zhenzhen-nemotron-asr-multilingual-fal

Comfly_nemotron_asr_multilingual_fal ("zhenzhen-nemotron-asr-multilingual-fal") does one thing and does it cleanly: speech-to-text. It's NVIDIA's Nemotron ASR model called as a cloud API, and it hands you back a transcript string - no model files, no local install, no GPU. Feed it an audio file from the canvas (or a URL) and a language, and out comes text. It's the node you reach for when a video pipeline needs captions, a voiceover needs a script back, or you want to pipe spoken content into a prompt-enhancer node downstream.

It's a FAL-class node in the T8mars/Comfyui-zhenzhen pack, the "Comfly" API-wrapper family that calls the author's paid reseller service. Key goes into api_key directly; FAL billing is pre-deduct and settles after the run. Submit → poll → download, same machinery as the pack's other FAL nodes - the difference is that the "result" here is text, not an image or video.

The inputs that matter

• audio_url - required, but note the tooltip: it's ignored when you connect an AUDIO input. For most people the workflow is: generate audio upstream, wire it in, done. The URL field is the fallback for hosted files.
• audio - the AUDIO input that takes precedence.
• language - auto or one of ~40 locale codes (en-US, en-GB, es-US, de-DE, fr-FR, ja-JP, ko-KR, …). auto is usually right; set it explicitly if you're getting mis-detected languages.
• acceleration - none, regular (default), high, or full. This is a quality/speed trade - more acceleration, faster but more error-prone. The default is a sane middle.
• audio_way - upload (default) or audio_url, mirroring how you're handing over the file.
• poll_interval / max_poll_attempts - default 6s × 600 = a one-hour timeout ceiling (generous for transcription).
• skip_error - fail soft for batch runs.

Outputs

• transcript - the transcribed text as a STRING. Wire it to a text display, a prompt field, or a save node.
• response - the raw API payload, which carries the timing/detail if you need it.

Installing it

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

Or ComfyUI Manager → "Comfyui-zhenzhen", then a key from the author's site into api_key.

The honest part

ASR is one of the more underrated cloud jobs - locally, a decent multilingual model wants a real setup, and this node is "wire audio in, get text out" with no fuss. The multilingual coverage and the explicit locale list are genuinely good. The caveats are the pack's usual ones: it's a reseller proxy, so your audio leaves your machine (fine for scripts, think twice about anything sensitive), and per-call pricing means you don't want to transcribe a ten-hour podcast here. For occasional captions and scripts, though, it's the least-effort ASR in ComfyUI you'll find.