EmAySee Scenema Audio Connector

Read the name as "Scenema Audio Connector," because that's the honest job: it doesn't synthesize anything itself. You give it a prompt, it POSTs it to a TTS API server - the default URL is http://10.0.0.71:8000/generate, a LAN address from the author's own setup - and it turns the returned WAV into a ComfyUI AUDIO tensor you can feed straight into a video or audio output node. No model downloads, no local TTS weights. Just a client for a server that's already running somewhere.

How it works

The prompt field is interesting: it's written in a <speak>-style markup with a voice attribute and <action> tags - the kind of rich, SSML-flavored prompt an expressive TTS model (think "Scenema"-style voice acting servers) expects. The default gives you a working example of the format. The node wraps your prompt, seed, steps, guidance, and sample rate into JSON, sends it to api_url, and decodes the base64 WAV response back into an AUDIO output. If you also feed an reference_audio (AUDIO) input, it re-encodes that clip and sends it along too - voice-cloning support, assuming your server accepts it.

The knobs that matter:

• api_url - the only input you're guaranteed to change. Point it at your TTS server's /generate endpoint. The default is someone's home network IP and will not work for you.
• prompt - the speech text plus voice/directing markup.
• seed (42), steps (50), guidance_scale (7) - sampler settings your server interprets.
• output_sample_rate (24000 / 44100 / 48000, default 44100) - what the server should produce.

Outputs: audio (AUDIO) and status (STRING) - "Success" on a good run, or an error string like Connection Failed: ... when the server is unreachable.

The two failure modes you'll actually hit

Server unreachable. The default IP is from the author's LAN. Unless your TTS server lives at that exact address, every run returns (None, "Connection Failed: ...") - the node won't hang, but it won't produce audio either. Fix api_url first, every time.

Server shape mismatch. This node assumes the response JSON has an "audio" key holding a base64 WAV, and that the request body matches {prompt, seed, steps, guidance_scale, sample_rate}. Not every TTS API speaks that contract. If your server uses different field names or returns raw audio, you'll see the "JSON missing 'audio' key" status. It's a connector for a specific API shape, not a universal wrapper.

It also needs requests installed in your ComfyUI Python env - the pack ships no requirements.txt, so if you get a requests import error, pip install requests (it's almost always already present).

Installing it

Part of ComfyUI_EmAySee_CustomNodes. ComfyUI Manager → search the pack title → Install, or:

cd ComfyUI/custom_nodes
git clone https://github.com/EmAySee/ComfyUI_EmAySee_CustomNodes
# restart ComfyUI

Under EmAySee/Audio, no model files.

Bottom line

Useful if you already run a compatible local TTS server (or want to stand one up) and you want its output to flow into ComfyUI video work. It's a client, not a generator - budget your time for wiring your server to its expected request/response shape.