KyutaiMoshi
MoshiVis
MoshiVis connects the Moshika spoken-dialogue model to image features. It lets a speech conversation refer to visual input while retaining the underlying model’s audio interaction design.
Research release
Only the connection layers need this training
The release combines roughly 7B Moshi parameters, a roughly 400M vision encoder and about 200M trained connection parameters. The visual adaptation does not establish a general tool-using assistant. PyTorch, Candle and MLX packages share the model identity.moshika-vis-pytorch-bf16: Kyutai model card
Specifications & access
| Detail | Value and conditions |
|---|---|
| Status | Research release |
| Access | Downloadable |
| Model identifiers | kyutai/moshika-vis-pytorch-bf16kyutai/moshika-vis-candle-bf16kyutai/moshika-vis-candle-q8kyutai/moshika-vis-mlx |
| Architecture | Frozen Moshi and SigLIP vision components with trained cross-attention layers.moshika-vis-pytorch-bf16: Kyutai model card |
| Terms | CC BY 4.0; Gemma Terms of Use apply through the PaliGemma 2 vision component.moshika-vis-pytorch-bf16: Kyutai model card |
Related models
- Moshi Moshika
Kyutai · Moshi
A Moshi speech-dialogue model that listens while speaking, using Kyutai's Moshika voice.
Launch date not confirmedResearch releaseDownloadable