A fly-brain simulation designs a buildable guitar

• The series reached a digital guitar with a body and headstock placed by a 166,700-neuron simulation, while scale-length math set the neck, as GlennHasABeard's fourth-round build describes.
• The outline comes from sparse light cues rather than a fully prescribed curve, and GlennHasABeard's 20-seed comparison tracks 20 distinct paths and their median.
• Six descending neurons became six strings, with GlennHasABeard's sound test counting 643 spike-triggered notes in 21 seconds.
• A separate drawing rig traces supplied wireframes, text, and artwork with a simulated foreleg, as hckmstrrahul's drawing-controller demo shows.

Janelia's MaleCNS download page exposes an interactive NeuPrint browser, Python and R packages, and bulk files. Google Research's release note documents the AI-assisted reconstruction of 3D neural shapes from millions of 2D electron-microscope images.

The shared dataset is a connectome of one male fly's central nervous system, spanning brain, optic lobes, and ventral nerve cord, according to Janelia's project announcement. GlennHasABeard layered eyes, a light target, movement dynamics, and a joystick onto that wiring, then dropped a block wherever the simulated fly walked.

His post separates the ingredients cleanly: the wiring comes from MaleCNS, while the behavioral dynamics, eyes, and controls are his own.

The fourth round assigns the guitar body to the fly's walk, then lets it place the headstock; GlennHasABeard's buildable-guitar post says scale-length math connects the points into a neck. The stated next stage is Blender and CAD, followed by a 3D-printed scale model, rather than a finished physical guitar.

An earlier in-progress pass used 19 light anchors, with every curve between them left to the system's movement, according to GlennHasABeard's progress update. The completed comparison used 23 anchors and 20 different noise seeds; its gold line is the median path across all 20 runs.

That turns the guitar silhouette into an ensemble output, not one fortunate trajectory.

For the audio experiment, GlennHasABeard selected six descending neurons from the cleanest run and mapped them to six strings. In GlennHasABeard's sound test, each of 643 notes is one real spike at its firing moment; the timing comes from the simulation while AI built the pitch mapping.

hckmstrrahul's drawing-controller demo presents another boundary around the same broad idea: a neural system drives motion, while the drawing task remains explicitly structured. The project's technical description calls it a fly-connectome-based controller for a simulated foreleg, and an earlier demo describes custom artwork as tracing trained paths.

The open-source Designerfly repository lays out the division of labor:

• An editor supplies the drawing path.
• UI transforms and a sequencer handle drawing, lifting, and travel phases.
• A 2,048-neuron, roughly 192,000-connection motor network reads target points, joint state, error, and contact force.
• MuJoCo simulates the three-joint foreleg and stylus, leaving ink only during contact.

The README describes the project as an embodied neural prototype with measured fly connectivity, learned parameters, and physics feedback.

The guitar work also surfaced a counterintuitive response: the optic lobe facing a light fell from about 7.1 to 5.6 Hz while the other side stayed near 7.1 Hz. GlennHasABeard's light-response explanation attributes the decrease to histamine released by photoreceptors, an inhibitory effect carried by the source wiring.

The fly has also become a quick visual premise for generative-media workflow demos. Fabian Stelzer's Glif workflow breaks one into four steps:

1. Ask the agent for an iPhone-style still.
2. Animate it with H3 Max.
3. Provide a short voice or dialogue reference to clone its vibe with Seed Audio.
4. Send the image and audio back as H3 Max references.

That workflow runs through Glif. Separately, Higgsfield AI's motion-design post labels a ChatGPT-plugin demo as a GPT-6 Astra upgrade, with a static fly image prompted into a 3D-style animation.