Nodes/Foundation-1/Foundation-1 Generate
ComfyUI Node

Foundation-1 Generate

Generates a tempo-synced musical sample loop using Foundation-1. Connect Foundation1ModelLoader → model input.

By Saganaki22·Created 5 months ago·Updated 5 months ago· 105
Foundation-1 Generate
  • model
  • audio
  • AUDIO
tagsSynth Lead, Warm, Wide, Bright, Clean, Melody
bpm140 BPM
bars8 Bars
keyE minor
steps250
cfg_scale7.0
seed0
sampler_typedpmpp-3m-sde
sigma_min0.300
sigma_max500
unload_after_generatefalse
torch_compilefalse
init_noise_level0.70
Categoryaudio/Foundation-1

Inputs (15)

NameTypeDefaultDescription
modelFOUNDATION1_MODELConnect to Foundation1ModelLoader.
tagsSTRINGSynth Lead, Warm, Wide, Bright, Clean, MelodyInstrument, timbre, FX, and notation tags separated by commas. See the Tag Reference Sheet. Do NOT put BPM, Bars, or Key here — use the dropdowns below.
bpmCOMBO140 BPMTempo of the generated loop.
barsCOMBO8 BarsLoop length in 4/4 bars.
keyCOMBOE minorMusical key for the generated loop.
stepsINT25010–500Diffusion steps. Higher = better quality but slower. 100–250 is the practical range. The training demo used 250.
cfg_scaleFLOAT7.01–15Classifier-free guidance scale. Higher = stronger prompt adherence, less variation. 6–8 recommended. Training demo used 7.
seedINT00–18446744073709550000Generation seed. Use ComfyUI's seed controls to randomise.
sampler_typeCOMBOdpmpp-3m-sdeDiffusion sampler. 'dpmpp-3m-sde' recommended for quality. 'dpmpp-2m-sde' is slightly faster.
sigma_minFLOAT0.3000.001–10Minimum noise level for k-diffusion schedule. Default 0.3.
sigma_maxFLOAT50010–1000Maximum noise level for k-diffusion schedule. Default 500.
unload_after_generateBOOLEANfalseAfter generation, move the model from VRAM to CPU RAM. Frees VRAM while keeping weights in memory for a faster next run (avoids a full disk reload). When enabled, ComfyUI's native 'Free Memory' button will also clear the CPU copy.
torch_compileBOOLEANfalseCompile the model with torch.compile before the first generation. The first run after enabling will be slower (compilation warmup). Every subsequent run in the same session will be faster. Compiled state is cached — toggling off requires a model reload. Requires PyTorch 2.0+. CUDA only. Disable if you see errors.
audiooptAUDIOOptional input audio for variation generation. Connect an AUDIO output from another node (e.g. LoadAudio, or a previous Foundation1Generate output). When connected, the model will create a variation/interpretation of this audio guided by your prompt. Leave disconnected for standard text-to-audio generation.
init_noise_leveloptFLOAT0.700.01–1Variation strength when an input audio is connected. Lower values (0.1–0.3) produce output close to the input. Higher values (0.5–0.9) give more creative interpretations. 1.0 = maximum variation. Only used when audio input is connected. Recommended: 0.5–0.75 for musical variations.

Outputs (1)

NameTypeDescription
AUDIOAUDIO