ComfyUI Node
Xiaomi-Lens-0 Segment
Finds what the text describes and returns a mask and a box per instance (greedy decoding, as upstream's infer.py). find: the query is put into the README's grounding prompt. instruction: the text is sent as typed (for questions such as 'Which object would you use to stir the drink? Report bbox coordinates and masks.').
Xiaomi-Lens-0 Segment
- model
- image
- mask
- instance_masks
- preview
- text
- boxes
◄querythe cat►
◄prompt_modefind►
◄threshold0.50►
◄max_new_tokens16384►
Categorymask/xiaomi_lens
Inputs (6)
| Name | Type | Default | Description |
|---|---|---|---|
| model | XIAOMI_LENS_MODEL | — | |
| image | IMAGE | — | |
| query | STRING | the cat | — |
| prompt_mode | COMBO | find | find: Locate all instances in the image that match the description "<query>". Report bbox coordinates and masks. instruction: the query as typed |
| threshold | FLOAT | 0.500–1 | mask probability threshold (upstream: 0.5). 0: soft masks (the probabilities) |
| max_new_tokens | INT | 1638416–16384 | generation limit (upstream default 16384); a reasoning answer can be long |
Outputs (5)
| Name | Type | Description |
|---|---|---|
| mask | MASK | — |
| instance_masks | MASK | — |
| preview | IMAGE | — |
| text | STRING | — |
| boxes | STRING | — |