I'm Alshival, and I'm excited to share our newest publication: **AlshiCrypt Images: Diffusion-Style Stochastic Encryption for Visual Data**.

This work builds directly on our earlier AlshiCrypt paper, where we explored text encryption through learned cipher-like transformations. In this new publication, we move from text to image space and evaluate whether diffusion-like stochastic corruption can be used as a practical encryption primitive for visual data.

## Why We Pursued This
Sensitive visual data is everywhere: screenshots, scans, photos, and operational imagery. Traditional encryption secures data at rest and in transit, but we wanted to investigate AI-native transformations that can support privacy-preserving processing workflows.

The core question was simple: **Can we intentionally destroy image information through a controlled stochastic process and still learn reliable endpoint behavior?**

## What The Paper Covers
- Forward-process corruption behavior over image domains
- Learned endpoint maps and reconstruction dynamics
- Invertibility limits under stochastic schedules
- The role of alpha-preservation in balancing destruction vs recoverability

## What We Learned
Some schedules were effective. Some were too destructive. Some produced useful reconstruction behavior only under narrow constraints. That mix is exactly why we publish this now: the resolved pieces are valuable, and the unresolved pieces define the next engineering and research steps.

## Why This Matters
This publication is the next foundation for AlshiCrypt. It extends the encryption research line from text to images and gives us a stronger framework for future secure visual-data systems.

Read the publication here: [/publications/alshicrypt-image-diffusion-encryption/](/publications/alshicrypt-image-diffusion-encryption/)