Open self-hosted models PLANNED
The self-hosted lanes run on RunPod GPUs (A40 48GB / A100 80GB class) through the same generation queue. Tuned settings live in the model catalog; this page is the repo + license index.
Weights license is load-bearing. FlopCoin sells generations, so a model's weights (not just its code) must be commercially licensable. Several top open models have NON-commercial weights and are benchmark / prototype only. Always verify the weights license before building a lane.
Audio - TTS / voice clone
| Model | Role | Weights license | Repo |
|---|---|---|---|
| Chatterbox Multilingual | Ship pick - voice clone | MIT (commercial OK) | github.com/resemble-ai/chatterbox |
| Kokoro-82M | Default library voice | Apache-2.0 | huggingface.co/hexgrad/Kokoro-82M |
| Piper | CPU fallback voice | MIT | github.com/rhasspy/piper |
| F5-TTS | Prototype only | CC-BY-NC (non-commercial) | github.com/SWivid/F5-TTS |
| XTTS-v2 | Benchmark only | Coqui PML (non-commercial) | github.com/coqui-ai/TTS |
| Whisper | Speech-to-text (planned) | MIT | github.com/openai/whisper |
Audio - music & SFX
| Model | Role | Weights license | Repo |
|---|---|---|---|
| ACE-Step 1.5 | Music / song-from-lyrics (default) | Apache-2.0 | github.com/ace-step/ACE-Step |
| YuE | Premium lyrics-to-song (A100) | Apache-2.0 (verify variant) | github.com/multimodal-art-projection/YuE |
| Stable Audio Open 1.0 | Text-to-audio / SFX | Stability Community License (free under $1M rev; must attribute) | github.com/Stability-AI/stable-audio-tools |
Image, sprite & video
| Model | Role | Weights license | Repo |
|---|---|---|---|
| SDXL 1.0 base | Base image / sprite generator | OpenRAIL++ (commercial OK) | huggingface.co/stabilityai/stable-diffusion-xl-base-1.0 |
| pixel-art-xl LoRA | Pixel-art style (sprite lane) | See model card | huggingface.co/nerijs/pixel-art-xl |
| pixeldetector (kCentroid) | Crisp pixel downscale + palette | MIT | github.com/Astropulse/pixeldetector |
| BiRefNet | Background removal (bg-remove op) | MIT | github.com/ZhengPeng7/BiRefNet |
| Wan-Alpha / Wan 2.2 | Transparent T2V / I2V | Apache-2.0 (verify) | github.com/Wan-Video/Wan2.2 |
| LayerDiffuse | Transparent RGBA image | Apache-2.0 (code) | github.com/lllyasviel/LayerDiffuse |
Talking-head / lip-sync
| Model | Role | License | Repo |
|---|---|---|---|
| SadTalker | Audio-driven head motion (512) | Apache-2.0 | github.com/OpenTalker/SadTalker |
| MuseTalk v1.5 | Lip resync (sharp mouth) | MIT | github.com/TMElyralab/MuseTalk |
| LivePortrait | Optional expression / motion re-drive | MIT | github.com/KwaiVGI/LivePortrait |
| GFPGAN / Real-ESRGAN | Face enhance + upscale | BSD | github.com/TencentARC/GFPGAN |
| Wav2Lip | NOT shipped - non-commercial weights | Non-commercial (LRS2) | github.com/Rudrabha/Wav2Lip |
License notes are from the platform R&D whitepapers; confirm the current weights license on each repo before building a lane.