Vision Settings

The Vision settings page controls how dIKta.me captures your screen, which AI models process images and video, and what happens to the results.

AI Models

Cloud Vision Provider / Model

The cloud provider and model used when the Cloud toggle is selected in the Vision Action panel.

• Default provider: Gemini
• Default model: gemini-2.5-flash

Any vision-capable model on your configured cloud provider works here:

Local Vision Model

The Ollama model used when the Local toggle is selected.

• Default: minicpm-v
• Must be a vision-capable model pulled into Ollama before use.

Capture Behaviour

Default Query

The text sent to the AI when you submit without typing or recording a question.

Default: Describe what you see and extract any visible text.

Auto-Record Voice Query

When enabled, the microphone starts recording automatically after a screenshot is taken so you can speak your question straight away.

• Default: On
• Stops after the Query Timeout elapses with no speech.

Query Timeout (seconds)

How long dIKta.me waits for voice input before proceeding with the default query.

• Default: 10 seconds

Max Image Dimension (px)

The longest side an image is allowed before dIKta.me resizes it before sending to the AI.

• Default: 2 048 px
• If the image still exceeds 1 MB after resizing, it is re-encoded as JPEG at 85 % quality.

Save Folder

Where screenshots and screen recordings are saved on disk.

• Default: %APPDATA%\DiktaMe\vision\
• Enter a custom path or click Browse to pick a folder.
• Leave empty to use the default location.

Output Behaviour

Each Vision action has its own Inject at cursor toggle. When on, the AI response is typed into the active window at your cursor. When off, the response goes to the clipboard only.

Video Recording

Video Quality

Frame rate is fixed at 30 fps.

Microphone Audio

Captures your microphone during recording.

• Default: On

When enabled, a Microphone Device dropdown appears so you can choose which mic to use for the recording.

System Audio

Captures audio playing on your computer (apps, browser tabs, etc.) via WASAPI loopback.

• Default: On

When enabled, an Output Device dropdown appears so you can choose which playback device to capture.

Webcam Bubble

Overlays a picture-in-picture webcam feed in the bottom-right corner of the recording.

• Default: On
• Size: Bubble width in pixels (default: 200 px, always 16:9 aspect ratio).
• dIKta.me automatically prefers a USB camera over the built-in webcam.

Max Recording Duration (seconds)

Recording stops automatically after this many seconds even if you haven't clicked Stop.

• Default: 120 seconds

Action Prompts

The Action Prompts section (collapsed by default — click to expand) lets you customise the instructions sent to the AI for each Vision action. Changes apply to both Cloud and Local providers.

The Cloud and Local tabs each also have their own System Prompt and Default Query for screenshot/image requests.

Advanced

Ollama Keep-Alive (seconds)

How long Ollama keeps the local vision model loaded in VRAM after the last inference call.

• Default: 300 seconds (5 minutes)
• Increase this if you take multiple screenshots in quick succession and want to avoid re-loading the model each time.

Max Response Tokens

Upper limit on the tokens the AI may return for a vision query.

• Default: 4 096

Temperature

Controls how literal vs. creative the AI's response is.

• Default: 0.3
• Keep low for OCR and table extraction. Raise slightly for descriptive tasks.