Qwen3-TTS Load Prompt

Qwen3-TTS Load Prompt is the reuse half of the pack's voice-library feature. If you've saved a voice embedding with Qwen3-TTS Save Prompt, this node is how you pull it back into a graph - a dropdown of every saved voice, one click, and the embedding is ready to feed a Voice Clone node. No reference audio to re-upload, no re-analysis, no recompute. If you've ever been annoyed that your cloned voice doesn't survive a workflow restart, this is the fix.

How it works

The node scans ComfyUI/models/Qwen3-TTS/prompts/ for .safetensors files - the exact folder Save Prompt writes to - and gives you a prompt_file dropdown to pick one. That's the only input, and it's the whole interaction. It reads the embedding back into a QWEN3_PROMPT (the same type Prompt Maker produces), and that output plugs straight into the prompt input on Qwen3-TTS Voice Clone.

The clever bit is invisible: Load Prompt rebuilds the full prompt object from the safetensors file, including its metadata - whether the original was an x-vector-only or in-context-learning prompt, and the reference text if there was one. So the embedding doesn't just carry the voice, it carries enough context to behave like it did the day you saved it. The dropdown also updates when files appear or change; it keys off file modification time so it won't silently serve you a stale cached read.

When you actually need it

This node only earns its place in the workflow when you're generating a lot of speech with a fixed voice, or coming back to a voice across sessions. The canonical setup:

1. Once: reference audio → Qwen3-TTS Prompt Maker → Save Prompt (my_voice.safetensors)
2. Every time after: Load Prompt (pick my_voice.safetensors) → Voice Clone → text → WAV

One honest note: it's a single-value node with a hard dependency on having saved something first. If you run it on a fresh install with an empty prompts folder, it shows a placeholder "no prompts saved yet" entry and errors if you try to run it. That's not a bug - it's the node telling you to run the save side first.

Installing & troubleshooting

Same pack-level install as the rest of ComfyUI-Qwen3-TTS: ComfyUI Manager (search "ComfyUI-Qwen3-TTS") or git clone https://github.com/DarioFT/ComfyUI-Qwen3-TTS into custom_nodes, then pip install -r requirements.txt yourself - ComfyUI won't auto-install it, and the qwen-tts dependency's transformers==4.57.3 pin can downgrade a shared environment.

The two errors you'll meet: "no prompts saved yet" (go run Save Prompt once) and "prompt file not found" (you deleted or renamed the file out from under the dropdown - refresh it). After that it's about as failure-proof as a node gets: it doesn't touch the GPU, doesn't load a model, doesn't download anything. It reads one file and hands you a prompt.