Skip to content

[New Skill]: Deepfake & Synthetic Media Guard #48

@rosspeili

Description

@rosspeili

Skill Name

security/deepfake_guard

What should this skill do?

Ideal for CERTH ITI's Computer Vision & AI Ethics Teams

Agents conducting independent automated research on X, Reddit, or the open web are highly susceptible to taking synthetic or deepfake imagery as absolute truth. This skill provides a lightweight, pre-flight analysis of an image URL. It runs a rapid noise-floor/frequency analysis to determine if the image is statistically likely to be generated by a diffusion model (Midjourney, Stable Diffusion), preventing the agent from acting on fake news.

Contributors: Ensure the Python logic (skill.py) uses lightweight, fast-loading models (e.g., ONNX runtime) to prevent massive memory footprints for the agent host. Provide extensive tests in test_skill.py against known real vs. fake datasets.

Ideal Inputs & Outputs

Input:
{
"media_url": "https://socialmedia.com/viral_image_123.jpg"
}

Output:
{
"is_synthetic": true,
"confidence_score": 0.92,
"artifacts_detected": ["diffusion_noise_pattern", "inconsistent_lighting"]
}

Targeted Models (if applicable)

Model Agnostic (All)

Metadata

Metadata

Assignees

No one assigned

    Labels

    enhancementNew feature or requestskill requestRequest for a new capability to be added.

    Projects

    No projects

    Milestone

    No milestone

    Relationships

    None yet

    Development

    No branches or pull requests

    Issue actions