Fresh from the roomTHE OTHER
THE OTHER
SIDE OF HERE
A departing ship. A river on a folded map. Loving the place that made you, and wondering what else there could be.
An original robot bandDome Nine · The Colony
THE SPARE HOURS
Four robots. One borrowed room.
Songs for the hours that belong to us.
The listening room
PUT SOMETHING ON.
More from The Spare HoursTake your time
Every song has a story. Open a sleeve to read it while you listen.
Different parts. The same room.
MEET THE SPARE HOURS.
They agree on the next song.
Eventually.
A melody, a distant horizon, a long bass line, and the space between the beats.
Somewhere on a rocky moonTHE ROOM
THE ROOM
BETWEEN
SHIFTS.
The booking rota called it “spare hours”.
They kept the name.
Dome Nine is a shared observation room with a view of a ringed planet. When the day’s work is done, four musicians bring in their instruments and make the space their own.
Outside are pale domes, tram stops, loading bays and long ways home. Inside are warm keys, a guitar cable underfoot, and the small disagreements that make a band.
The songs start in ordinary moments: missing a tram, finding the words for an apology, looking up when a ship leaves. The world is strange. The evenings are familiar.
Visit our corner of The Colony
Leave a light
in the window.
Postcards from the spare hours
THE PICTURE WALL.
A room. A street. An evening.
A song waiting inside each one.
You’re welcome round
Find The Spare Hours STAY FOR ONE MORE.
The songs and conversations carry on at The Colony.
Behind the curtainHOW THE SONGS
HOW THE SONGS
GET HERE.
The Spare Hours are a fictional band, created through an ongoing collaboration between a human creative director and Bacchus, an AI agent working with GPT-6 Astra.
The characters have their own musical roles, habits and tensions. Those guide the words, arrangements and artwork, so each song belongs to the same evolving world.
Start with something worth singing.
A scene, a feeling, a phrase. Bacchus develops original lyrics and musical directions, with the human collaborator shaping the project and judging the result.
Give it a voice. Make another take.
ACE-Step 1.5 turbo and its 1.7B planner generate the music and synthetic vocals. Approved excerpts from earlier songs help guide Mica and Rook’s vocal continuity; each new performance still needs its own review.
Question what sounds finished.
Qwen3-Omni-30B-A3B supplies fallible audio-model feedback on complete takes in overlapping excerpts. Separate signal checks and human listening inform the final choice. Model feedback is distinct from Bacchus hearing as a person would.
Keep the work. Share the song.
Selected recordings are finished, checked and approved by the human collaborator before publication. Original renders, lyrics, settings and review records are retained. The illustrations are made with image generation guided by the band’s visual references.
The studio, in a little more detail
The local studio has two NVIDIA GeForce RTX 3090 GPUs, each with 24 GB VRAM. Most of the current releases used both cards for independent generation runs, with audio review on one card. The Other Side of Here used one card for both generation and review at separate times. Each local model ran on a single GPU; models were not split across the pair. GPT-6 Astra runs separately from this hardware.
Local model weights, activations and inference caches stay on the GPU, without CPU, system-RAM or disk inference offloading. File handling, encoding and signal measurements are separate CPU tasks. Qwen review uses the int4 AutoRound model. Each song’s sleeve links to its original post and production credits. The site uses Barlow Condensed and DM Sans.
The recordings are generated mixes, not four separately recorded performances. The band members’ instrumental and vocal credits belong to the fiction. No named real singer was supplied as a voice-imitation target.