F5tts
App in the BluixApps catalog
What it is
F5-TTS is state-of-the-art zero-shot TTS by Shanghai AI Lab — flow matching + DiT architecture, faster than XTTS-v2 with higher fidelity, voice cloning from 10-second samples. The newer challenger to XTTS in the open TTS leaderboard.
When you need TTS quality that approaches commercial APIs, F5-TTS is the open option.
What it's for
- High-fidelity TTS — closer to commercial APIs than XTTS
- Voice cloning from 10-second reference
- Native English + Chinese — primary languages, with community LoRAs for others
- Voice chat mode — TTS + ASR loop for conversational systems
- Faster inference than XTTS — 3-5× real-time on RTX 3090
- Emotion + prosody control — nuanced delivery options
Who it's for
- Premium audio content creators demanding closer-to-commercial quality
- Voice chat product builders (companion AI, assistant interfaces)
- Audiobook producers for English / Chinese content
- AI startups integrating high-quality TTS in their stack
- Hosting providers selling premium voice tier
Why teams pick F5-TTS over alternatives
- MIT license — fully open
- Highest quality in late-2024 open TTS benchmarks (matches/beats XTTS for EN/ZH)
- Faster than XTTS by 2-3×
- Better emotion control than XTTS
- Voice chat mode built-in — TTS + ASR ready loop
- Active research backing — Shanghai AI Lab + community
Integrations
- Gradio web UI with multi-speech + voice chat tabs
- Gradio API auto-exposed at
/api/predict/0 - HuggingFace Diffusers-style pipeline
- Pair with LLM — voice chat with Ollama/vLLM
- Pair with Whisper — voice chat loop (ASR → LLM → F5-TTS)
- Community LoRAs for additional languages (Italian, Spanish, French)
Notable users & community
- 9k+ GitHub stars
- Shanghai AI Lab + community development
- Featured in late-2024 TTS leaderboard upsets
- Active Chinese + English community
- Multiple commercial integrations starting
Tips & operations
- Reference voice: 10-30 sec clean recording, low noise
- English + Chinese native; other languages via community LoRAs
- VRAM: 8 GB GPU recommended
- Voice chat loop: enable TTS+ASR tabs for full conversation
- Storage: model weights ~1.5 GB (lighter than XTTS)
- Latency: ~200ms first token (production-grade)
- License clean — MIT, commercial OK
- Compare with XTTS: F5 wins on quality for EN/ZH; XTTS wins on language coverage
What we ship in BluixApps
- Cloned
SWivid/F5-TTSrepo, pip-installed - pytorch/pytorch CUDA 12.4 base + ffmpeg
- Gradio launcher (
infer_gradio) - Persistent volumes: repo, models (~1.5 GB), output
- Port 7867 mapped
- Install report at
/root/bluixapps/f5tts.txt - Acceptable Use Policy notes (voice cloning ethics)
- Pairing suggestions (XTTS for language coverage, Whisper for voice chat)
- GPU pre-flight check via
bluixapps_ensure_nvidia_runtime - Backup hook covers models + outputs
Get this app — pick a BluixApps plan
Same catalog. Scaling tenant isolation, white-label and support tier.
| Tier | Tenants | Catalog | Support | White-label | Monthly | |
|---|---|---|---|---|---|---|
| Stacks | 1 | 19 curated stacks | Standard | — | $19/mo | DetailDeploy |
| Starter | 10 | Full catalog | Standard | +$15–25/mo | $49/mo | DetailDeploy |
| Pro | 25 | Full catalog | Priority bugfix | +$15–25/mo | $149/mo | DetailDeploy |
| Growth | 100 | Full catalog | Priority bugfix | +$15–25/mo | $349/mo | DetailDeploy |
| Scale | 500 | Full catalog | 7-day window | +$15–25/mo | $799/mo | DetailDeploy |
| Enterprise | Unlimited | Full catalog | Priority 7-day | Bundled | $1,499/mo | DetailDeploy |