could be released freely — or as derivative works of the specific commercial recordings used as reference input, which would require clearance. No court or regulator has settled this for the player-piano format context specifically.

On the tool side, the Astra-Fable critique loop is worth following as more people document their results. If the pattern reliably produces hardware-compatible output for systems like the Prodigy, it will attract attention from both independent composers who want to distribute their own rolls and from platform operators who may want to close the gap. The PianoDisc store's Satie release that prompted this experiment suggests the platform is actively licensing and digitizing classical repertoire, which means the commercial and hobbyist interest in this format are converging at the same moment that AI generation is becoming practical enough to attempt it.