Supercapybara / Railway Templates

qwen3-tts

Self-hosted Qwen3-TTS on a pure-C CPU inference engine — no Python, no PyTorch, no GPU. OpenAI-compatible endpoint included.

Category: AI · One-click template for Railway by SuperSlowSloth

Alibaba's Qwen3-TTS served from a pure-C inference engine (a bon5co fork with a Linux server segfault fix): ~4–6 GB RAM for the 1.7B int8 model, ~3 GB for the 0.6B, weights (~3.4 GB) persisted on a volume at /models so redeploys don't re-download, health at GET /v1/health, and a streaming endpoint for lower perceived latency on CPU.

$13.88 in its first 30 days on the 2026-09-20 snapshot — the strongest young earner in the fleet and the clearest proof the tail can pay when the product is a component.

Deploy qwen3-tts on Railway

Disclosure: deploy links carry the author's Railway referral code — the author receives a usage kickback when you deploy. The template itself is free to deploy; Railway usage charges apply.

Earnings snapshot 2026-09-20: $13.88 in the last 30 days, 2 active deploys.

Wrapper repository: github.com/bon5co/qwen3-tts