ByteDanceDoubao Embedding
Doubao Embedding Vision
Embedding Vision represents mixed text, image and video inputs. It supports dense vectors and multiple vectors for one input.
Technical guide
The guide lets clients set retrieval instructions and control video frame sampling. Multiple-vector output was introduced in the 251215 release. BytePlus lists the related service under the Skylark name.Doubao multimodal embedding guideBytePlus ModelArk current model catalog
Practical use
Practical guidance. Build an evaluation set with real search queries and difficult near matches. Keep the same instruction and frame policy when creating the stored index and evaluating new retrieval behavior.
Specifications & access
| Detail | Value and conditions |
|---|---|
| Access | API |
| Model identifiers | doubao-embedding-vision-251215skylark-embedding-vision-251215 |
| Other output | Sparse embeddings are supported for text inputs.Doubao multimodal embedding guide |
Related models
- Seed 2.1 Lite
ByteDance ยท Seed 2.1
Multimodal Seed 2.1 model with a separate System One interface for fixed-choice probabilities and scores.
Launch date not confirmedAPI