Skip to main content
Pika’s audio nodes turn a written description into finished audio without a local GPU. Pika Music writes a song from a style prompt plus optional lyrics, at a length you choose from 10 to 360 seconds, and it can cover an uploaded track at that track’s own length. Pika SFX returns one sound effect from 1 to 20 seconds, for foley, UI blips, risers and trailer hits. Neither node downloads weights. Both hand back a single AUDIO output, run on your Comfy account, and are billed in credits by the second of generated audio. What these nodes are for
  • A cue, jingle or instrumental bed cut to the exact length of a video edit
  • A song from lyrics you already wrote, in a style you can change without touching the words
  • A restyled version of a track you own: new genre, new tempo, the same song underneath
  • Foley, impacts, whooshes, room tone and game event sounds for a finished picture
To use the Partner Nodes, you need to ensure that you are logged in properly and using a permitted network environment. Please refer to the Partner Nodes Overview section of the documentation to understand the specific requirements for using the Partner Nodes.
Make sure your ComfyUI is updated.Workflows in this guide can be found in the Workflow Templates. If you can’t find them in the template, your ComfyUI may be outdated.If nodes are missing when loading a workflow, possible reasons:
  1. You are not using the latest ComfyUI version (Nightly version)
  2. Some nodes failed to import at startup

​
Get started

  1. Update ComfyUI to 0.45.0 or later
  2. Go to Template Library, search for Pika Music or Pika SFX
  3. Write the prompt, set duration, then run the workflow

​
Pika Music: text to music

Write the arrangement, the mood and the instrumentation into the prompt, then add lyrics if you want a vocal. The node returns one finished track, mixed by the model. Pika Music text to music workflow preview

Download Workflow

Download JSON or search “Pika Music” in Template Library
  1. Choose text to music in the Pika Music node.
  2. Write the brief into prompt: genre and era, BPM and key, the instruments that carry the hook, the vocal character, and how the song moves from section to section. Up to 4,000 characters reach the model.
  3. Add lyrics with section tags such as [Verse] and [Chorus], or leave the field empty for an instrumental. Up to 8,000 characters.
  4. Set duration between 10 and 360 seconds, 30 by default. Turn on rewrite_prompt only when your prompt is a few words and you want a model to expand it first.
  5. Run the workflow. Save Audio (Advanced) writes an MP3 to ComfyUI/output/audio/.

​
Pika Music: cover a song

Switch the same node’s mode to cover to restyle a song you already have instead of writing a new one. Connect an audio file to source_audio, then write the brief as the arrangement you want instead: a different genre, a different tempo, a different band. Cover ignores duration. The result keeps the length of the source, clamped to 5 to 360 seconds, so a three-minute upload comes back three minutes long. The lyrics input still applies, which means you can keep the original words or hand the model a new set for the same arrangement. A cover works from a mix, so separation is the model’s job. A centered lead vocal and a simple arrangement survive a restyle more reliably than a dense master.

​
Pika SFX: text to sound effects

Name the sound in one line and the node returns a single effect. Pika follows the brief without adding speech, music or background noise you did not ask for, so an effect drops into a timeline without a bed you have to strip out. Pika SFX text to sound effect workflow preview

Download Workflow

Download JSON or search “Pika SFX” in Template Library
  1. Write the sound in the Pika SFX prompt, for example an old timber door creaking open on rusty hinges, then a heavy latch clicking shut.
  2. Set duration between 1 and 20 seconds, 5 by default, then run the workflow. The MP3 lands in ComfyUI/output/audio/.
Each run produces one effect at 44.1 kHz stereo. For a sequence of sounds, run the node once per sound and place them on the timeline yourself. Pika SFX does not write music or sung vocals; use Pika Music for that.

​
Writing a music prompt

Build the brief in this order: genre and era, BPM and key, the instruments that carry the hook, vocal character, arrangement arc, mix notes. The model reads all 4,000 characters, so detail earns its place; a stack of adjectives does not. The default prompt in the template shows the shape:
More starting points:
  • Mainstream dance pop, four-on-the-floor beat at 124 BPM in A minor, bright female lead in an upper-mid register, sidechained warm pads, plucked synth hook, filtered verse lifting into a festival-sized chorus, glossy radio mix.
  • Sparse solo grand piano at 68 BPM, felt hammers and close mic, rubato intro resolving into a steady left-hand arpeggio, no drums, no vocals.
  • Cinematic orchestral build at 90 BPM, taiko and low-string ostinato, brass swell around twenty seconds, choir pad underneath, wide trailer mix, instrumental.
  • Lofi hip hop at 82 BPM, dusty boom-bap drums, vinyl crackle, warm electric piano loop, upright bass, no vocals, loopable.
What changes the result:
  • Ask for the arc, not just the palette. “Filtered verse lifting into a full chorus” gives the track somewhere to go; a long instrument list on its own tends to come back as a static texture.
  • Say what you do not want. no vocals, no ad-libs, dry close-mic lead are all read as direction.
  • Name the delivery. Register, gender and how forward the vocal sits in the mix move the track more than one more instrument does.
  • Match the length to the job. A 10-second sting and a 180-second cue want different briefs, so generate the length you need instead of cutting a long song down.
  • Iterate one dimension at a time. Keep the seed while you edit the prompt, so the difference you hear comes from the change you made.

​
Lyrics and section tags

lyrics is a separate input from the prompt, so the same words can come back as a ballad, a dance single or a rock performance when only the brief changes.
  • An empty lyrics field returns an instrumental on the same brief, with no vocal.
  • Tags set the structure. They are read as phrasing and section boundaries rather than text to sing. [Verse], [Pre-Chorus] and [Chorus] cover most songs; add [Verse 2], [Bridge] or [Final Chorus] to name repeats and the ending.
  • Keep one lyric line per line. Line breaks are what the model phrases against.

​
Writing a sound effect prompt

One line, and four things in it: what the material is, what happens to it, the space it happens in, and how close the mic is. Intensity goes last. A close, dry description reads better than a list of adjectives.
  • One event per run. A door opening and a latch closing share one prompt only when they land in that order inside the duration you set.
  • Say no music, no voices when the effect has to sit under dialogue. It keeps the render clean.
  • Match duration to the event. A click is one second; a riser has to fill the gap in front of the cut.
  • Audition variants. Re-run with a new seed, then change one dimension: material first, then space, then intensity.

​
Duration, limits and credits

Both nodes bill in credits by the second of generated audio, so a 10-second sting costs a fraction of a 3-minute song. See Partner node pricing for the current rate. The same prompt, lyrics, duration and seed with rewrite_prompt off reproduce the same result. With rewrite_prompt on, a model rewrites the brief before generation, so the seed alone no longer pins the output. Leave it off when you want a reproducible workflow. Both nodes start at seed 42, so hold it while you edit the prompt and change it only to audition another take.

​
Output and saving

Both nodes output AUDIO, and the templates save it with Save Audio (Advanced): prefix audio/Pika_Music or audio/Pika_SFX, format mp3 at V0, which lands in ComfyUI/output/audio/. Switch the format to flac when you want a lossless file to master or re-cut. The output is an ordinary audio stream, so it composes with the rest of the graph:
  • Connect it to Create Video to attach the track to a set of frames
  • Run Pika SFX once per hit and layer the results in an editor, since one run gives one effect
  • Set the track duration to the length of the cut instead of trimming a longer song in the timeline

​
Which Pika audio node to use

For an effect timed to action on screen, the video path is Pika Soundtrack on the Pika H3 Flex page, which reads the clip instead of a description.