ComfyUI Node
Load Embedding Model
A ComfyUI node in Eric/SemanticSearch with 5 inputs and 1 output.
Load Embedding Model
- model
◄model_nameQwen3-VL-Embedding-8B►
◄deviceauto►
◄max_resolution1024x1024 (1MP)►
◄attention_typesdpa►
◄embedding_dimFull (4096/2048)►
CategoryEric/SemanticSearch
Inputs (5)
| Name | Type | Default | Description |
|---|---|---|---|
| model_name | COMBO | Qwen3-VL-Embedding-8B | 2 options: Qwen3-VL-Embedding-2B, Qwen3-VL-Embedding-8B |
| device | COMBO | auto | 5 options: auto, cuda, cuda:0, cuda:1, cpu |
| max_resolutionopt | COMBO | 1024x1024 (1MP) | Maximum image resolution for encoding. Higher = more accurate but slower. |
| attention_typeopt | COMBO | sdpa | Attention implementation. sdpa=default, eager=fallback, sage=SageAttention (if installed) |
| embedding_dimopt | COMBO | Full (4096/2048) | Embedding dimension. Full=native (4096 for 8B, 2048 for 2B). Reduced dims use Matryoshka representation for smaller/faster indexes. |
Outputs (1)
| Name | Type | Description |
|---|---|---|
| model | EMBEDDING_MODEL | — |