GROQ Audio Processor

Here's the honest review before the how-to: this node doesn't do anything yet. GROQ Audio Processor ships as a UI shell with no implementation behind it. The function that's supposed to process your audio is literally an empty pass in the source. You can wire it up, queue your workflow, and nothing will come out - on a modern ComfyUI it errors out rather than silently skipping. So read this as "what this node is supposed to be," and don't build a workflow on it until the author finishes it.

What it's reaching for

The idea is neat: feed ComfyUI's AUDIO type into a node, and let GROQ's hosted speech models transcribe it. That's a genuinely useful thing to do in a ComfyUI graph - think turning a voiceover track into a text string you can feed downstream, or captioning an audio source the way you'd caption an image. The node exposes four task modes - transcribe, translate, summarize, and sentiment - which tells you the ambition: not just speech-to-text, but post-processing the transcript.

What the schema says

The inputs are straightforward. api_key (leave empty to use GROQ_API_KEY), audio_data (an AUDIO-type connection - in ComfyUI that usually comes from an audio loader or a Video Helper Suite node), task, language (an ISO code like en or es), and temperature (0–1). The two outputs are text (the transcript) and confidence (a float for how sure the model is).

For comparison, a real implementation would call GROQ's audio API - client.audio.transcriptions.create with a whisper model like whisper-large-v3-turbo - then hand the returned text back. None of that code exists in this pack; only the process_audio stub does.

Install

Same story as the rest of this pack (ComfyUI_ComfyGroq / GROQ in ComfyUI Manager, or clone https://github.com/downlifted/ComfyUI_GROQ-PromptWizard into custom_nodes). It needs pip install groq and a GROQ API key. No model downloads - everything runs on GROQ's cloud.

If you actually need audio transcription

Don't wait on this node. GROQ's hosted whisper is fast and the free tier is generous, but the direct route is just the Python SDK in a script. Inside ComfyUI, a local whisper setup will cost you VRAM but keeps audio on the box. Either way you'll get a working transcript today instead of a pass statement. The one thing this node has going for it is the confidence output - few whisper UIs bother exposing it - so if the author ever finishes it, it might genuinely earn its place.