Create Your First Recipe (OV80i)

This is where your camera becomes an AI inspector. A recipe is a complete package (image settings, alignment, regions of interest (ROIs), AI model, and output rules) bundled together for one specific inspection task.

You can have as many recipes as you want on a camera. Each one can be saved, backed up, transferred to other cameras, and version controlled.

Before you start: remember the waterfall

Everything in this section follows the Waterfall Principle. You'll go through six steps in order. Don't skip ahead. Verify each step works before moving to the next.

1

Image Settings

Exposure, Gain, Lighting

2

Template & Aligner

Capture, Align

3

Regions of Interest (ROIs)

Draw ROIs

4

AI Training

Label, Train

5

Output Rules

Pass/Fail, IO

6

Deploy!

Activate, Verify

Create a new recipe

1. Go to All Recipes in the left sidebar (this is also the landing page when you open the camera)
2. Click + New
3. Give it a name (e.g., "Surface Defect Inspection")
4. Choose the recipe type: Segmentation vs. Classification (explained below)

What do you want this recipe to do?

Each OV80i recipe uses one model type. Pick based on the question you need answered. You can change this later.

Recommended

Surface check, pass / fail per panel

Classification

Decide whether each part, or each region of interest, belongs to a category.

Defect detectionPass / failAssembly verificationPresence / absenceSortingCosmetic checks

Best for: verdicts and known categories, where the answer is one label per region.

Defect localization, pixel-level masks

Segmentation

Find and outline features, defects, or regions at the pixel level.

CountingContaminant detectionMeasurementPositionDefect localizationCoverage

Best for: locating, measuring, or counting features whose shape and position matter.

Classification, in 60 seconds

Read Understanding Classifier →

A classifier looks at each region you draw and assigns one label from a list you define. The model returns one verdict per ROI: this region is a "pass", or "missing", or "scratched". Most OV80i recipes start here.

Example, missing fastener

Four ROIs over four screw locations. Two classes: present and missing. The model returns one label per ROI.

Example, surface pass / fail

One ROI over the panel surface. Two classes: clean and blemished. The model returns one label for the panel.

Want a side-by-side breakdown? See Classifier vs. Segmenter. Want a deeper walkthrough of either model type? Read Understanding Classifier for verdicts and labels, or Understanding Segmenter for pixel-level masks, counts, and measurements. 5. Click Activate to enter the recipe editor

Multi-model recipes

The OV80i supports multi-model recipes. You can combine classifiers and segmenters in a single recipe for comprehensive inspection. For example, use a classifier to check part presence and a segmenter to detect surface defects, all in one recipe.

Not sure which to choose? See Classifier vs. Segmenter or ask the AI Assistant at tools.overview.ai.

Multi-model flexibility

The OV80i supports multi-model recipes, allowing you to combine classifiers and segmenters in a single recipe for comprehensive inspection. For example, use a classifier to verify part presence and a segmenter to detect surface scratches, all running on the same capture.

Now follow the six steps:

Step 1: Image Settings

Full guide: Image Settings

Get your camera image looking clean and consistent. Adjust exposure, gain, white balance, and, critically, enable lens distortion correction if you're using a wide-angle lens.

Key settings:

• Exposure: How long the sensor captures light. Higher = brighter but more motion blur
• Gain: Digital brightness boost. Higher = brighter but noisier
• Lens Correction: Fixes barrel distortion from wide-angle lenses. Enable this now if applicable. Don't skip it
• External lighting: Verify your lighting is even and consistent before proceeding

Verify before moving on: Click Live Preview. The image should be sharp, well-lit, and consistent shot to shot.

Step 2: Template Image & Alignment

Full guide: Alignment Explained

This is the step most customers find challenging, and the one that makes the biggest difference. The aligner is the foundation of your entire inspection. It dynamically moves your inspection boxes to track the part as it shifts and rotates on the conveyor. Good alignment lets you draw smaller ROIs, which means less training data and more accurate AI. Aligner → ROIs → Classifier/Segmenter: if the first link is weak, everything downstream breaks.

The short version:

1. Capture a template image of a good part
2. Place 2-3 small template regions on features that never change (strong edges, corners, holes)
3. Place them as far apart as possible on the part
4. Clean up noisy edges with the Ignore tool
5. Save, then test with Live Preview. Move the part around and verify the alignment tracks it

The #1 alignment mistake

Never anchor the aligner to defects, labels, stickers, or anything that can move independently of the part. Only align to permanent, rigid features (machined edges, drilled holes, PCB outlines). If you align to a barcode sticker and someone places it crooked, the camera shifts all your inspection boxes to the wrong position.

Try it yourself: Use the simulator below to see what happens when a part shifts on the conveyor. Toggle the aligner off, then move the sliders to watch the inspection boxes lose tracking.

Camera Settings

Aligner Enabled (Track Part)

Status: Tracking Locked / Pass

---

Simulate Real World

Move the part coming down the line.

X Offset 0px

Y Offset 0px

Rotation 0°

Legend

Inspection Region

Alignment locked

Alignment lost

ROI

Read Alignment Explained for the full walkthrough. This is the most important page in this documentation.

Step 3: Regions of Interest (ROIs)

Full guide: Regions of Interest (ROIs)

Now draw the areas where the AI will actually inspect. These are your Regions of Interest (ROIs).

The critical rule: Keep ROIs as small as possible. This is the second biggest source of customer issues. Read Why ROI Size Matters to understand why.

The short version:

1. Create an Inspection Type (e.g., "Surface Quality") with your expected classes (e.g., "good", "defective")
2. Draw rectangular ROIs on each location you want to inspect
3. Make them just big enough to contain the feature, no bigger
4. Name them descriptively (e.g., "Surface_Center")

Step 4: Train Your AI Model

Full guide: Training Your AI

Label a few images and train your first model.

The short version:

1. Start with 10-15 images per class. Don't over-collect
2. Double-check every label before training (one mislabel can ruin your model)
3. Train (~30 seconds) to check the signal
4. Test with Live Preview. Try to break it
5. Add targeted data where it fails, retrain

Step 5: Output Rules (IO Block)

Full guide: Setting Up Outputs

Define what happens when the AI makes its decision.

Basic Mode: Set rules for pass/fail. The simplest setup: all ROIs must pass for a global pass. That single binary result gets sent to your PLC, HMI, or output.

Advanced Mode (Node-RED): For anything beyond simple pass/fail: custom dashboards, time-series logic, data routing to MES systems, barcode scanner integration (external reader required), and more. Use tools.overview.ai to generate Node-RED flows from plain English descriptions.

Step 6: Deploy and verify

1. Activate your recipe
2. Set your trigger mode (manual, hardware sensor, PLC, or interval)
3. Run test parts through the system
4. Verify the pass/fail output matches your expectations
5. Check edge cases, the parts that are hardest to classify

Congratulations! You now have a running AI inspection.

Recipe checklist

Before moving on, confirm:

• New recipe created and named
• Image settings configured: sharp, well-lit, consistent
• Alignment set up and tracking reliably
• Regions of Interest (ROIs) drawn: small, well-positioned, named
• AI model trained and tested with Live Preview
• Output rules configured: pass/fail matches expectations
• Recipe activated and deployed with correct trigger mode

What's next?

• Improving Your Model: How to keep your AI getting better over time
• Troubleshooting & FAQ: Common issues and quick fixes
• Classifier vs. Segmenter: Detailed guide on when to use which