Multi-Image Analysis with GPT-4 Vision

Comparing Multiple Images

def compare_images(images: list[str], comparison_prompt: str) -> str:
    """Compare multiple images."""

    content = [{"type": "text", "text": comparison_prompt}]

    for i, img_path in enumerate(images):
        content.append({
            "type": "text",
            "text": f"Image {i+1}:"
        })
        content.append({
            "type": "image_url",
            "image_url": {"url": f"data:image/png;base64,{encode_image(img_path)}"}
        })

    response = client.chat.completions.create(
        model="gpt-4-vision-preview",
        messages=[{"role": "user", "content": content}],
        max_tokens=2000
    )

    return response.choices[0].message.content

# Before/After comparison
result = compare_images(
    ["before.jpg", "after.jpg"],
    "Compare these images and describe all differences you observe."
)

Temporal Sequence Analysis

def analyze_sequence(image_paths: list[str], context: str) -> dict:
    """Analyze a sequence of images over time."""

    prompt = f"""Analyze this sequence of {len(image_paths)} images.
    Context: {context}

    Describe:
    1. What's happening in each image
    2. The progression/changes between images
    3. Any trends or patterns
    4. Conclusions you can draw

    Return as JSON."""

    content = [{"type": "text", "text": prompt}]
    for path in image_paths:
        content.append({
            "type": "image_url",
            "image_url": {"url": f"data:image/png;base64,{encode_image(path)}"}
        })

    response = client.chat.completions.create(
        model="gpt-4-vision-preview",
        messages=[{"role": "user", "content": content}],
        response_format={"type": "json_object"},
        max_tokens=3000
    )

    return json.loads(response.choices[0].message.content)

Multi-Angle Product Analysis

def analyze_product_images(images: list[str]) -> dict:
    """Analyze product from multiple angles."""

    prompt = """Analyze this product from multiple angles and provide:
    1. Product description
    2. Key features visible
    3. Condition assessment
    4. Any defects or issues
    5. Overall quality rating (1-10)

    Return as structured JSON."""

    content = [{"type": "text", "text": prompt}]
    for img in images:
        content.append({
            "type": "image_url",
            "image_url": {"url": f"data:image/png;base64,{encode_image(img)}"}
        })

    response = client.chat.completions.create(
        model="gpt-4-vision-preview",
        messages=[{"role": "user", "content": content}],
        response_format={"type": "json_object"}
    )

    return json.loads(response.choices[0].message.content)

Best Practices

1. Order matters - Present images in logical order
2. Label images - Help model distinguish between them
3. Limit count - 4-6 images optimal for most tasks
4. Use consistent sizing - Resize for uniformity
5. Provide context - Explain what you’re comparing

Conclusion

Multi-image analysis unlocks powerful comparison and temporal understanding capabilities. Structure your prompts to guide the model through the images effectively.