Skip to main content
Glama
Sign In

FilmLab

by com.thefilmradar
OverviewDiscussionsAdmin

Generate Sound

lab_sfx

Generate or CLEAN UP a SOUND track other than music — sound effects, ambience/atmosphere, dialog, a voiceover, FOLEY scored to an existing clip, or the repair kinds that take an EXISTING recording apart instead of inventing one: denoise (take the noise out), split (separate it into stems), isolation (keep the voice, drop the rest). All of it goes through the same audio pipeline as lab_generate_music. ASYNCHRONOUS: the call returns immediately with a row whose audio_url is null and metadata.status is 'pending'. Nothing is playable yet. Poll lab_list_audio (matching audio_type) until that row's status is 'completed' and audio_url is set; a failed row refunds its credits and carries metadata.error. The backend picks the provider for the chosen art on its own — there is no tool parameter here.

Input Schema

NameRequiredDescriptionDefault
artYesKind of track — sfx (a single sound effect), ambience (a continuous atmosphere/soundscape), dialog (a spoken line), voiceover (narration), foley (sound derived FROM an existing clip and synced to what happens in it — needs `video_id`, ignores `duration_seconds`, and takes its length from the source). The repair kinds work on a recording you already have and need exactly one source (`audio_id`, `video_id` or `audio_url`), never a prompt: denoise (take out hiss, hum and room noise), split (separate the recording into stems — one audio row per stem), isolation (keep only the voice, drop room, noise and music). Their length follows the source, so `duration_seconds` is ignored. Not 'music' — that stays lab_generate_music, it binds a different stem
stemsNoSPLIT ONLY: which parts to separate out. Left out, you get the voice and the rest of the recording, each part as its own audio row. Asking for a subset drops what you leave out — there is no combined everything-but-the-voice file, that would be a mixdown the provider never made.
promptNoWhat the sound should be. For sfx/ambience: name the source and character of the sound. For dialog/voiceover: the line or text to speak. Required for every art EXCEPT foley — foley watches the clip, so leaving it empty means 'take what you see', and a prompt merely steers it
audio_idNoREPAIR KINDS ONLY, and the best of the three ways to name a source: an existing audio row of this project (from lab_list_audio), which has an owner and an address that belongs to us. Chain it — denoise a row, then split the result. The row must already carry an audio_url; a pending one has no file yet.
get_costNo`true` reports the price and your balance WITHOUT generating anything and without spending quota. Not drawn from the lab_list_models catalog (that covers image/video models only) — audio pricing follows the provider FilmLab picks for `art`, and each kind is billed by its own unit: per run, per second, or per character. For the spoken kinds the price follows the length of your text, which the cost endpoint cannot take: the answer then names the provider and your balance and reports `cost_generations` as null rather than inventing a number — the real amount comes back with the call.
video_idNoThe clip to work from. For FOLEY it is required to RUN one; for the repair kinds it is one of the three ways to name a source, and the backend extracts the audio track from it. From lab_list_assets or the row an animate call returned; the clip must belong to this project. The backend measures its length and rejects sources it will not process, before anything is charged. A price question (`get_cost: true`) does NOT need it — the price follows a length, not a particular clip.
audio_urlNoREPAIR KINDS ONLY, and the last resort: a bare address of a sound file. It carries no owner, so nothing can be checked against the project — prefer `audio_id` or `video_id`. A provider address is copied into our own storage first, and if that fails the call is refused rather than stored with a deadline.
scene_refNoOptional scene label, stored on the row. Useful to tell several tracks of one project apart
project_idYesProject id (from lab_list_projects) — positive integer as string
duration_secondsNoDesired length in seconds. The backend clamps this to what the chosen provider actually supports for this `art` — dialog/voiceover ignore it entirely, the length follows the text, and foley ignores it too, its length follows the clip. The repair kinds ignore it as well: you clean up a file, not a stretch of it. One exception, and only for asking: in a `get_cost` query on foley or a repair kind, naming it prices THAT length instead of the ceiling. There is no lab_list_models entry for audio: the provider isn't a choice you make here, it's picked automatically per `art`.
nur_eigene_datenNoOnly for a model that sends inputs to China or a US recipient without a documented basis, when the step carries images, clips, voices or faces. Means: this material shows or contains no real person other than the human using this account. Set it ONLY after the human has said exactly that for THIS step; never assume it, never reuse it for the next step. Not needed when the human has declared this material in the FilmLab app: the backend reads that declaration, also for material derived from it; only the app can store one. An LK reference, a licensed likeness or another person's agreement does not replace it. Without it the backend refuses with drittland_dritte_gesperrt before any charge; then use a non-gated model or text only. Matters only for the repair kinds: a source recording can carry voices. Text-to-sound never needs it.

Schema Changelog

Changes observed during successful MCP inspections.

  1. Changed1 schema field changed
    • changedInput schema / properties / nur_eigene_daten / description
      Previous value: -"Only for a model that sends inputs to China or a US recipient without a documented basis, when the step carries images, clips, voices or faces. Means: this material shows or contains no real person other than the human using this account. Set it ONLY after the human has said exactly that for THIS step; never assume it, never reuse it for the next step. An LK reference, a licensed likeness or another person's agreement does not replace it. Without it the backend refuses with drittland_dritte_gesperrt before any charge; then use a non-gated model or text only. Matters only for the repair kinds: a source recording can carry voices. Text-to-sound never needs it."New value: +"Only for a model that sends inputs to China or a US recipient without a documented basis, when the step carries images, clips, voices or faces. Means: this material shows or contains no real person other than the human using this account. Set it ONLY after the human has said exactly that for THIS step; never assume it, never reuse it for the next step. Not needed when the human has declared this material in the FilmLab app: the backend reads that declaration, also for material derived from it; only the app can store one. An LK reference, a licensed likeness or another person's agreement does not replace it. Without it the backend refuses with drittland_dritte_gesperrt before any charge; then use a non-gated model or text only. Matters only for the repair kinds: a source recording can carry voices. Text-to-sound never needs it."
  2. Changed1 schema field changed
    • addedInput schema / properties / nur_eigene_daten
      Added value: +{
+  "description": "Only for a model that sends inputs to China or a US recipient without a documented basis, when the step carries images, clips, voices or faces. Means: this material shows or contains no real person other than the human using this account. Set it ONLY after the human has said exactly that for THIS step; never assume it, never reuse it for the next step. An LK reference, a licensed likeness or another person's agreement does not replace it. Without it the backend refuses with drittland_dritte_gesperrt before any charge; then use a non-gated model or text only. Matters only for the repair kinds: a source recording can carry voices. Text-to-sound never needs it.",
+  "type": "boolean"
+}
  3. Changed1 schema field changed
    • addedInput schema / additionalProperties
      Added value: +false
  4. First observed

TDQS

A4.4/5.0
Behavior4/5

Does the description disclose side effects, auth requirements, rate limits, or destructive behavior?

Annotations only declare readOnlyHint=false and destructiveHint=false, so the description carries the burden and does it well: it discloses the asynchronous contract (row returned with audio_url null, status 'pending'), the credit refund on failure with metadata.error, and that no `tool` parameter exists because the backend picks the provider per `art`. The one residual gap is that it doesn't quantify expected latency or poll cadence, though it does name the polling tool.

Agents need to know what a tool does to the world before calling it. Descriptions should go beyond structured annotations to explain consequences.

Conciseness4/5

Is the description appropriately sized, front-loaded, and free of redundancy?

Front-loaded with purpose, then async lifecycle, then provider note — a sensible order. Dense and every clause is load-bearing, though the heavy capitalization and triple enumeration of the repair kinds make it longer than strictly necessary.

Shorter descriptions cost fewer tokens and are easier for agents to parse. Every sentence should earn its place.

Completeness5/5

Given the tool's complexity, does the description cover enough for an agent to succeed on first attempt?

For an 11-parameter, no-output-schema, credit-spending generation tool with minimal annotations, the description supplies what's required: the async return shape, the polling loop, the failure/refund semantics, and the provider-selection rule. An agent could invoke and follow up correctly from this text alone.

Complex tools with many parameters or behaviors need more documentation. Simple tools need less. This dimension scales expectations accordingly.

Parameters3/5

Does the description clarify parameter syntax, constraints, interactions, or defaults beyond what the schema provides?

Schema description coverage is 100% and the per-parameter docs are already exhaustive, so the description's parameter contribution is limited to reinforcing the art-vs-music boundary and stating there is no `tool` parameter. Baseline 3 is appropriate since the schema, not the description, does the semantic heavy lifting.

Input schemas describe structure but not intent. Descriptions should explain non-obvious parameter relationships and valid value ranges.

Purpose5/5

Does the description clearly state what the tool does and how it differs from similar tools?

Specific verb+resource ('Generate or CLEAN UP a SOUND track other than music') with the full enumeration of kinds (sfx, ambience, dialog, voiceover, foley, denoise, split, isolation). It explicitly excludes music and routes it to lab_generate_music, so the agent can distinguish it from the nearest sibling without opening a schema.

Agents choose between tools based on descriptions. A clear purpose with a specific verb and resource helps agents select the right tool.

Usage Guidelines5/5

Does the description explain when to use this tool, when not to, or what alternatives exist?

Explicit when-to-use (generate vs repair kinds), when-not ('Not music — that stays lab_generate_music'), and what to do after the call (poll lab_list_audio matching audio_type until completed). It also names get_cost as the price-check route without generating.

Agents often have multiple tools that could apply. Explicit usage guidance like "use X instead of Y when Z" prevents misuse.

All FilmLab toolsInspect
Try in Browser
One endpoint for every MCP toolRayrun runs every MCP server behind one URL and holds the credentials. Each call is decided per client, tool, and argument, then recorded.ray.runAd

Glama MCP Gateway

Add one secure layer between your agents and this server.

Resources