Docker image available: s2-tts with Q8_0 GGUF for 12GB VRAM

[orrinwitt]

Hey — I built a Docker image that packages s2.cpp with the Q8_0 GGUF model for easy deployment on GPUs with 12GB VRAM.

Image: ghcr.io/orrinwitt/s2-tts:latest
Repo: https://github.com/orrinwitt/s2-tts

Features:

• Q8_0 model and tokenizer baked into the image (no external downloads)
• CUDA support via NVIDIA Container Toolkit
• HTTP server mode on port 3030 (/generate endpoint)
• Voice cloning via multipart form data
• S2-Pro [bracket] emotion tag syntax supported
• Runs on RTX 3060 12GB with room to spare

Quick start:

services:
  s2-tts:
    image: ghcr.io/orrinwitt/s2-tts:latest
    ports:
      - "3030:3030"
    deploy:
      resources:
        reservations:
          devices:
            - driver: nvidia
              device_ids: ['1']
              capabilities: [gpu]

API-only (no WebUI) — designed for programmatic use. See the README for full docs including emotion tags, voice cloning, and text formatting rules.

Thanks for the great work on s2.cpp!

[rodrigomatta]

Hey @orrinwitt, that's really cool, thanks for putting this together! A ready-to-run Docker image with the model baked in is a great way to lower the barrier for people who just want to get started quickly.

I'll take a closer look when I get a chance. If you'd be open to opening a PR, I'd love to find a way to reference this in the repo, even if just linking it from the README. Either way, appreciate the effort!