Llama 3.2 Vision

Run Llama 3.2 Vision via OpenRouter.

v2

Ask a question to Llama 3.2 Vision model.

You can specify arbitrary text prompts or predefined ones, the block supports the following types of prompt:

• Open Prompt (unconstrained) - Use any prompt to generate a raw response

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

• Visual Question Answering (visual-question-answering) - Model answers the question you submit in the prompt

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

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

• Single-Label Classification (classification) - Model classifies the image content as one of the provided classes

• Multi-Label Classification (multi-label-classification) - Model classifies the image content as one or more of the provided classes

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

• Structured Output Generation (structured-answering) - Model returns a JSON response with the specified fields

🛠️ API providers and model variants

Llama 3.2 Vision is exposed via OpenRouter. By default this block uses the Roboflow-managed OpenRouter key and bills your Roboflow credits - no extra setup needed. To bypass Roboflow billing, paste your own sk-or-... key into the api_key field.

The privacy_level field controls which OpenRouter providers may serve the request:

• No data collection (default) – providers may not train on your inputs.

• Allow data collection – broader provider pool.

• Zero data retention – strictest, restricts to providers that retain nothing.

💡 Further reading and Acceptable Use Policy

Model license

Check the Llama 3.2 license before use.

Type identifier

Use the following identifier in step "type" field: roboflow_core/llama_vision@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..	❌
api_key	str	OpenRouter API key. Defaults to Roboflow's managed key, billed in credits via Roboflow. Provide your own sk-or-... key to call OpenRouter directly without Roboflow billing..	✅
privacy_level	str	Provider privacy filter. Stricter levels reduce the pool of providers and may increase per-call cost on the managed key..	❌
max_tokens	int	Maximum number of tokens the model can generate in its response..	❌
temperature	float	Temperature to sample from the model - value in range 0.0-2.0, the higher - the more random / "creative" the generations are..	✅
max_concurrent_requests	int	Number of concurrent requests for batches of images. If not given - block defaults to value configured globally in Workflows Execution Engine. Restrict if you hit rate limits..	❌
task_type	str	Task type to be performed by model. Value determines required parameters and output response..	❌
prompt	str	Text prompt to the Llama model.	✅
output_structure	Dict[str, str]	Dictionary with structure of expected JSON response.	❌
classes	List[str]	List of classes to be used.	✅
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

requires_internet - air-gapped / offline deployments : This block depends on a service that is not reachable from fully offline / air-gapped deployments.

Input and Output Bindings

The available connections depend on its binding kinds. Check what binding kinds Llama 3.2 Vision in version v2 has.

Input and output bindings

• input

  • api_key (Union[ROBOFLOW_MANAGED_KEY, secret, string]): OpenRouter API key. Defaults to Roboflow's managed key, billed in credits via Roboflow. Provide your own sk-or-... key to call OpenRouter directly without Roboflow billing..

  • temperature (float): Temperature to sample from the model - value in range 0.0-2.0, the higher - the more random / "creative" the generations are..

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

  • prompt (string): Text prompt to the Llama model.

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

  • model_version (string): Model to be used.

• output

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

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

Example JSON definition

{
	    "name": "<your_step_name_here>",
	    "type": "roboflow_core/llama_vision@v2",
	    "api_key": "rf_key:account",
	    "privacy_level": "<block_does_not_provide_example>",
	    "max_tokens": "<block_does_not_provide_example>",
	    "temperature": "<block_does_not_provide_example>",
	    "max_concurrent_requests": "<block_does_not_provide_example>",
	    "images": "$inputs.image",
	    "task_type": "<block_does_not_provide_example>",
	    "prompt": "my prompt",
	    "output_structure": {
	        "my_key": "description"
	    },
	    "classes": [
	        "class-a",
	        "class-b"
	    ],
	    "model_version": "11B - OpenRouter"
	}

v1

Ask a question to Llama 3.2 Vision model with vision capabilities.

You can specify arbitrary text prompts or predefined ones, the block supports the following types of prompt:

• Open Prompt (unconstrained) - Use any prompt to generate a raw response

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

• Visual Question Answering (visual-question-answering) - Model answers the question you submit in the prompt

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

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

• Single-Label Classification (classification) - Model classifies the image content as one of the provided classes

• Multi-Label Classification (multi-label-classification) - Model classifies the image content as one or more of the provided classes

• Structured Output Generation (structured-answering) - Model returns a JSON response with the specified fields

Issues with structured prompting

Model tends to be quite unpredictable when structured output (in our case JSON document) is expected. That problems may impact tasks like structured-answering, classification or multi-label-classification.

The cause seems to be quite sensitive "filters" of inappropriate content embedded in model.

🛠️ API providers and model variants

Llama Vision 3.2 model is exposed via OpenRouter API and we require passing OpenRouter API Key to run.

There are different versions of the model supported:

• smaller version (11B) is faster and cheaper, yet you can expect better quality of results using 90B version

• Regular version is paid (and usually faster) API, whereas Free is free for use for OpenRouter clients (state at 01.01.2025)

As for now, OpenRouter is the only provider for Llama 3.2 Vision model, but we will keep you posted if the state of the matter changes.

API Usage Charges

OpenRouter is external third party providing access to the model and incurring charges on the usage. Please check out pricing before use:

• Llama 3.2 Vision 11B (Regular)

• Llama 3.2 Vision 90B (Regular)

💡 Further reading and Acceptable Use Policy

Model license

Check out model license before use.

Click here for the original model card.

Usage of this model is subject to Meta's Acceptable Use Policy.

Type identifier

Use the following identifier in step "type" field: roboflow_core/llama_3_2_vision@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 Llama model.	✅
output_structure	Dict[str, str]	Dictionary with structure of expected JSON response.	❌
classes	List[str]	List of classes to be used.	✅
api_key	str	Your Llama Vision API key (dependent on provider, ex: OpenRouter API key).	✅
model_version	str	Model to be used.	✅
max_tokens	int	Maximum number of tokens the model can generate in it's response..	❌
temperature	float	Temperature to sample from the model - value in range 0.0-2.0, the higher - the more random / "creative" the generations are..	✅
max_concurrent_requests	int	Number of concurrent requests that can be executed by block when batch of input images provided. If not given - block defaults to value configured globally in Workflows Execution Engine. Please restrict if you hit limits..	❌

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

Runtime compatibility

requires_internet - air-gapped / offline deployments : This block depends on a service that is not reachable from fully offline / air-gapped deployments.

Input and Output Bindings

The available connections depend on its binding kinds. Check what binding kinds Llama 3.2 Vision in version v1 has.

Input and output bindings

• input

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

  • prompt (string): Text prompt to the Llama model.

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

  • api_key (string): Your Llama Vision API key (dependent on provider, ex: OpenRouter API key).

  • model_version (string): Model to be used.

  • temperature (float): Temperature to sample from the model - value in range 0.0-2.0, the higher - the more random / "creative" the generations are..

• output

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

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

Example JSON definition

{
	    "name": "<your_step_name_here>",
	    "type": "roboflow_core/llama_3_2_vision@v1",
	    "images": "$inputs.image",
	    "task_type": "<block_does_not_provide_example>",
	    "prompt": "my prompt",
	    "output_structure": {
	        "my_key": "description"
	    },
	    "classes": [
	        "class-a",
	        "class-b"
	    ],
	    "api_key": "xxx-xxx",
	    "model_version": "11B (Free) - OpenRouter",
	    "max_tokens": "<block_does_not_provide_example>",
	    "temperature": "<block_does_not_provide_example>",
	    "max_concurrent_requests": "<block_does_not_provide_example>"
	}