ComfyUI-MiMoASR CLI

Drives the MiMoASRLoader + MiMoASRTranscribe nodes over ComfyUI's HTTP + WebSocket API. Same ComfyClient template used across the other ComfyUI-* packages.

Quick start

# uv handles dependencies via the PEP 723 inline script metadata
uv run python mimo_asr_cli.py /path/to/speech.wav --ip 10.8.0.12

# Explicit Chinese tag (useful for accented or noisy speech)
uv run python mimo_asr_cli.py speech.wav --language Chinese --ip 10.8.0.12

# Pass model + repo paths if your ComfyUI server can't see the defaults
uv run python mimo_asr_cli.py speech.wav \
    --model-path /mnt/models/MiMo-V2.5-ASR \
    --tokenizer-path /mnt/models/MiMo-Audio-Tokenizer \
    --repo-path /mnt/repos/MiMo-V2.5-ASR \
    --ip 10.8.0.12

# Save transcript to file
uv run python mimo_asr_cli.py speech.wav --out transcript.txt --ip 10.8.0.12

Flags

Flag	Default	Notes
`audio` (positional)	—	Local audio file. Uploaded to ComfyUI's `input/`.
`--ip`	`127.0.0.1`	ComfyUI host.
`--port`	`8188`	ComfyUI port.
`--server`	—	Full `host:port`, overrides `--ip`/`--port`.
`--out`	—	Write transcript to file.
`--model-path`	`./models/MiMo-V2.5-ASR`	Relative to ComfyUI cwd, or absolute.
`--tokenizer-path`	`./models/MiMo-Audio-Tokenizer`	Same.
`--repo-path`	(env `MIMO_ASR_REPO`)	Local clone of XiaomiMiMo/MiMo-V2.5-ASR.
`--language`	`Auto`	`Auto` / `Chinese` / `English`.
`--dump-workflow PATH`	—	Save the API-format JSON for inspection.
`--no-run`	off	Combine with `--dump-workflow` to skip submission.
`--timeout`	`600`	Seconds.

Server-side checklist

# 1. ComfyUI is up
curl -s -o /dev/null -w "HTTP %{http_code}\n" http://SERVER:8188/

# 2. Nodes loaded
curl -s http://SERVER:8188/object_info | python3 -c "
import json, sys
d = json.load(sys.stdin)
for n in ('MiMoASRLoader', 'MiMoASRTranscribe', 'LoadAudio', 'PreviewAny'):
    print(f'  {n}: {\"✓\" if n in d else \"✗\"}')"

PreviewAny ships with rgthree-comfy (and several others). If it's not installed, swap the preview node in build_workflow() for whatever your environment provides — ShowText|pysssss is a drop-in replacement.

Audio format notes

/upload/image accepts any file (the endpoint's name is misleading). Audio gets stored under ComfyUI's input/ folder, and LoadAudio handles decoding via torchaudio — so wav / mp3 / flac / m4a / ogg all work.