# Image Recognition Bot Templates and OCR: Asset Lab
> Build image recognition bot assets in Asset Lab: capture an emulator or cloud device screen, pick tap coordinates, test an OCR region live and crop templates.
Source: https://automationmacro.com/docs/asset-lab (Studio, updated 2026-09-04)
Asset Lab is the helper app in Macro Automation Studio (MAS) where an image recognition bot gets its assets. It captures a device screen, then lets you pick coordinates, test OCR regions and crop template images into your Image Library. Use it before you write `find_object` or `read_text` calls, so the numbers in your script come from real pixels.
## Before you start
- MAS is installed and you are signed in. Asset Lab reuses your session.
- A device to capture: a local emulator with ADB switched on, or a cloud device in the ready state.
- On a Mac, adb must be found by MAS. See [Install MAS](/docs/install) if MAS reports "ADB executable not found".
## Open Asset Lab
Asset Lab is a separate window. On Windows it is `AssetHelper.exe` next to the app; on a Mac it is `AssetLab.app` inside the bundle. MAS launches it with your sign-in, adb and Tesseract already configured. Three places open it:
1. In the Code Editor, open the **Assets** panel and click **Open Asset Helper**.
2. In **Device Groups**, click **Open in Asset Helper** on a cloud device card that is ready. Asset Lab opens connected to that device with a fresh capture.
3. In the Code Editor's cloud device view, click **Open in Asset Helper**.
## Connect and capture
The connection area has two source tabs: **Local** and **Cloud**.
1. Click the scan button, then pick the emulator's port from the list. Type the port if the scan does not find it.
2. Click **Connect**.
3. Click **Capture Screenshot** or press F5. Navigate the emulator to the screen you want and capture again whenever you need a fresh frame.
1. Pick a device from the list. Only devices in the ready state can be selected; others show their state, such as stopped or starting. Start a stopped device from **Cloud Devices** in MAS.
2. Click **Connect**. Asset Lab captures the screen at once.
3. Click **Live** to see and control the device. The stream runs over WebRTC at the device's native resolution. Navigate to the screen you want, then click **Freeze & Capture**. The capture is a real adb screenshot, not a video frame, so it matches what `find_object` sees during a run.
Opening Live takes over the stream from anyone else watching the device, including the device view in MAS.
Captures are always native resolution. The footer shows the coordinates under the mouse and how many points, paths and regions you have drawn.
## The tools
| Tool | Key | Use it for |
|---|---|---|
| **Points** | P | Coordinates for `mas.click` and `mas.swipe` |
| **OCR** | R | A region for `mas.read_text`, tested live |
| **Swipe** | S | Start and end coordinates of a drag |
| **Crop** | C | A template image for `mas.find_object` |
Tab toggles the side panel, Esc deselects the tool, and ? lists the shortcuts.
### Pick a point
1. Press P and click the element. Click its center, not its edge.
2. The **Points** panel lists each point. Click **Copy** to put the coordinates on the clipboard as JSON.
Points suit elements that never move. For anything that moves, crop a template instead.
### Test an OCR region
1. Press R and drag a rectangle around the text.
2. The **OCR** panel shows the cropped region and runs Tesseract on this computer. The text appears under **OCR Results**, with the model that read it.
3. Adjust the settings until the text reads correctly, then click **Copy** for the region coordinates or **Copy Text** for the text.
The settings map to `mas.read_text` arguments:
| Asset Lab setting | SDK argument |
|---|---|
| **OCR Model**: Default (Balanced), Fast, Best | `model="eng"`, `"eng_fast"`, `"eng_best"` |
| **Page Segmentation Mode (PSM)**: 0 to 13 | `psm=7` for one line, `8` for one word, `6` for a block, `11` for sparse text |
| **Image Preprocessing**: Grayscale | `color_conversion=ColorConversion.BGR_TO_GRAY` |
| **Image Preprocessing**: Binary (Black/White) | `color_conversion=ColorConversion.BLACK_WHITE` |
> [!NOTE]
> Asset Lab starts with the Default model; the SDK's default is `eng_best`. Pass `model` explicitly when you want the run to match what you tested.
### Crop a template
1. Press C and drag a rectangle around the element.
2. In the dialog, choose **Save As...** to keep a PNG on your computer, or **Upload to Server** to add it to your Image Library.
3. For an upload, enter a file name and confirm. The dialog reports the library name and the image ID, for example "Saved to your image library as play_button (id 42)".
Back in the Code Editor, the **Assets** panel refreshes when the window regains focus, and **Copy ID** puts the ID on the clipboard. The library accepts images up to 10 MB in jpg, jpeg, png, gif or webp; crops from Asset Lab are PNG.
## Use the image ID in an image recognition bot
Declare IDs once with `mas.images` and use the names everywhere else. `find_object_retry` tries three times, two seconds apart, and returns `None` when nothing matches.
```python
import mas
images = mas.images({"play_button": 42, "close_x": 43})
match = mas.find_object_retry(images.play_button, total_tries=3, time_sleep=2.0)
if match:
mas.click(match.x, match.y)
else:
mas.log("Play button not on screen", level="warning")
```
`find_object` accepts a match at a threshold of 0.8 by default. Pass `search_region=Region(x1, y1, x2, y2)` to limit the search to the part of the screen where the element lives; it is faster and avoids look-alikes elsewhere. When a control has several looks, list them with `find_any_object_retry`.
The `mas.images` call also matters outside the script: MAS reads the IDs from it when it publishes a project to the Marketplace and when it packs a cloud run, so the templates travel with the code. Only images in your own library resolve.
For a region you tested in the OCR panel:
```python
import mas
from mas import Region, ColorConversion
gold = mas.read_text(
region=Region(x1=610, y1=24, x2=760, y2=64),
model="eng",
psm=7,
color_conversion=ColorConversion.BLACK_WHITE,
)
print(gold.text, gold.confidence)
```
## Template matching tips for a game bot
- Crop tight. Include the element and a few pixels of margin, not the background around it.
- Avoid text and numbers that change. Crop the icon or the button frame, not the counter next to it.
- Capture at the resolution the macro runs at. A template from a 1080p emulator misses on a 720p one, and a cloud device profile differs from your emulator. Recapture after you change the emulator's resolution or DPI.
- Prefer static, high-contrast elements. Animated areas produce templates that match on some frames only.
- Make one template per state. A greyed-out button and an active one are two images.
- Start at the 0.8 threshold. Raise it when the wrong element matches; lower it a little when a correct element is missed.
- Name images by what they are. `mas.images` names must be valid Python identifiers such as `login_btn`.
## Troubleshooting
### Asset Helper not found
Windows: `AssetHelper.exe` is missing next to the app. Mac: `AssetLab.app` is missing from the bundle. Reinstall MAS from the [download page](/download).
### ADB not found
MAS could not locate adb, so Asset Lab cannot talk to a local emulator. On Windows, reinstall MAS. On a Mac, install adb with Homebrew as described on [Install MAS](/docs/install) and restart MAS.
### No cloud devices, or the device cannot be selected
The Cloud tab lists devices on your account and greys out any that are not ready. Create or start the device from **Cloud Devices** in MAS; the list refreshes on its own.
### Cloud capture failed
The device may have stopped between listing and capture. Refresh the list and retry. If the message says the device's host needs a refresh, stop and start the device in **Cloud Devices**, then capture again.
## Next
Turn a template into a loop on [Image recognition macros](/docs/guides/image-recognition-macros), and read the full argument lists on [Vision](/docs/sdk/vision).