Skip to main content
With Pawa AI’s vision capabilities, your applications can go beyond text-only understanding and incorporate visual information into reasoning, analysis, and decision-making.
Pawa AI Vision Example
Vision Example For more specific use cases related to document parsing and data extraction, we recommend exploring our Document Parsing documentation here.

​
Vision Based Models in Pawa AI.

Currently we have one vision model:
  • pawa-v1-blaze-20250318: A powerful small language model (SLM) optimized for reasoning, complex generation, multimodal, tools understanding, agentic workflow, and advanced knowledge tasks.

​
How to send multimodal chat request with Pawa AI ?

curl https://api.pawa-ai.com/v1/chat/request \
  -H "Content-Type: application/json" \
  -H "Authorization: Bearer $PAWA_AI_API_KEY" \
  -d '{
    "model": "pawa-v1-blaze-20250318",
    "messages": [
      {
        "role": "user",
        "content": [
          { "type": "text", "text": "Describe the main objects in this image and their colors." },
          { "type": "image_url", "image_url": { "url": "https://example.com/path/to/image.jpg" } }
        ]
      }
    ]
  }'

curl https://api.pawa-ai.com/v1/chat/request \
  -H "Content-Type: application/json" \
  -H "Authorization: Bearer $PAWA_AI_API_KEY" \
  -d '{
    "model": "pawa-v1-blaze-20250318",
    "messages": [
      {
        "role": "user",
        "content": [
          { "type": "text", "text": "What is the difference of these images?" },
          { "type": "image_url", "image_url": { "url": "https://example.com/path/to/image.jpg" } }
          { "type": "image_url", "image_url": { "url": "https://example.com/path/to/image.jpg" } }
          { "type": "image_url", "image_url": { "url": "https://example.com/path/to/image.jpg" } }
        ]
      }
    ]
  }'

​
Use Cases

Classify the overall content of an image into one or more categories.
   curl https://api.pawa-ai.com/v1/chat/request \
  -H "Content-Type: application/json" \
  -H "Authorization: Bearer $PAWA_AI_API_KEY" \
  -d '{
    "model": "pawa-v1-blaze-20250318",
    "messages": [
      {
        "role": "user",
        "content": [
          { "type": "text", "text": "Classify this image by scene and mood." },
          { "type": "image_url", "image_url": { "url": "https://example.com/path/to/image.jpg" } }
        ]
      }
    ]
  }'
Example output:
{
  "labels": ["outdoor", "campfire", "night"],
  "confidence": {"outdoor": 0.97, "campfire": 0.94, "night": 0.90},
  "summary": "A night outdoor scene with a campfire and people nearby."
}
Detect and locate multiple objects in the image with rough descriptions.
curl https://api.pawa-ai.com/v1/chat/request \
  -H "Content-Type: application/json" \
  -H "Authorization: Bearer $PAWA_AI_API_KEY" \
  -d '{
    "model": "pawa-v1-blaze-20250318",
    "messages": [
      {
        "role": "user",
        "content": [
          { "type": "text", "text": "List the objects you see and their approximate positions." },
          { "type": "image_url", "image_url": { "url": "https://example.com/path/to/store-shelf.jpg" } }
        ]
      }
    ]
  }'
Example output:
{
  "objects": [
    {"label": "laptop", "bbox": [120, 80, 460, 320]},
    {"label": "coffee-cup", "bbox": [480, 260, 560, 340]}
  ],
  "summary": "A laptop on a desk with a coffee cup on the right."
}
Extract text or structured fields from receipts, invoices, or forms. For advanced usage you can combine with response_format
curl https://api.pawa-ai.com/v1/chat/request \
  -H "Content-Type: application/json" \
  -H "Authorization: Bearer $PAWA_AI_API_KEY" \
  -d '{
    "model": "pawa-v1-blaze-20250318",
    "messages": [
      {
        "role": "user",
        "content": [
          { "type": "text", "text": "Extract the total, date and vendor name from this receipt." },
          { "type": "image_url", "image_url": { "url": "https://example.com/path/to/receipt.jpg" } }
        ]
      }
    ]
  }'
Example output:
{
  "vendor": "Moshi Supermarket",
  "date": "2025-03-18",
  "total": 16450,
  "currency": "TZS"
}

​
Best practices

  • Resize large images to a reasonable resolution (e.g., 1024px max on the longest side) to reduce latency without losing salient details.
  • Prefer image_url for repeatable, production traffic. Use signed URLs that expire quickly if assets are private.
  • Provide clear textual instructions that set expectations (“list objects with colors and approximate positions”).
  • For OCR, favor sharp, high‑contrast images. If possible, crop to the relevant region before sending.
  • When sending multiple images, order them logically and reference them in the text (“in the second photo, compare…”).

​
Supported formats and limits

  • Formats: jpg/jpeg, png
  • Max image size: 25 MiB per image
  • Number of images per request: no hard limit; practical limits depend on latency and payload size
  • Base64 payloads should include the MIME type in the content object

​
Privacy & security

  • Assets sent via image_url are fetched server‑side only for the duration of the request. Use time‑limited signed URLs for private content.
  • Inputs may be logged for abuse monitoring and quality unless you enable data controls in your dashboard.

​
Troubleshooting

  • Image not found: Ensure the image_url is publicly reachable or signed correctly; test with curl -I <url>.
  • Payload too large: Downscale images or switch to image_url.
  • Poor OCR quality: Increase resolution, improve lighting/contrast, or crop tighter around text.
  • Slow responses: Host images on a nearby region/CDN and reduce image dimensions.
⌘I