Describe the drama.
The pipeline builds the rest.
ScriptCast turns a single YAML script into a fully mixed, mastered, chaptered
audio drama — using local AI inference and ffmpeg. No cloud. No per-minute billing.
No third parties between your script and the final master.
Zero spam. We'll only notify you when the preview build is available.
One YAML file describes the full production: characters, dialogue, ambience, music, spatial positioning, mastering. The pipeline produces the audio drama.
From script to master, deterministically.
Every stage is declarative, reproducible, and runs entirely on your machine. No API keys, no job queues in someone else's cloud, no per-render costs.
Parse
ScriptCast reads the YAML script and validates every parameter — pan ranges, reverb mix, ducking ratios, chapter progression, character registration.
Render
The orchestrator dispatches each line and cue to local ComfyUI via HTTP and WebSocket, injecting text, instructions, seeds, and output paths into the workflow graph.
Normalize
Every stem is measured (EBU R128) and normalized to its category reference level. Measurements are cached in a per-project YAML file, so subsequent runs skip already-normalized files.
Mix
A deterministic ffmpeg filter graph assembles dialogue, ambience, music, and SFX with spatial positioning, environment reverb, sidechain ducking, and an EBU R128 loudness pass on the master bus.
Package
The final master is delivered as FLAC, MP3 with ID3v2 tags, and M4B with chapters and cover art — ready to upload to any distribution platform.
Studio-grade DSP, engineered in code.
Every production decision is declared in the script. Every acoustic effect is computed by deterministic filters. Nothing relies on post-hoc tweaking in a DAW.
Declarative production script
Characters, dialogue, ambience, music, SFX, spatial positioning, mastering — everything lives in a single versionable YAML file. Diffs and code review work the same as with source code.
Deterministic local orchestration
ComfyUI workflows are mutated in memory and dispatched via HTTP and WebSocket to a local instance. Seed values, text, and output paths are injected per line and per cue — no external API calls anywhere.
Multi-model synthesis
Supports multiple local voice and audio generation backends through a transformer registry. Swap models by changing the workflow file, without touching the pipeline.
Spatial audio by construction
Pan, distance, occlusion, and pace are first-class parameters on every dialogue line and cue. Environment presets — room, hall, cave, telephone — apply parametric reverb filters computed in the graph.
Deterministic sidechain ducking
Instead of ffmpeg's opaque sidechaincompress, ScriptCast builds an explicit control envelope from the dialogue track and multiplies the target. The applied attenuation matches the authored duckAmountDb exactly.
Chaptered, tagged deliverables
Outputs include FLAC masters, MP3 with ID3v2 metadata, and M4B audiobook containers with chapters, cover art, and media_type=2 — ready for direct upload.
Currently forging: The Empty House.
ScriptCast's first full-cast production is an adaptation of a classic supernatural short story. It doubles as the pilot that validates the entire pipeline end-to-end.
The Empty House
A dramatized full-cast adaptation of one of Blackwood's most restrained ghost stories. Multiple voices, ambience beds, music cues, SFX, spatial positioning, and mastering — all described in a single YAML script, all produced by the pipeline.
The production is in progress. A preview will be released alongside the first ScriptCast build.
How ScriptCast is different.
An honest look at the tools and workflows used today to produce full-cast audio fiction, and where a deterministic pipeline fits.
| Approach | Production model | Strength | Friction |
|---|---|---|---|
| Manual DAW editing | Human performs every cut, mix, and cue | Maximum creative control | Hours per episode, not reproducible, single-editor bottleneck |
| Cloud TTS + manual assembly | Generate voice lines via API, mix by hand | Easy access to high-quality voices | Per-minute billing, uploads content to third parties, no pipeline |
| Cloud audio drama SaaS | Web UI, subscription, hosted inference | No local setup required | Zero privacy, ongoing cost, vendor lock-in, limited scripting |
| ScriptCast | Declarative YAML script, local inference, deterministic pipeline | Reproducible, scriptable, fully local, fixed cost | Requires a local GPU and initial pipeline setup |
Built for creators who think in scripts, not sessions.
ScriptCast is designed for producers who want their audio fiction to be described, versioned, and regenerated — not manually sculpted one take at a time.
Audio fiction podcasters
You produce full-cast audio dramas and want a repeatable pipeline instead of assembling each episode by hand.
Indie audio studios
You want a programmable production layer that integrates with local models and respects your existing asset library.
Audiobook producers
You need multi-voice, dramatized renditions of public-domain works at scale, with chapter metadata and cover art embedded.
Engineers who write fiction
You prefer describing a scene in YAML over clicking through a DAW, and you want the whole thing in version control.
A desktop pipeline, priced like one.
ScriptCast is distributed as professional desktop software, not as a per-minute or per-seat subscription service.
- One-time purchase. The version you buy is yours to keep forever.
- Optional annual upgrade plan for access to new major releases — cancel anytime and keep the last version you paid for.
- Same model professional IDEs use. No lockout, no forced renewal, no cloud dependency.
- Volume licensing available for studios and teams, without telemetry or online activation.
Honest about where we are.
ScriptCast is under active development. Here is a transparent view of what exists today and what is planned. No ship dates, no promises.
Declarative YAML script parsing
Full schema validation across dialogue, cues, spatial parameters, mastering, chapters, and character registration.
Local ComfyUI orchestration
Deterministic workflow graph mutation and dispatch over HTTP and WebSocket, with multi-model transformer registry.
Mix, master, and package
Deterministic ffmpeg filter graph, EBU R128 loudness normalization, true-peak limiting, and multi-format packaging with chapters and tags.
First full production
End-to-end validation of the pipeline with a complete full-cast adaptation of "The Empty House".
Preview and distribution builds
Short preview samplers exported automatically alongside the master, for use as discovery assets on any distribution platform.
Extended environment and DSP presets
Additional acoustic environments, reverb algorithms, and mastering targets to cover a wider range of source material.