← Gameplay preview
Ahmed Kart / Dream Circuit · Sound workshop 02

More character.
Less gravel.

Revised Stable Audio takes, plus a complete set of Astra-designed effects: bright rewards, springy jumps, rounded turbo energy and distinct damage cues.

New proposals for your ears. The first Stable bank missed the mark. This revision changes the generation settings and gives you the full source takes to compare. Astra’s proposals are composed synthesis, not output from an Astra audio model. The Mac game still uses its existing sounds.

Hear the sounds together

Eleven-second listening mixes: espresso → boost and coins → jump → pizza shield → impact and heart loss → recovery. Constructed examples, not gameplay recordings. Start at a comfortable volume.

Reference

Current game

The existing V2 sounds, with V3’s event gains.

Astra / composed synthesis

Astra-designed

26 purpose-built effects: tuned resonators, pitch gestures and shaped air textures.

Local model / revised settings

Stable revised

Short effects selected from seven-second sources, using default guidance and simpler prompts.

Local model + composed synthesis

MOSS + Astra

The ten existing MOSS proposals, with Astra for cues outside that set.

Explore each cue

Compare Original, Astra and Stable; MOSS is available for ten actions. Repetition exposes tiring tails. Boost sustain loops until you press Stop. “Stable full take” lets you hear what the model actually generated before editing.

Loading the sound bank…

What changed in Stable?

A controlled coin test changed from a sustained low-frequency texture at 1.48 seconds to discrete transients at seven seconds. Lower guidance alone did not fix the short render. The audio codec also reconstructed a known tone accurately. That points to our generation setup as a contributor; it does not yet establish that the new effects sound right.

Listen to the controlled comparisons

Same coin prompt and seed, two lengths and two guidance settings. Each full take has the same peak level for playback. The clap and water controls use seven seconds and guidance 1. Judge the character and timing, not just loudness.

Read the diagnosis, settings and limits

Choose a cue to listen.