Skip to main content
Get up and running with Qwen3-VL in just a few steps. This guide will walk you through your first image inference.

​
Installation

1

Install Transformers

Qwen3-VL requires transformers 4.57.0 or higher:
pip install "transformers>=4.57.0"
2

Install Optional Dependencies

For optimal performance, install these recommended packages:
pip install accelerate torch torchvision

​
Your First Inference

Here’s a complete example to perform image understanding with Qwen3-VL: 
from transformers import AutoModelForImageTextToText, AutoProcessor

# Load the model on available devices
model = AutoModelForImageTextToText.from_pretrained(
    "Qwen/Qwen3-VL-8B-Instruct", 
    dtype="auto", 
    device_map="auto"
)

processor = AutoProcessor.from_pretrained("Qwen/Qwen3-VL-8B-Instruct")

# Prepare your message with an image
messages = [
    {
        "role": "user",
        "content": [
            {
                "type": "image",
                "image": "https://qianwen-res.oss-cn-beijing.aliyuncs.com/Qwen-VL/assets/demo.jpeg",
            },
            {"type": "text", "text": "Describe this image."},
        ],
    }
]

# Prepare inputs for inference
inputs = processor.apply_chat_template(
    messages,
    tokenize=True,
    add_generation_prompt=True,
    return_dict=True,
    return_tensors="pt"
)
inputs = inputs.to(model.device)

# Generate the output
generated_ids = model.generate(**inputs, max_new_tokens=128)
generated_ids_trimmed = [
    out_ids[len(in_ids):] for in_ids, out_ids in zip(inputs.input_ids, generated_ids)
]
output_text = processor.batch_decode(
    generated_ids_trimmed, 
    skip_special_tokens=True, 
    clean_up_tokenization_spaces=False
)
print(output_text)
We recommend starting with the 8B model for development. For production, see the Model Variants guide.

​
Image Input Formats

Qwen3-VL supports multiple image input formats: 
{
    "type": "image",
    "image": "https://example.com/image.jpg"
}

​
Model Selection

Choose a model size based on your use case:

2B / 4B Models

Edge deployment - Run on consumer GPUs or mobile devices

8B Model

Balanced - Best for most applications (24GB VRAM)

32B Model

High performance - Production use cases (80GB VRAM)

235B MoE Model

Maximum capability - Research and specialized tasks (8x80GB)

​
Performance Optimization

For better performance, enable Flash Attention 2: 
pip install flash-attn --no-build-isolation
Then load the model with: 
import torch
from transformers import AutoModelForImageTextToText

model = AutoModelForImageTextToText.from_pretrained(
    "Qwen/Qwen3-VL-8B-Instruct",
    torch_dtype=torch.bfloat16,
    attn_implementation="flash_attention_2",
    device_map="auto"
)
Flash Attention 2 is especially beneficial for multi-image and video scenarios, reducing memory usage and improving speed.

​
Next Steps

Image Processing

Learn about multi-image inference and resolution control

Video Processing

Process videos with frame sampling

Capabilities

Explore OCR, grounding, document parsing, and more

Deployment

Deploy with vLLM or SGLang for production