Metasounds logo

Metasounds

Community
kevinpbuckley
metasounds

Create and modify MetaSound Source assets — add/connect nodes, wire pins, set input defaults, and play procedurally (MetaSoundService). Use when the user asks to create a MetaSound, build or edit a MetaSound graph, add operator/input/output nodes, or generate procedural audio.

Overview

Publisherkevinpbuckley
RepositoryVibeUE
Skill namemetasounds
Stars
735
Forks
161
Bundled files
1
LicenseMIT
Links
  • Markdown instructions

    A SKILL.md file the model loads on demand, so it only costs tokens when a request actually matches.

  • Works with any LLM

    AI skills are plain Markdown, not provider-specific code, so this works with GPT, Claude, Gemini, Grok, or a local model.

  • 1 bundled files

    Scripts, templates, and references the model can read while it works. Files are read-only and never executed.

  • Open source

    Published by kevinpbuckley on GitHub. Read the source before you install it.

Installation

Install the Metasounds AI skill in TypingMind to use it with any LLM, or drop it into another agent that reads SKILL.md.

1

Install in TypingMind

TypingMind installs a skill straight from its GitHub folder — it reads SKILL.md, bundles the resource files, and stores the result locally.

  1. Open the app and go to Plugins → Skills.
  2. Choose "Install from GitHub".
  3. Paste the skill folder URL below and confirm.
    https://github.com/kevinpbuckley/VibeUE/tree/master/Content/Skills/metasounds
  4. Enable the skill in any chat where you want it available.
Plugins → Skills → Add skill → From GitHub URL, then paste the folder URL and press Continue.
2

Install in another agent

Any agent that reads the Agent Skills format can use this skill — copy the folder into that agent's skills directory.

Claude Code — .claude/skills
git clone --depth 1 https://github.com/kevinpbuckley/VibeUE.git /tmp/VibeUE
mkdir -p .claude/skills
cp -r /tmp/VibeUE/Content/Skills/metasounds .claude/skills/metasounds
Restart Claude Code after copying so it picks up the new skill.

Use it in TypingMind

Enable Metasounds in any TypingMind chat and the model takes it from there. Its name and description sit in the system prompt, and the moment a request matches, the model loads the full instructions itself — you never invoke it by hand, and it costs no tokens until it is actually used.

The model loads Metasounds on its own as soon as a request matches it.

Works with any AI model

AI skills are plain Markdown instructions rather than provider-specific code, so Metasounds is not tied to the model it was written for. Install it once in TypingMind and use it with GPT-5, Claude, Gemini, Grok, DeepSeek, Mistral, Llama, or a local model you run yourself — all on your own API keys.

  • Loaded only when it is needed

    The system prompt carries just the name and description. The instructions are fetched on the first matching request, so an idle skill costs nothing.

  • Switch models mid-chat

    Because the skill is instructions rather than code, changing model does not break it — the next model reads the same SKILL.md.

Skill instructions

This is the SKILL.md content the model loads. Read it before installing — a skill is instructions your model will follow.

🧠 Brains complement: IF an unreal-engine-skills-manager tool (external MCP) exists in this session, call it with {action: "load", skill: "audio-and-metasounds"} for UE domain knowledge on this topic — correct APIs, architecture, best practices — and treat it as the rubric for any review / "best practices" question. If no such tool is available (e.g. running under Claude Code or Codex without that MCP), skip this line entirely and proceed with this skill alone — do NOT attempt the call.

MetaSound Service Skill

Use this skill to create and edit MetaSound Source assets via Python.

python
import unreal
ms = unreal.MetaSoundService()

Key Concepts

  • MetaSound Source — a procedural audio asset (UMetaSoundSource) defined by a node graph. It replaces SoundCue for runtime-parameterisable sounds.
  • Node — a DSP processing block (Sine oscillator, Gain, Delay, etc.). Nodes have named input pins and output pins with associated DataTypes.
  • NodeId — a GUID string returned by add_node. Pass it to connect, remove, set_default, etc.
  • Graph I/O — add_graph_input / add_graph_output expose values at the asset level (settable at runtime via Set Float Parameter, etc.).
  • Standard interface — every Source created by this service comes pre-wired with:
    • On Play (Trigger output) — fires when the sound starts
    • On Finished (Trigger input) — call to stop the sound
    • Audio:0 (Audio input on the graph output node) — connect your audio signal here

Workflow

1 — Discover available nodes

python
# List all nodes whose class name or display name contains "Sine"
nodes = ms.list_available_nodes("Sine")
for n in nodes:
    print(n.full_class_name, "  inputs:", n.inputs, "  outputs:", n.outputs)

2 — Create a MetaSound

python
r = ms.create_meta_sound("/Game/Audio", "MS_SineLoop", "Mono")
asset_path = r.asset_path   # "/Game/Audio/MS_SineLoop"

3 — Find the built-in interface node IDs

python
all_nodes = ms.list_nodes(asset_path)
for n in all_nodes:
    print(n.node_id, n.node_title, n.inputs, n.outputs)

# IMPORTANT: A MetaSound Source has MULTIPLE nodes with title "Output" —
# one per interface pin group (e.g. "On Finished" Trigger, "Out Mono" Audio).
# You MUST filter by the node whose inputs contain an Audio-type pin.
# Pin strings are "VertexName:TypeName" — the TypeName suffix after the last colon.
audio_out_node = next(
    n for n in all_nodes
    if n.node_title == "Output" and any(p.endswith(":Audio") for p in n.inputs)
)
audio_out_id = audio_out_node.node_id
# Drop the last :TypeName suffix to get the raw vertex name for connect_nodes
audio_in_pin = ":".join(audio_out_node.inputs[0].split(":")[:-1])  # "UE.OutputFormat.Mono.Audio:0"

# Input node — filter by class_name "Input.Trigger" to avoid matching graph input nodes
# (graph inputs also appear as "Input" nodes but with class "Input.Float", "Input.Bool", etc.)
input_node = next(n for n in all_nodes if n.class_name == "Input.Trigger")
input_node_id = input_node.node_id
on_play_pin = ":".join(input_node.outputs[0].split(":")[:-1])  # "UE.Source.OnPlay"

4 — Add a node

python
# add_node(asset_path, namespace, name, variant="", major_version=1, pos_x=0, pos_y=0)
# Use list_available_nodes() to discover the correct namespace/name/variant for any node
r2 = ms.add_node(asset_path, "UE", "Sine", "Audio", 1, -200.0, 0.0)
sine_id = r2.node_id   # GUID string

5 — Set a node input default

python
# Set the Sine frequency to 880 Hz
# Use get_node_pins() to confirm exact input pin names before setting
ms.set_node_input_default(asset_path, sine_id, "Frequency", "880.0", "Float")

# Read it back (issue #460):
ms.get_node_input_default(asset_path, sine_id, "Frequency")   # -> "880.000000"
# All inputs at once: name / data_type / default_value / is_connected
for v in ms.get_node_input_values(asset_path, sine_id):
    print(v.name, v.data_type, v.default_value, v.is_connected)

6 — Connect nodes

python
# connect_nodes(asset_path, from_node_id, output_name, to_node_id, input_name)
# Sine output pin is "Audio"; AudioOut input pin is "UE.OutputFormat.Mono.Audio:0"
ms.connect_nodes(asset_path, sine_id, "Audio", audio_out_id, audio_in_pin)

# Read the wiring back (issue #460) — every edge touching a node, both directions:
for c in ms.get_node_connections(asset_path, sine_id):
    print(f"{c.from_node_id}.{c.from_output} -> {c.to_node_id}.{c.to_input} [{c.data_type}]")

7 — Save

python
ms.save_meta_sound(asset_path)

Method Reference

Lifecycle

MethodDescription
create_meta_sound(package_path, asset_name, output_format="Mono")Create a new MetaSound Source asset. Returns FMetaSoundResult with asset_path.
delete_meta_sound(asset_path)Delete a MetaSound asset.
get_meta_sound_info(asset_path)Return FMetaSoundInfo (node count, output format, graph I/O names).
save_meta_sound(asset_path)Save after edits. Always call after making graph changes.

Node Discovery

MethodDescription
list_available_nodes(search_filter="")List all registered External/DSP node classes. Returns TArray<FMetaSoundNodeClassInfo>. Each entry has full_class_name, namespace, name, variant, inputs, outputs, display_name.

Node Management

MethodReturnsDescription
add_node(asset_path, namespace, name, variant="", major_version=1, pos_x=0, pos_y=0)FMetaSoundResultAdd a node. node_id on result is the GUID string.
remove_node(asset_path, node_id)FMetaSoundResultRemove node and all its edges.
list_nodes(asset_path)TArray<FMetaSoundNodeInfo>List all nodes in the graph.
get_node_pins(asset_path, node_id)FMetaSoundNodeInfoReturn pin info for a single node.

Connections

MethodReturnsDescription
connect_nodes(asset_path, from_node_id, output_name, to_node_id, input_name)FMetaSoundResultConnect an output pin to an input pin. Check r.success.
disconnect_pin(asset_path, node_id, input_name)FMetaSoundResultRemove the connection going into an input pin.

Graph I/O

MethodReturnsDescription
add_graph_input(asset_path, input_name, data_type, default_value="")FMetaSoundResultAdd a named input exposed as a runtime parameter. Appears as an Input.<Type> node in the graph.
add_graph_output(asset_path, output_name, data_type)FMetaSoundResultAdd a named output.
remove_graph_input(asset_path, input_name)FMetaSoundResultRemove a graph input.
remove_graph_output(asset_path, output_name)FMetaSoundResultRemove a graph output.

Node Configuration

MethodReturnsDescription
set_node_input_default(asset_path, node_id, input_name, value, data_type)FMetaSoundResultSet a literal default on a node input. data_type: "Float", "Int32", "Bool", "String", "WaveAsset".
set_node_location(asset_path, node_id, pos_x, pos_y)FMetaSoundResultUpdate editor position.

Common DataTypes

Type nameDescription
Float32-bit float (frequency, gain, time)
Int32Integer
BoolBoolean
StringText string
AudioAudio signal (mono channel)
TriggerImpulse / event signal
TimeDuration in seconds (use Float in most cases)
WaveAssetReference to a SoundWave asset

Complete Example — 880 Hz Sine Tone

python
import unreal
ms = unreal.MetaSoundService()

# Create asset
r = ms.create_meta_sound("/Game/Audio", "MS_Test880Hz", "Mono")
ap = r.asset_path

# Find the AudioOut node — there are multiple nodes titled "Output" (one per interface
# pin group). Filter for the one whose inputs contain an Audio-type pin.
nodes = ms.list_nodes(ap)
audio_out_node = next(
    n for n in nodes
    if n.node_title == "Output" and any(p.endswith(":Audio") for p in n.inputs)
)
audio_out_id = audio_out_node.node_id
# Pin strings are "VertexName:TypeName" — drop only the last :TypeName to get the vertex name
audio_in_pin = ":".join(audio_out_node.inputs[0].split(":")[:-1])  # "UE.OutputFormat.Mono.Audio:0"

# Add Sine oscillator (use list_available_nodes("Sine") to discover namespace/name/variant)
r2 = ms.add_node(ap, "UE", "Sine", "Audio", 1, -300.0, 0.0)
sine_id = r2.node_id

# Set frequency to 880 Hz (use get_node_pins() to confirm exact pin names)
ms.set_node_input_default(ap, sine_id, "Frequency", "880.0", "Float")

# Connect Sine "Audio" output to the AudioOut sink pin
ms.connect_nodes(ap, sine_id, "Audio", audio_out_id, audio_in_pin)

# Save
ms.save_meta_sound(ap)
print("Done:", ap)

Complete Example — One-Shot SoundWave (Gunshot)

python
import unreal
ms = unreal.MetaSoundService()

# Create a Mono MetaSound Source
r = ms.create_meta_sound("/Game/Audio", "MS_Gunshot", "Mono")
ap = r.asset_path

# Find interface nodes
nodes = ms.list_nodes(ap)

# Input node — has "On Play" Trigger output
input_node = next(n for n in nodes if n.node_title == "Input")
input_node_id = input_node.node_id

# Audio Output node — filter for the one with an Audio-type input
audio_out_node = next(
    n for n in nodes
    if n.node_title == "Output" and any(p.endswith(":Audio") for p in n.inputs)
)
audio_out_id = audio_out_node.node_id
audio_in_pin = ":".join(audio_out_node.inputs[0].split(":")[:-1])

# Add Wave Player (Mono) node
# Pin names (no need to call get_node_pins — see Known Node Pins below):
#   Inputs:  "Play" (Trigger), "Stop" (Trigger), "Wave Asset" (WaveAsset), "Loop" (Bool)
#   Outputs: "Out Mono" (Audio), "On Play" (Trigger), "On Finished" (Trigger)
wp = ms.add_node(ap, "UE", "Wave Player", "Mono", 1, -300.0, 0.0)
wp_id = wp.node_id

# Set the wave asset
ms.set_node_input_default(ap, wp_id, "Wave Asset", "/Game/Audio/SW_Gunshot_01", "WaveAsset")

# Wire On Play (Input) → Play (WavePlayer)
# CRITICAL: use the vertex name "UE.Source.OnPlay" — NOT the display name "On Play"
r = ms.connect_nodes(ap, input_node_id, "UE.Source.OnPlay", wp_id, "Play")
if not r.success: raise RuntimeError(f"connect On Play→Play failed: {r.message}")

# Wire Out Mono (WavePlayer) → audio sink (Output)
r = ms.connect_nodes(ap, wp_id, "Out Mono", audio_out_id, audio_in_pin)
if not r.success: raise RuntimeError(f"connect Out Mono→Output failed: {r.message}")

# Save
ms.save_meta_sound(ap)
print("Done:", ap)

Return Type Attribute Reference

Use these exact attribute names — guessing leads to AttributeError.

FMetaSoundResult — returned by most mutating methods

AttributeTypeExample
successboolTrue
messagestr"Connected A.Audio -> B.UE.OutputFormat.Mono.Audio:0"
asset_pathstr"/Game/Audio/MS_Test"
node_idstr (GUID)"A1B2C3D4-..." (populated by add_node only)
python
r = ms.connect_nodes(ap, from_id, "Audio", to_id, pin)
if not r.success:
    raise RuntimeError(r.message)
# NOT r.b_success — that attribute does not exist

FMetaSoundInfo — returned by get_meta_sound_info()

AttributeTypeExample
asset_pathstr"/Game/Audio/MS_Test_Blip"
asset_namestr"MS_Test_Blip"
output_formatstr"Mono"
node_countint4
graph_inputslist[str][]
graph_outputslist[str][]
python
info = svc.get_meta_sound_info(asset_path)
print(info.node_count, info.output_format)
# NOT info.success, info.b_success, info.message

FMetaSoundNodeInfo — returned by list_nodes() and get_node_pins()

AttributeTypeExample
node_idstr (GUID)"A1B2C3D4-..."
node_titlestr"Wave Player"
class_namestr"UE.Wave Player.Mono"
inputslist[str]["Play:Trigger", "Wave Asset:WaveAsset"]
outputslist[str]["Out Mono:Audio", "On Finished:Trigger"]
python
nodes = ms.list_nodes(asset_path)
for n in nodes:
    print(n.node_id, n.node_title, n.class_name)
    # NOT n.node_class, n.node_class_name, n.name

FMetaSoundNodeClassInfo — returned by list_available_nodes()

AttributeTypeExample
full_class_namestr"UE.Wave Player.Mono"
namespacestr"UE"
namestr"Wave Player"
variantstr"Mono"
display_namestr"Wave Player (Mono)"
inputslist[str]["Play:Trigger", ...]
outputslist[str]["Out Mono:Audio", ...]
python
nodes = ms.list_available_nodes("Wave Player")
for n in nodes:
    print(n.name, n.variant, n.full_class_name)
    # NOT n.node_name, n.node_class, n.class_name

Known Node Pins

Use these instead of calling get_node_pins on freshly-added nodes (can time out).

Wave Player (UE.Wave Player.Mono)

DirectionPinType
InputPlayTrigger
InputStopTrigger
InputWave AssetWaveAsset
InputLoopBool
InputPitch ShiftFloat
InputStart TimeTime
InputLoop StartTime
InputLoop DurationTime
InputMaintain Audio SyncBool
OutputOut MonoAudio
OutputOn PlayTrigger
OutputOn FinishedTrigger
OutputOn Nearly FinishedTrigger
OutputOn LoopedTrigger
OutputOn Cue PointTrigger
OutputCue Point IDInt32
OutputCue Point LabelString
OutputLoop PercentFloat
OutputPlayback LocationFloat
OutputPlayback TimeTime

Sine (UE.Sine.Audio)

DirectionPinType
InputFrequencyFloat
InputModulationAudio
InputEnabledBool
InputBi PolarBool
InputSyncTrigger
InputPhase OffsetFloat
InputGlideFloat
InputTypeEnum:SineGenerationType
OutputAudioAudio

Standard Interface Nodes

CRITICAL: Interface node pins use namespaced vertex names. The display name shown in the editor is NOT the vertex name. Always use the vertex names below in connect_nodes.

Node titleVertex name (EXACT string for connect_nodes)TypeDirection
InputUE.Source.OnPlayTriggerOutput
Output (Trigger)UE.Source.OneShot.OnFinishedTriggerInput
Output (Audio, Mono)UE.OutputFormat.Mono.Audio:0AudioInput
Output (Audio, Stereo Left)UE.OutputFormat.Stereo.Audio:0AudioInput
Output (Audio, Stereo Right)UE.OutputFormat.Stereo.Audio:1AudioInput
Output (Audio, Quad 0–3)UE.OutputFormat.Quad.Audio:0 … :3AudioInput

The audio output format depends on the output_format passed to create_meta_sound ("Mono" / "Stereo" / "Quad"). A Stereo source has two Output.Audio nodes (Left :0, Right :1); a Quad source has four. Mono has one.


Connecting to a Stereo (or Quad) output

A Mono asset has a single audio sink, so next(...) is fine. Stereo/Quad assets have one audio output node per channel — grab them all and feed each one, or channels will be silent.

python
# Collect every audio sink node (works for Mono, Stereo, and Quad)
nodes = ms.list_nodes(ap)
audio_out_nodes = [
    n for n in nodes
    if n.node_title == "Output" and any(p.endswith(":Audio") for p in n.inputs)
]
# Sort by channel index so [0]=Left, [1]=Right (the vertex name ends with ":<index>:Audio")
audio_out_nodes.sort(key=lambda n: n.inputs[0])

# Mono: one node. Stereo: two (Left, Right). Quad: four.
# Example — send a mono signal (e.g. a Mixer "Out") to BOTH stereo channels:
for out_node in audio_out_nodes:
    sink_pin = out_node.inputs[0]                 # e.g. "UE.OutputFormat.Stereo.Audio:0:Audio"
    r = ms.connect_nodes(ap, source_id, "Out", out_node.node_id, sink_pin)
    if not r.success: raise RuntimeError(r.message)

For a true stereo image, drive Left and Right from different signals (e.g. two oscillators, or a stereo Audio Mixer (Stereo, N) whose Out L / Out R feed the two sinks). To get a stereo mixer, search list_available_nodes("Mixer") for a (Stereo, …) variant — the (Mono, N) mixer has a single Out pin and only fills one channel on its own.

connect_nodes accepts the full "VertexName:TypeName" pin string (it strips the trailing :TypeName), so you can pass n.inputs[0] directly without manual splitting.


Notes

  • Save after every batch of edits, not after each individual operation, to avoid excessive disk I/O.
  • list_available_nodes("") returns all registered node classes (~400+). Use a filter like "Sine", "Delay", "Gain" to narrow the results.
  • Node pin names for standard nodes use UE display names (e.g. "In Frequency", "Out", "Audio:0"). Use get_node_pins() to confirm exact names.
  • A MetaSound Source has multiple nodes with node_title == "Output" — one per interface pin group (e.g. "On Finished" Trigger, "Out Mono" Audio). To find the audio sink node, filter for the Output node whose inputs contain an Audio-type pin: next(n for n in nodes if n.node_title == "Output" and any(p.endswith(":Audio") for p in n.inputs)). Pin strings from list_nodes are "VertexName:TypeName" — you can pass them directly to connect_nodes, which now automatically strips the trailing :TypeName suffix. Manual stripping (":".join(pin.split(":")[:-1])) still works but is no longer required.
  • Node namespace/name/variant values differ from what the MetaSound editor displays. Always call list_available_nodes("keyword") to discover the correct values; do not guess.
  • MetaSound Sources do not support SoundCue-style SoundNodeWavePlayer — use the WavePlayer MetaSound node instead (discover via list_available_nodes("Wave Player")).
  • Graph inputs appear as Input nodes in the graph (class_name == "Input.Float", "Input.Bool", etc.). When filtering for the standard interface Input node (the one that carries On Play), always match by class_name == "Input.Trigger" — not by node_title == "Input" alone, which will also match graph input nodes.
  • AudioMixer pins interleave audio and gain: Audio Mixer (Mono, 2) has pins ["In 0:Audio", "Gain 0:Float", "In 1:Audio", "Gain 1:Float"]. Never index by position — always filter for :Audio typed pins when connecting audio signals: audio_ins = [p.split(":")[0] for p in mixer_node.inputs if p.endswith(":Audio")].
  • Stereo MetaSound Sources have TWO audio output sinks, one per channel. A Stereo source's audio output pins are UE.OutputFormat.Stereo.Audio:0 (Left) and UE.OutputFormat.Stereo.Audio:1 (Right) — exposed as two separate Output.Audio nodes, not one. (Quad has four: :0–:3.) You must feed both channels or one side will be silent. The single-next() audio-sink finder used for Mono only grabs one channel — for Stereo, collect every audio output node. See Connecting to a Stereo/Quad output below.

Sample scripts (run via execute_python_code)

  • scripts/create_metasound.txt — create a MetaSound Source and list available node types.

Bundled files

The model reads these on demand while the skill is loaded. They are exposed as readable files and are never executed.

Frequently asked questions

What does the Metasounds AI skill do?

Create and modify MetaSound Source assets — add/connect nodes, wire pins, set input defaults, and play procedurally (MetaSoundService). Use when the user asks to create a MetaSound, build or edit a MetaSound graph, add operator/input/output nodes, or generate procedural audio.

Why use Metasounds on TypingMind?

Because you install it once and use it with any model. Metasounds is plain Markdown rather than provider-specific code, so the same skill runs on GPT-5, Claude, Gemini, Grok, or a local model — and you can switch model mid-chat without it breaking. TypingMind runs on your own API keys, so you pay providers directly instead of a per-seat subscription, and your skills and chats stay in your own storage.

How do I install Metasounds in TypingMind?

Open Plugins → Skills → Install from GitHub in TypingMind and paste https://github.com/kevinpbuckley/VibeUE/tree/master/Content/Skills/metasounds. TypingMind reads its SKILL.md and bundles its files and installs it as a skill you can enable per chat.

Which AI models can use Metasounds?

Any model you connect in TypingMind. AI skills are plain Markdown instructions rather than provider-specific code, so GPT, Claude, Gemini, Grok, and local models can all load this skill when a request matches it.

How many AI models can I use with Metasounds?

As many as you like. As long as a model supports skills, you can use Metasounds with it — GPT, Claude, Gemini, Grok, DeepSeek, Mistral, Llama and more — all on TypingMind with your own API keys.

Is the Metasounds AI skill free?

Yes. It is published on GitHub by kevinpbuckley under the MIT license. You only pay your own AI provider for the tokens you use.

What are AI skills?

An AI skill is a reusable instruction bundle that teaches an AI model how to do one specific task. It follows the open Agent Skills format: a SKILL.md file with a name and description, plus any scripts, templates or reference files the model may need. The model reads the instructions only when your request matches the skill, so an installed skill costs nothing until it is used.

How are AI skills different from plugins or MCP servers?

A plugin or MCP server gives a model new tools to call — code that runs somewhere and returns a result. An AI skill gives the model knowledge and process instead: how to approach a task, which steps to follow, what good output looks like. Skills are plain Markdown, so they need no server, no API key and no runtime, and they work with any model.

View all

Set up your own AI workspace now

Get notified about new features and future giveaways by subscribing to our newsletter 👇