ComfyUI Node Runs on cloud
Paligemma
A ComfyUI node in VLM Nodes/Paligemma with 16 inputs and 3 outputs.
Paligemma
- image
- description
- mask
- visualization
◄model_idgokaygokay/sd3-long-captioner-v2►
◄custom_model_id►
◄task_typeCaptioning►
◄promptDescribe this image in detail.►
◄precisionbfloat16►
◄devicecuda►
◄quantizationNone►
◄max_tokens256►
◄min_tokens10►
◄temperature0.7►
◄fill_maskTrue►
◄mask_colorwhite►
◄mask_opacity0.5►
◄mask_threshold0.5►
◄mask_blur0►
CategoryVLM Nodes/Paligemma
Inputs (16)
| Name | Type | Default | Description |
|---|---|---|---|
| image | IMAGE | Input image to process with Paligemma | |
| model_id | COMBO | gokaygokay/sd3-long-captioner-v2 | Select the Paligemma model to use |
| custom_model_id | STRING | Enter Hugging Face model repo (e.g., 'gokaygokay/sd3-long-captioner-v2') or full URL (e.g., 'https://huggingface.co/gokaygokay/sd3-long-captioner-v2') | |
| task_type | COMBO | Captioning | Select task type: - Captioning: Generates detailed image descriptions - Segmentation: Creates masks for specified objects/regions (use ft-refcoco-seg model) - Question Answering: Answers questions about the image |
| prompt | STRING | Describe this image in detail. | Input text based on task type: - For Captioning: Describe what aspects to focus on - For Segmentation: Describe what to segment (e.g., 'segment the dog', 'find the red car') - For Q&A: Enter your question about the image |
| precisionopt | COMBO | bfloat16 | Model precision - affects VRAM usage and processing speed |
| deviceopt | COMBO | cuda | Device to run model on. 'cuda' for best performance if available |
| quantizationopt | COMBO | None | Model quantization for reduced VRAM usage |
| max_tokensopt | INT | 2561–1024 | [Captioning/Q&A] Maximum number of tokens to generate |
| min_tokensopt | INT | 101–512 | [Captioning/Q&A] Minimum number of tokens to generate |
| temperatureopt | FLOAT | 0.70–2 | [Captioning/Q&A] Temperature for text generation |
| fill_maskopt | COMBO | True | [Segmentation Only] Fill the segmented regions with semi-transparent color overlays |
| mask_coloropt | STRING | white | [Segmentation Only] Color for segmentation mask (e.g., 'white', 'black', '#FF0000') |
| mask_opacityopt | FLOAT | 0.50–1 | [Segmentation Only] Opacity of the segmentation overlay (0.0 = transparent, 1.0 = solid) |
| mask_thresholdopt | FLOAT | 0.50–1 | [Segmentation Only] Threshold for binary mask creation |
| mask_bluropt | INT | 00–64 | [Segmentation Only] Blur radius for mask edges (0 for no blur) |
Outputs (3)
| Name | Type | Description |
|---|---|---|
| description | STRING | — |
| mask | MASK | — |
| visualization | IMAGE | — |