Florence-2 Model

Run Florence-2 on an image

v2

Dedicated inference server required (GPU recommended) - you may want to use dedicated deployment

This Workflow block introduces Florence 2, a Visual Language Model (VLM) capable of performing a wide range of tasks, including:

• Object Detection

• Instance Segmentation

• Image Captioning

• Optical Character Recognition (OCR)

• and more...

Below is a comprehensive list of tasks supported by the model, along with descriptions on how to utilize their outputs within the Workflows ecosystem:

Task Descriptions:

• Custom Prompt (custom) - Use free-form prompt to generate a response. Useful with finetuned models.

• Text Recognition (OCR) (ocr) - Model recognizes text in the image

• Text Detection & Recognition (OCR) (ocr-with-text-detection) - Model detects text regions in the image, and then performs OCR on each detected region

• Captioning (short) (caption) - Model provides a short description of the image

• Captioning (detailed-caption) - Model provides a long description of the image

• Captioning (long) (more-detailed-caption) - Model provides a very long description of the image

• Unprompted Object Detection (object-detection) - Model detects and returns the bounding boxes for prominent objects in the image

• Object Detection (open-vocabulary-object-detection) - Model detects and returns the bounding boxes for the provided classes

• Detection & Captioning (object-detection-and-caption) - Model detects prominent objects and captions them

• Prompted Object Detection (phrase-grounded-object-detection) - Based on the textual prompt, model detects objects matching the descriptions

• Prompted Instance Segmentation (phrase-grounded-instance-segmentation) - Based on the textual prompt, model segments objects matching the descriptions

• Segment Bounding Box (detection-grounded-instance-segmentation) - Model segments the object in the provided bounding box into a polygon

• Classification of Bounding Box (detection-grounded-classification) - Model classifies the object inside the provided bounding box

• Captioning of Bounding Box (detection-grounded-caption) - Model captions the object in the provided bounding box

• Text Recognition (OCR) for Bounding Box (detection-grounded-ocr) - Model performs OCR on the text inside the provided bounding box

• Regions of Interest proposal (region-proposal) - Model proposes Regions of Interest (Bounding Boxes) in the image

Type identifier

Use the following identifier in step "type" field: roboflow_core/florence_2@v2 to add the block as a step in your workflow.

Properties

Name	Type	Description	Refs
name	str	Enter a unique identifier for this step..	❌
task_type	str	Task type to be performed by model. Value determines required parameters and output response..	❌
prompt	str	Text prompt to the Florence-2 model.	✅
classes	List[str]	List of classes to be used.	✅
grounding_detection	Optional[List[float], List[int]]	Detection to ground Florence-2 model. May be statically provided bounding box [left_top_x, left_top_y, right_bottom_x, right_bottom_y] or result of object-detection model. If the latter is true, one box will be selected based on grounding_selection_mode..	✅
grounding_selection_mode	str	.	❌
model_id	str	Model to be used.	✅

The Refs column marks possibility to parametrise the property with dynamic values available in workflow runtime. See Bindings for more info.

Runtime compatibility

hard - runtime self_hosted_cpu; execution local : Requires a GPU; run_locally() loads a model that needs CUDA.

Input and Output Bindings

The available connections depend on its binding kinds. Check what binding kinds Florence-2 Model in version v2 has.

Input and output bindings

• input

  • images (image): The image to infer on..

  • prompt (string): Text prompt to the Florence-2 model.

  • classes (list_of_values): List of classes to be used.

  • grounding_detection (Union[instance_segmentation_prediction, keypoint_detection_prediction, list_of_values, object_detection_prediction]): Detection to ground Florence-2 model. May be statically provided bounding box [left_top_x, left_top_y, right_bottom_x, right_bottom_y] or result of object-detection model. If the latter is true, one box will be selected based on grounding_selection_mode..

  • model_id (roboflow_model_id): Model to be used.

• output

  • raw_output (Union[string, language_model_output]): String value if string or LLM / VLM output if language_model_output.

  • parsed_output (dictionary): Dictionary.

  • classes (list_of_values): List of values of any type.

Example JSON definition

{
	    "name": "<your_step_name_here>",
	    "type": "roboflow_core/florence_2@v2",
	    "images": "$inputs.image",
	    "task_type": "<block_does_not_provide_example>",
	    "prompt": "my prompt",
	    "classes": [
	        "class-a",
	        "class-b"
	    ],
	    "grounding_detection": "$steps.detection.predictions",
	    "grounding_selection_mode": "first",
	    "model_id": "florence-2-base"
	}

v1

Dedicated inference server required (GPU recommended) - you may want to use dedicated deployment

This Workflow block introduces Florence 2, a Visual Language Model (VLM) capable of performing a wide range of tasks, including:

• Object Detection

• Instance Segmentation

• Image Captioning

• Optical Character Recognition (OCR)

• and more...

Below is a comprehensive list of tasks supported by the model, along with descriptions on how to utilize their outputs within the Workflows ecosystem:

Task Descriptions:

• Custom Prompt (custom) - Use free-form prompt to generate a response. Useful with finetuned models.

• Text Recognition (OCR) (ocr) - Model recognizes text in the image

• Text Detection & Recognition (OCR) (ocr-with-text-detection) - Model detects text regions in the image, and then performs OCR on each detected region

• Captioning (short) (caption) - Model provides a short description of the image

• Captioning (detailed-caption) - Model provides a long description of the image

• Captioning (long) (more-detailed-caption) - Model provides a very long description of the image

• Unprompted Object Detection (object-detection) - Model detects and returns the bounding boxes for prominent objects in the image

• Object Detection (open-vocabulary-object-detection) - Model detects and returns the bounding boxes for the provided classes

• Detection & Captioning (object-detection-and-caption) - Model detects prominent objects and captions them

• Prompted Object Detection (phrase-grounded-object-detection) - Based on the textual prompt, model detects objects matching the descriptions

• Prompted Instance Segmentation (phrase-grounded-instance-segmentation) - Based on the textual prompt, model segments objects matching the descriptions

• Segment Bounding Box (detection-grounded-instance-segmentation) - Model segments the object in the provided bounding box into a polygon

• Classification of Bounding Box (detection-grounded-classification) - Model classifies the object inside the provided bounding box

• Captioning of Bounding Box (detection-grounded-caption) - Model captions the object in the provided bounding box

• Text Recognition (OCR) for Bounding Box (detection-grounded-ocr) - Model performs OCR on the text inside the provided bounding box

• Regions of Interest proposal (region-proposal) - Model proposes Regions of Interest (Bounding Boxes) in the image

Type identifier

Use the following identifier in step "type" field: roboflow_core/florence_2@v1 to add the block as a step in your workflow.

Properties

Name	Type	Description	Refs
name	str	Enter a unique identifier for this step..	❌
task_type	str	Task type to be performed by model. Value determines required parameters and output response..	❌
prompt	str	Text prompt to the Florence-2 model.	✅
classes	List[str]	List of classes to be used.	✅
grounding_detection	Optional[List[float], List[int]]	Detection to ground Florence-2 model. May be statically provided bounding box [left_top_x, left_top_y, right_bottom_x, right_bottom_y] or result of object-detection model. If the latter is true, one box will be selected based on grounding_selection_mode..	✅
grounding_selection_mode	str	.	❌
model_version	str	Model to be used.	✅

The Refs column marks possibility to parametrise the property with dynamic values available in workflow runtime. See Bindings for more info.

Runtime compatibility

hard - runtime self_hosted_cpu; execution local : Requires a GPU; run_locally() loads a model that needs CUDA.

Input and Output Bindings

The available connections depend on its binding kinds. Check what binding kinds Florence-2 Model in version v1 has.

Input and output bindings

• input

  • images (image): The image to infer on..

  • prompt (string): Text prompt to the Florence-2 model.

  • classes (list_of_values): List of classes to be used.

  • grounding_detection (Union[instance_segmentation_prediction, keypoint_detection_prediction, list_of_values, object_detection_prediction]): Detection to ground Florence-2 model. May be statically provided bounding box [left_top_x, left_top_y, right_bottom_x, right_bottom_y] or result of object-detection model. If the latter is true, one box will be selected based on grounding_selection_mode..

  • model_version (string): Model to be used.

• output

  • raw_output (Union[string, language_model_output]): String value if string or LLM / VLM output if language_model_output.

  • parsed_output (dictionary): Dictionary.

  • classes (list_of_values): List of values of any type.

Example JSON definition

{
	    "name": "<your_step_name_here>",
	    "type": "roboflow_core/florence_2@v1",
	    "images": "$inputs.image",
	    "task_type": "<block_does_not_provide_example>",
	    "prompt": "my prompt",
	    "classes": [
	        "class-a",
	        "class-b"
	    ],
	    "grounding_detection": "$steps.detection.predictions",
	    "grounding_selection_mode": "first",
	    "model_version": "florence-2-base"
	}