Music 2.6 (BudgetPixel)

MiniMax Music 2.6 is the full-length song generator of the pack: give it a style prompt and it produces a complete track, and if you don't have lyrics it will write or polish them for you. The default lyrics_optimizer setting is on, which is the killer feature if you're not a songwriter - you type "a melancholy indie ballad about a road trip," and the model handles the words and the music together. Flip it off and it will sing whatever lyrics you paste in, verbatim. At 200 credits per generation it's the pack's premium music tier, on par with Music 3.0.

As with every BudgetPixel node, nothing runs locally. The request goes to the API, the track renders server-side, and a WAV (default) or MP3 comes back as a core AUDIO value in your graph. No GPU, no weights, no VRAM - just a BudgetPixel API key and a credit balance.

The inputs

• prompt - the style/mood description, up to 2000 characters. Optional if you're providing lyrics.
• lyrics - your own words, 1–3500 characters. Required unless instrumental is on or lyrics_optimizer is on (the optimizer writes from the prompt).
• lyrics_optimizer - default true. Lets the model write or polish lyrics from your prompt. Set it false to force it to sing exactly what you wrote.
• instrumental - skips vocals entirely for a pure instrumental track.
• format - wav (default, lossless) or mp3 (smaller file).

Two outputs: audio and video. The video socket is a shared schema artifact that stays unconnected for audio-only models. Wire audio into Save Audio.

Music 2.6 vs Music 3.0

The pack ships both, at the same 200 credits. 3.0 is the newer MiniMax music model and the default pick for most people - if you're choosing, start there. 2.6 remains a useful fallback when 3.0's output feels wrong for the job or when you've built a workflow around its behavior and don't want to re-tune. Honestly, for one-off tracks the differences are subtle enough that either works; pick 3.0 unless you have a reason not to.

Install and common issues

Install 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

Restart and find it under Add Node → BudgetPixel → Audio. Set your API key via BUDGETPIXEL_API_KEY or a budgetpixel.json file - there's no key widget, deliberately, so your key never rides along in shared workflow JSON.

The pack-wide caveats: you need a 2025+ ComfyUI for the AUDIO type, and Stop doesn't cancel a submitted render - it abandons the wait while the job keeps running and still bills the full 200 credits. And the one real user-facing trap here is the lyrics logic: if you've written lyrics but lyrics_optimizer is still on, the model may "polish" them into something you didn't write. Turn it off when the exact words matter.