HttpTTSToAudio

You've built the video workflow. Now the character needs to talk, the tutorial needs a voiceover, the storyboard needs narration. That's what HttpTTSToAudio is for: it turns plain text into a spoken WAV file right inside your ComfyUI graph, using Microsoft Edge's neural voices. No API key, no GPU cost, no giant model download - just an internet connection and a small local server you start once. It ships from GeekatPlay Studio, a YouTube creator who publishes ComfyUI tutorial content, and it's clearly built to be demonstrated rather than just parked in a repo.

What it actually does

The name says "Http" for a reason - this node is an HTTP client, not a TTS engine. The real work happens in tts_server.py, a tiny Flask server that runs separately on http://127.0.0.1:5002. The node POSTs your text, voice, rate, and volume as JSON to /tts, the server synthesizes audio, and the node copies the resulting WAV into your ComfyUI output directory (or wherever you point it) with a timestamped, descriptive filename.

Under the hood the server uses edge-tts, the Python library that talks to Microsoft Edge's free online TTS endpoint. It's unofficial and requires internet, but it's been the community's favorite free neural-voice hack for years. If Edge is unreachable, the server silently falls back to pyttsx3, which uses your OS's system voices - offline, but noticeably more robotic. That fallback is nice to have; just don't expect Neural quality from it.

The inputs that matter

• text - multiline, your actual words. The only thing you'll touch most runs.
• voice - a 17-option dropdown of Edge Neural voices across US, GB, AU, CA, and IN English. Default is en-US-AriaNeural, which is a perfectly pleasant female voice. Names follow the en-US-GuyNeural pattern, so you can tell dialect and gender at a glance.
• rate (50–400, default 180) and volume (0.0–1.0) - the server converts these to Edge's percentage format relative to the 180 baseline, so you get real control, not made-up numbers.
• text_file_path and output_directory - file pickers. Leave the output directory blank and it lands in ComfyUI's default output folder, which is what you want 90% of the time.
• timeout_seconds + auto_timeout - for long scripts. Leave auto_timeout on; it scales the request timeout with word count (roughly words ÷ 2.5 seconds, plus a buffer) so you don't get a false failure on a 2,000-word script.

The output is a single audio_path STRING pointing at the WAV. That's the pack's own recommended wiring: feed it into VHS_LoadAudio's audio_file input, then out through WAS_SaveAudio. The pack's example workflow does exactly this, and it's the cleanest way to get generated speech into a video pipeline.

Installing it - the part people miss

Install is easy: ComfyUI Manager (search "ComfyUI-Text2Speech") or cd ComfyUI/custom_nodes && git clone https://github.com/GeekatplayStudio/ComfyUI-Text2Speech, then restart. Dependencies are light - edge-tts, pyttsx3, flask, requests - no torch, no model files. On the standalone Windows build you may need install.bat or pip with the embedded Python.

Here's the trap: the node does nothing until the server is running. Start it with python tts_server.py (or double-click run_tts.bat on Windows) and leave it in its own terminal. You're now running two processes for one feature - slightly clunky, but it's also why the node is so fast.

Troubleshooting

• "Could not connect to TTS server" - the server isn't up. Start it, and make sure nothing else grabbed port 5002.
• No audio - check your internet. Edge TTS is online-only; the pyttsx3 fallback kicks in but uses system voices.
• Timeout on long text - keep auto_timeout on. For genuinely huge scripts (10k+ words) the README suggests splitting into chunks.

Honestly, this isn't the best-sounding TTS in existence - F5-TTS and the new local diffusion models are better. But you don't want to babysit a 335M-parameter model inside every workflow just for narration. This is free, instant, and genuinely good enough for voiceovers, character lines, and accessibility. It's the one I'd reach for when I want speech without thinking about it.