Stable-audio-3-medium support added
Just added support for the medium model.
Note: the medium model only works on NVIDIA GPU
| Model | Best for | Hardware | Max length | Practical choice |
|---|---|---|---|---|
small-music |
Songs, beats, loops, musical ideas | CPU works | 120 sec | Default choice for music |
small-sfx |
Sound effects, hits, whooshes, impacts, ambience | CPU works | 120 sec | Default choice for SFX |
medium |
Higher quality music, longer generations | NVIDIA CUDA only in this launcher | 380 sec | Best quality local option if your GPU supports it |
small-music and small-sfx are specialized models, not quality tiers. Use small-music for musical output and small-sfx for non-musical sound design.
medium is the higher-quality local option, but it is only available in this launcher on supported NVIDIA Windows/Linux systems because it requires CUDA and Flash Attention.
small-music and small-sfx are not really “better/worse” versions of each other. They are specialized: use
small-music for music, use small-sfx for non-musical sound design.
medium is the upgrade path for quality and longer output, but it is less compatible. In this launcher it only
appears on NVIDIA Windows/Linux because it needs CUDA and Flash Attention. It also downloads much more data
and uses more VRAM.
large is not supported locally by this repo. It is API-only, so users will not see it as a Pinokio launch
option here.