zhenzhen-whisper-1-lowprice

Comfly_whisper_1_lowprice is the most boring node in the Comfyui-zhenzhen pack, and I mean that as a compliment. It takes an audio input and returns the transcription as text. That's it. No images, no video, no aspect ratios, no drama. In a pack full of big generative showpieces, this is the quiet utility that just works - and because it's an API call to OpenAI's Whisper-1 through Zhenzhen's domestic storefront, it's essentially free.

Why would you want this inside ComfyUI instead of a website? Because you're building pipelines. Transcription output feeds prompt-generation nodes, subtitle workflows, or even a caption that gets routed back into an image node. When the whole chain lives in the graph, the transcription belongs in the graph too.

How it works

Connect an AUDIO to the required audio input and the node uploads it, runs Whisper-1, and returns text (the transcription) plus response (the full API payload). The only real choice is response_format:

• json - plain transcription with segment offsets.
• verbose_json - the full detail: segments, per-word timing, confidence.
• srt / vtt - subtitle files, ready to write to disk for captioning.
• text - the bare string, if that's all you need.

model is locked to whisper-1. That's the whole interface.

Installing and setting up

Install via ComfyUI Manager (search Comfyui-zhenzhen) or:

cd ComfyUI/custom_nodes
git clone https://github.com/T8mars/Comfyui-zhenzhen

Restart ComfyUI. No models to download. This is a low-price-family node, so it takes no direct API key - connect the Comfly Seedance2 Low Price Settings node's api_config output, or set SEEDANCE_API_KEY. If you're already using any other node in this family, the settings node is already on your canvas; just wire it in.

Troubleshooting

The family's one tripwire applies: run it without api_config and it raises "empty api_key." Other than that, the common complaint is audio format or duration - very long audio can exceed upload limits, so chunk long recordings before transcribing. And if you're feeding it a video's audio, extract it first (any audio-extraction node) rather than expecting the node to read a video. Whisper handles most languages but its timing data is only in verbose_json or srt - if you need offsets, don't use plain json. skip_error is here too, but for a node this cheap and deterministic you can honestly leave it off.