Objects & Products
Granular identification of specific items, materials, and textures. Recognizes brands, models, and proprietary products with high precision.
Our proprietary computer vision engine processes millions of pixels per second to generate a structured, hierarchical taxonomy for every asset in your library.
The Pipeline
Nexa doesn't just guess; it analyzes. Our multi-stage pipeline breaks down every image into semantic components, ensuring that every tag is contextually accurate and hierarchically sound.
Raw files are normalized, resized, and color-converted to a standard sRGB space. We strip EXIF metadata to protect privacy while preserving visual data. This ensures consistent input for every model in the pipeline.
Supported formats: JPEG, PNG, WebP, TIFF, RAW (CR2, NEF, ARW), HEIC, SVG.
Nexa runs parallel neural networks: a CNN for object detection, a Vision Transformer for semantic understanding, and a color science model for palette extraction. These outputs are fused into a single context vector.
Output: A JSON object containing bounding boxes, tags, colors, and sentiment scores.
Tag Taxonomy
Beyond simple object recognition, Nexa understands context, emotion, and style.
Granular identification of specific items, materials, and textures. Recognizes brands, models, and proprietary products with high precision.
Demographic analysis including age range, gender, and expression. Privacy-safe by default; no biometric data is stored or used for training.
Sentiment analysis of facial expressions and overall scene atmosphere. Tags like "joyful", "melancholic", "energetic", or "serene" are generated automatically.
Extracts dominant and accent colors using Lab color space analysis. Generates CSS-compatible hex codes and named color tags.
Geospatial tagging for street scenes and indoor environments. Recognizes architectural styles, weather conditions, and time of day.
Style classification including photography genre, art movement, and composition techniques (e.g., "Golden Ratio", "Rule of Thirds", "Bokeh").
Precision Control
Control the balance between recall and precision with confidence thresholds, and train the model on your specific brand assets.
Every tag comes with a confidence score (0.0 to 1.0). Use the threshold slider to filter out low-confidence noise. Set a high threshold for strict cataloging, or a low one to capture every potential match.
Upload a CSV or JSON file of your specific product SKUs, brand names, or campaign terms. Nexa will retrain the model on your dataset within minutes, ensuring your unique terminology is recognized instantly.
Enterprise only. Requires a minimum of 50 labeled examples per custom class.
Performance
Nexa outperforms manual tagging and general-purpose competitors in both speed and accuracy.
| Metric | Manual | Nexa AI | Competitor X |
|---|---|---|---|
| Tagging Accuracy | ~65% | 98.2% | 91.5% |
| Processing Speed | ~4 hrs / 1k | 0.3s / img | 1.2s / img |
| Contextual Understanding | Low | High | Medium |
| Custom Vocabulary | Manual Entry | Auto-Training | Manual Entry |
Developers
A RESTful API designed for speed and reliability. Get structured JSON responses directly from your code.
{
"image_url": "https://cdn.nexa.ai/assets/2023/summer-campaign.jpg",
"threshold": 0.85,
"return_bounding_boxes": true
}
Response (200 OK)
{
"id": "tag_8f92a",
"status": "completed",
"processing_time_ms": 312,
"tags": [
{ "name": "product", "confidence": 0.98, "category": "object" },
{ "name": "summer", "confidence": 0.94, "category": "season" },
{ "name": "golden_hour", "confidence": 0.91, "category": "lighting" }
]
}
See the difference precision makes. Upload your first batch and get instant, structured insights.