Abilities
An Ability is a preconfigured AI task that analyzes visual media and returns structured output.
Abilities allow developers to add visual intelligence to applications without training or deploying their own models.
Typical workflow:
Create your ability on EyePop.ai
Image/Video/Live Stream → Ability → Structured Output
For examples of Abilities and their output, see the Abilities Hub
What is an Ability?
An Ability is a specific visual analysis capability such as:
Detecting objects
Classifying scenes
Extracting structured information from images
Understanding events in video
Developers call an Ability using the EyePop API and receive structured results such as bounding boxes, classifications, or extracted text.
Anatomy of an Ability
An Ability is defined by a small set of configuration parameters that control how visual media is analyzed. These parameters determine how the model interprets the task, how much compute is used, and how frequently media is analyzed.
Ability Components
Field
Description
name
Unique identifier for the Ability. Used when calling the API.
description
Human-readable explanation of the task the Ability performs. This is used by the Prompt Creation Agent to generate or refine prompts.
image_size
The resolution images are resized to before being analyzed. Smaller sizes reduce compute cost and increase speed.
prompt
The instructions given to the vision-language model describing what it should detect or classify.
model
The underlying AI model used to perform the analysis.
fps
Frames per second to analyze when processing video or livestreams. Controls how frequently frames are sampled.
Name
The name uniquely identifies the Ability.
Example:
Names are used in API calls and should clearly reflect the task being performed.
Description
The description explains the purpose of the Ability.
This field is especially important because it is used by the Prompt Creation Agent to help generate or improve prompts.
Example:
Good descriptions are:
clear
specific
task-oriented
Avoid vague descriptions such as:
Image Size
The image_size parameter controls how images are resized before inference.
Example:
Reducing image size:
decreases compute usage
increases inference speed
often improves detection consistency
Typical production sizes:
Use Case
Image Size
Object detection
512-640
Find event in video
512-640
Document analysis
768-1024
Prompt
The prompt defines what the model should analyze in the image or video frame. For examples of Ability prompts, see the Abilities Hub.
Example:
Prompts should:
clearly define the task
restrict possible outputs
avoid unnecessary complexity
Model
The model specifies which AI model runs the Ability.
Example:
Different models may vary in:
reasoning ability
speed
compute cost
Abilities typically use a model optimized for vision-language tasks.
FPS (Frames Per Second)
The fps parameter determines how frequently frames are analyzed when processing video or livestreams.
Example:
This means the Ability will analyze 5 frames per second of video.
Choosing the correct FPS helps balance detection accuracy and compute cost.
Use Case
Recommended FPS
Security monitoring
2–5
Sports analytics
5–10
Industrial monitoring
1–3
Lower FPS reduces compute usage while still capturing most events.
Helpful Mental Model
You can think of an Ability as three main parts:
Where:
Model determines reasoning capability
Prompt defines the task
Sampling (fps + image_size) controls performance and compute usage
What Tasks Are Abilities Good For?
Abilities are designed for real-world visual intelligence workloads.
Common use cases include:
Security and Surveillance
Person detection
Intrusion alerts
PPE detection
Retail and Commerce
Product recognition
Shelf monitoring
Customer analytics
Sports Analytics
Player detection
Action classification
Event segmentation
Document Processing
Driver's license extraction
Receipt parsing
Title or invoice extraction
Industrial Automation
Quality inspection
Object counting
Safety compliance monitoring
What is a Compute Unit?
A Compute Unit (CU) represents the cost of running one AI inference task.
The resolution of the image or video does not effect the compute units used as it's resized to the image scaling resolution you define in your ability. The bigger you make this resolution, the more compute units will be used to achieve your task. Typically, accuracy will increase with larger window sizes.
Each time an Ability processes an image or frame, compute resources are consumed.
Video workloads consume compute based on the number of frames analyzed.
Estimating Compute Unit Usage
Estimate pricing coming soon
Overage Costs
Each EyePop plan includes a monthly allocation of compute units.
If usage exceeds the included amount, additional compute is billed automatically.
Example:
4,000
5,200
1,200 units billed at the end of the month (+$60)
This allows applications to scale without interruption.
Optimizing Compute unit Usage
You can reduce compute costs and improve performance by optimizing inputs.
Reduce Image Size
Recommended image width:
640px – 1280px
Very large images increase compute cost.
Reduce Video FPS
Typical production settings:
Security monitoring
2–5
Sports analytics
5–10
Industrial monitoring
1–3
Crop Regions of Interest
Instead of analyzing the entire frame, crop the relevant region before sending it to the Ability.
This reduces compute and improves accuracy. Region of interest (ROI) is supported through the SDK.
Ability Prompt Creation Agent
The Ability Prompt Creation Agent helps developers generate reliable prompts for visual AI tasks.
Prompt design is one of the hardest parts of working with vision-language models. Small wording changes can significantly affect accuracy, consistency, and cost.
The Prompt Creation Agent analyzes the task you want to perform and generates a production-ready prompt optimized for EyePop Abilities.
Why It Exists
Vision models are extremely sensitive to how instructions are written.
Poor prompts can cause issues such as:
inconsistent classifications
overly verbose outputs
hallucinated results
unpredictable formatting
increased compute cost
The Prompt Creation Agent helps avoid these problems by generating prompts that follow tested patterns.
How to run the Ability Prompt Creation Agent
Coming soon to the Dashboard
Last updated