🌸 OVERTLI Pollinations Text to Music

The audio corner of the Overtli Studio Suite splits into two jobs, and this node owns the musical one. GZ_TextToAudio sends a text description to Pollinations' music generation endpoint and returns a track as a native ComfyUI AUDIO output - no Suno subscription, no local ACE-Step install, no GPU. You prompt for a genre and a mood, you get a music bed you can drop under a video in the same graph.

Keep the name honest though: "Text to Audio" is doing a lot of work, and in practice this is text-to-music-plus-soundscape. The model catalog is tagged [music] (and some entries double as [tts]), and the mode presets - Audio Prompt Enhance, Music Direction, Ambience and Foley - are written for exactly that: shaping a prompt for music generation, or steering toward layered environmental sound for foley work.

The inputs that matter

• text - your musical idea. "Lo-fi chill track, 80 BPM, warm vinyl crackle" beats "something nice" by a mile, and music_style_preset has you covered if you'd rather pick a direction (Cinematic Atmosphere, Epic Trailer, Synthwave Drive, Festival EDM Lift) than write one.
• mode_preset - Off, Audio Prompt Enhance, Music Direction, or Ambience and Foley. These rewrite your rough text into generation-ready language before the request goes out.
• model - 33 audio-capable entries fetched live from Pollinations' catalog; default is a free one tagged for both music and TTS.
• response_format - mp3 (default), wav, opus, aac, flac, pcm. Pick wav/flac if the track's going into serious post-processing; mp3 is fine for a draft bed.
• duration - seconds, default 10, up to 600. This is the knob that actually shapes a music bed - a 10-second loop and a two-minute piece are different jobs.
• instrumental - on to keep vocals out, which for music beds is usually what you want.
• speed - 0.25–4x playback speed adjustment (default 1).
• custom_instructions - free-form direction.

Output

One audio socket of type AUDIO. Wire it into an audio preview, or combine it under your video track. Like the other media nodes in the pack, the returned audio is a real tensor in current ComfyUI builds, so downstream audio nodes can touch it - though for heavy mastering you'd export and work in a DAW anyway.

Install and the realistic read

Standard pack install: search "Overtli" in ComfyUI Manager, or clone into custom_nodes, pip install -r requirements.txt, restart. No models to download - the generation happens on the provider's side.

The realistic read comes from the wider audio landscape (audio-generation.md): local open models like ACE-Step are where people go when they want control - training, consistent artist styles, everything offline. This node is the opposite end of the trade: fast, zero-setup, cloud, free-tier. For a quick instrumental bed under a draft video it's genuinely hard to beat, and foley-ish ambience presets cover a niche most local setups don't. What you give up is controllability and consistency - hosted music generation is effectively a dice roll per render, so don't expect the same track twice even with a fixed prompt, and the free tier is as reliable as any free hosted service (which is to say: sometimes slow, occasionally down). For temp audio and rapid iteration it's the right tool; for a final soundtrack you care about, treat it as a sketch pad rather than the master.