Lyria 3 (BudgetPixel)

Lyria is Google's take on music generation, and version 3 is the one worth actually using: high-fidelity tracks from a plain text description, optional lyrics, and - the part that separates it from most music nodes - up to ten reference images that condition the mood of the track. Want a song that sounds like a rainy Kyoto backstreet you have a photo of? Feed the photo in alongside the prompt. It's a level of control that the run-of-the-mill text-prompt-only music nodes don't give you.

It's an API node, like the whole BudgetPixel pack: the generation happens server-side on Google's model, billed at a flat 100 credits per track, and the MP3 comes back watermarked with SynthID. No local GPU, no model download, no VRAM anxiety - you just need a BudgetPixel API key. For the cost of a few Nano Banana images you get a full song, which makes it one of the pack's better credit-to-output ratios if you need music at all.

The inputs

• prompt (required) - a description of the music, 10–2000 characters. Style, tempo, genre, instrumentation: all fair game.
• lyrics - optional, up to 3000 characters. Provide your own words or leave it blank and let the model figure out a vocal-less arrangement.
• images - up to ten reference images to set the track's mood or theme. In the ComfyUI graph these are normal IMAGE connections, so you can pipe in anything you've already generated.

The node has two outputs: audio and video. For an audio-only model like this one the video socket just stays unconnected - it's a shared socket from the pack's schema builder, not a promise of a music video.

What to wire it into

The audio output is a ComfyUI core AUDIO value, so it plugs straight into a Save Audio node. From there you could drop it into a video's soundtrack, send it through a music-reactive video pipeline, or just render it to a file for a listen. The whole point of this node existing inside ComfyUI rather than a web page is that the track lands in your graph as a value, ready to be composited, trimmed, or looped by other nodes.

Install and common issues

Install the pack via ComfyUI Manager (search "BudgetPixel") or:

cd ComfyUI/custom_nodes
git clone https://github.com/budgetpixel-ai/comfyui-budgetpixel.git
pip install -r comfyui-budgetpixel/requirements.txt

Then restart, and the node is under Add Node → BudgetPixel → Audio. You'll need a BudgetPixel API key from a paid plan, set as BUDGETPIXEL_API_KEY or in a budgetpixel.json file - the nodes deliberately have no key widget, so your key never ends up in a shared workflow.

The usual failure modes are shared with the rest of the pack. Missing key means a clear MissingAPIKeyError. A recent ComfyUI (2025+) is required for the AUDIO and VIDEO output types. And because generation is async server-side, remember that stopping the node while it's polling only abandons the wait - a submitted job keeps running and still bills its 100 credits, so don't spam Stop to "cancel" a track you regret.

One tip: if the lyrics matter more than the vibe, Lyria shares the pack with Mureka V9, which sings your exact lyrics with structure tags and costs less - Lyria's edge is quality and the image conditioning, not lyric fidelity.