Qwen TTS 🐑

If you've built a video or avatar workflow in ComfyUI and hit the point where you need a voice to go with it, this node gives you one without leaving the graph: text in, audio out, generated by Alibaba's Qwen3-TTS-Flash model over their API. It's a cloud call, not a local model - nothing downloads, nothing runs on your GPU. That's a genuinely different category from most nodes in this pack's ecosystem, and worth knowing going in.

What it does and how. You type or wire in text, pick a voice and a language_type, and the node sends that off to Qwen's TTS endpoint and hands back an audio output you can plug straight into a Save Audio node, a video's audio track, or anything else downstream that expects ComfyUI's AUDIO type. Alibaba's Tongyi lab - the umbrella behind Qwen, Wan, and Z-Image, and the source most of the local generation community has defaulted to for open weights through 2025–2026 - ships Qwen3-TTS as a hosted API product rather than an open checkpoint, so this node is calling out to Alibaba's cloud the same way the pack's LLM connector nodes call out to SiliconFlow, DeepSeek, or Gemini. It needs a valid API key to do anything at all.

The parameters that matter.

• text (multiline, required) - what gets spoken. Ships with a Chinese placeholder ("你好，我是通义千问语音合成助手" - "Hi, I'm the Tongyi Qianwen speech synthesis assistant"), which is a decent hint about which language this model was built around first, even though it explicitly supports English too.
• voice - a dropdown of 24 preset voices (default Cherry; others include Ethan, Chelsie, Serena, Dylan, and more). Pick one that suits the register you're after; there's no custom-voice cloning here, just the preset roster.
• language_type - Chinese (default) or English. Match this to your text, since it almost certainly steers pronunciation and prosody rather than being purely cosmetic.
• model - currently locked to qwen3-tts-flash, the only option on the schema right now.
• api_key (optional STRING) - paste your Alibaba API credential here directly for a quick test.
• tts_connector (optional, type TTSConnector) - a companion connector node elsewhere in this pack presumably handles centralized key management, the same pattern the LLM connector nodes use with mie_llm_keys.json. That connector node isn't covered in this write-up, so if you want to avoid pasting a raw key into every TTS node in your graph, look for it under this pack's node categories in ComfyUI Manager or the node search.

Installing. ComfyUI Manager: search ComfyUI_MieNodes and install. Manual: cd ComfyUI/custom_nodes && git clone https://github.com/MieMieeeee/ComfyUI-MieNodes, then restart ComfyUI. No model weights to download - the heavy lifting happens on Alibaba's servers, not your machine.

Worth flagging plainly: neither this node nor Qwen's TTS product shows up anywhere in the pack's README or in the knowledge base this article draws on - the README documents ten-plus LLM text services in detail but says nothing about TTS at all. Everything above beyond the raw schema is inferred from how the rest of this pack's cloud-connector nodes behave and from Alibaba Tongyi's general API pattern, not confirmed against author documentation for this specific node.

Troubleshooting.

• Node errors out immediately with no audio. Almost certainly a missing or invalid api_key - with no tts_connector wired in and an empty api_key field, there's nothing to authenticate the request with.
• Audio comes back but pronunciation sounds off. Check language_type actually matches your text - feeding English text through the Chinese setting (or vice versa) is a likely cause given the two are separate, explicit choices rather than auto-detected.
• Long scripts get cut off or the call fails. Cloud TTS APIs generally cap input length per request; if you're voicing a long passage, split text into chunks and generate multiple AUDIO clips rather than sending the whole script in one call.
• This is a paid, metered API call, not a free local model. Every generation costs against your Alibaba account's usage - budget for it the same way you would any other cloud LLM or TTS call in this pack, and don't expect it to behave like the free-tier models Alibaba is known for on the image and video side.