AceStepCohere_Lyrics

The interesting thing about Cohere's node in this pack isn't the default model - it's the dropdown. Alongside the Command family sit the Aya models, which are Cohere's multilingual line. If you want lyrics in Portuguese, Spanish, Japanese, or a dozen other languages that the OpenAI/Claude nodes handle with a heavy English accent, this is the node to pick. Given this pack's prompt library leans hard into Brazilian music (Funk Carioca, Forró, Sertanejo...), the fit is obvious.

The mechanism

Identical family skeleton: text holds your song description, the node wraps it in the ACE-Step production-tag lyricist prompt, and it calls Cohere's API with your api_key, model, max_tokens (1024 default), seed, and instrumentality_balance. Output is a lyrics STRING ready for your text conditioning. Nothing exotic under the hood - urllib straight to Cohere's endpoint, no SDK.

Model choices worth knowing

Default is command-a-03-2025, Cohere's flagship, and it's a solid generalist for lyrics. But the dropdown's personality is in the specialized entries:

• command-a-reasoning - for when you want the model to think through structure (verses/bridges/chorus mapping) before writing. Marginal benefit on a lyric sheet, honestly.
• command-a-translate - a translation specialist. Interesting if you draft in one language and want the model to handle localization... except it writes in one shot, so you'd pass the source text as your text input and let it translate into the format.
• aya-expanse-32b / 8b - the multilingual workhorses. The 32B is genuinely good in several languages and is the one to reach for on non-English songs.
• aya-vision / command-a-vision - vision models; in a text-only node they're a curiosity more than a tool.

Gotchas

• No temperature parameter is sent, so Cohere's defaults apply.
• The specialized Command variants (translate, reasoning, vision) are listed because they're on Cohere's API, not because they're tuned for songwriting. The node sends the same lyricist prompt to all of them; the model does the interpreting.
• Standard family caveats apply: api_key is a node input that rides along in saved workflows, and API errors come back as the lyrics string.

Install is the pack standard - ComfyUI Manager (search "JK AceStep Nodes") or git clone https://github.com/jeankassio/JK-AceStep-Nodes.git into ComfyUI/custom_nodes/, restart. The node adds no extra dependencies beyond the pack's torch/numpy/tqdm/librosa.

Honest verdict: for English lyrics, OpenAI's node is the smoother default. This one earns its place the moment your song idea isn't in English.