Japanese Stable VLM
Japanese Stable VLM generates Japanese descriptions and answers from an image, with optional text instructions. It supports captioning, tag-guided descriptions and visual questions.
Model design and access
Training first adapts an MLP connector, then updates the connector and language model. The card lists translated CC12M captions, STAIR captions and Japanese Visual Genome questions. The model produces text from images; it is not an image generator.Japanese Stable VLM official model card
The language component has earlier Mistral ancestry. Japanese Stable LM Instruct Gamma 7B is based on Mistral-7B-v0.1, with Japanese instruction tuning by Stability AI Japan. The VLM adds its own visual connector and multimodal training to that language component.Japanese Stable LM Instruct Gamma 7B: Mistral ancestry and Japanese training
Practical use
Practical guidance. This release can support Japanese image-description research without a translation stage after generation. Review concrete visual claims against the image, especially small objects and text. A caption that sounds natural can contain inferred details, so keep descriptive output separate from verified extraction.
Specifications & access
| Detail | Value and conditions |
|---|---|
| Access | Downloadable |
| Architecture | LLaVA 1.5.Japanese Stable VLM official model card |
| Language component | Japanese Stable LM Instruct Gamma 7B.Japanese Stable VLM official model card |
| Vision component | OpenAI CLIP ViT-L/14.Japanese Stable VLM official model card |
| License | Stability AI Community License.Japanese Stable VLM official model card |
Related models
- Japanese Stable CLIP ViT-L/16
Stability AI · Japanese Stable CLIP
Japanese Stable CLIP maps images and Japanese text into a shared embedding space.
Launch date not confirmedDownloadable - Japanese Stable LM 2 Instruct 1.6B
Stability AI · Stable LM
Japanese Stable LM 2 Instruct 1.6B adapts the small Stable LM 2 family for Japanese instructions.
Launch date not confirmedDownloadable