audioTTSVoice CloneVoice DesignAudio

Qwen 3 TTS

Text-to-speech with voice cloning and voice design using Qwen3-TTS models. Supports three modes: Voice Design (generate speech from text + voice description), Voice Clone (clone from reference audio), and Custom Voice (predefined speakers with style instructions).

Launch on GPU Get Started

Features

Voice Design - Generate speech from text + natural language voice description (1.7B only)
Voice Clone - Clone a voice from reference audio (0.6B or 1.7B)
Custom Voice - Use predefined speakers with optional style instructions (0.6B or 1.7B)

Requirements

GPU	16GB+ VRAM (24GB recommended for 1.7B models)
Docker Image	`ghcr.io/nedos/qwen3-tts-gradio:latest`
Port	7860

Available Models

Model	Size	Type
`Qwen3-TTS-12Hz-0.6B-Base`	0.6B	Base
`Qwen3-TTS-12Hz-1.7B-Base`	1.7B	Base
`Qwen3-TTS-12Hz-1.7B-CustomVoice`	1.7B	CustomVoice
`Qwen3-TTS-12Hz-1.7B-VoiceDesign`	1.7B	VoiceDesign

Usage

Python Client

python

from hypercli import HyperCLI

client = HyperCLI()

# Launch Qwen 3 TTS
job = client.jobs.create(
    image="ghcr.io/nedos/qwen3-tts-gradio:latest",
    gpu_type="l40s",
    ports={"lb": 7860},
    env={"PRELOAD_MODELS":"CustomVoice:1.7B,Base:1.7B"},
    auth=True,
)

print(f"Job {job.job_id} launched!")
print(f"Access at: https://{job.job_id}.job.hypercli.com")

Install with: pip install hypercli-sdk

CLI

bash

hyper instances launch ghcr.io/nedos/qwen3-tts-gradio:latest \
  --gpu l40s \
  --port lb:7860 \
  --env PRELOAD_MODELS="CustomVoice:1.7B,Base:1.7B"

Install with: pip install hypercli-cli

Docker (Local)

bash

docker run --gpus all -p 7860:7860 \
  -e PRELOAD_MODELS="CustomVoice:1.7B,Base:1.7B" \
  ghcr.io/nedos/qwen3-tts-gradio:latest

Web Console

Use the web console to launch with a pre-filled form:

Open in Console

Environment Variables

Variable	Example
`PRELOAD_MODELS`	`CustomVoice:1.7B,Base:1.7B`

Ready to run Qwen 3 TTS?

Deploy on HyperCLI GPUs and access via web browser.

Launch on GPU View Documentation