KINESYS improves image editing using references to guide effects
KINESYS: Test-Time Adaptation for Reference-Based Effect Editing in Images
Computer Vision and Pattern Recognition
Summary
Sometimes changing an image to add a visual effect is hard to explain with words, but easier if you show a picture as a reference. The authors created KINESYS, a method that adapts during editing to copy the effect from a reference image onto another, while keeping the original picture’s content and layout intact. This technique uses feedback from both the original image and the reference to improve the effect transfer without unwanted changes. Users and automated tests found it keeps the source image’s important details better than previous methods and still does a good job applying the new effect.
What this means in practice
- •For photo editing software developers: Add fine-tuned effect transfer features that accurately replicate visual styles from example images while maintaining original content.$Commercial implications: This enables photo editing apps to offer advanced style transfer tools that produce higher-quality edits, appealing to professional and hobbyist users.
- •For video production teams: Enhance visual effects workflows by applying reference-based effect edits to individual frames while preserving scene composition.
Authors
Noah Frahm, Luchao Qi, Haoze Li, Dan B Goldman, Sambit Bhattacharya, Roni Sengupta
Abstract
Visual effects are often easier to demonstrate via reference than to describe in words. We study effect-reference editing, where a reference image shows a desired visual effect and language identifies which properties should guide the edit. Existing single-pass editing techniques often fail to make fine-grained edits, resulting in unintended changes to source content or deviations from the reference effect. We introduce KINESYS, a test-time adaptation framework for effect-reference editing that combines reference-aware and source-aware feedback to transfer the desired effect while preserving the composition and content of the original shot. To ensure KINESYS can produce a diverse set of edits that preserve the source, we develop a robust optimization strategy in which the generative model is fine-tuned across multiple noise samples and optimization runs. Evaluation with VLM judges, source preservation metrics (LPIPS, TPIPS), and a user study shows that KINESYS outperforms state-of-the-art image editors in source preservation; on effect transfer, it remains competitive with Nano Banana 2 and outperforms open-weight editors.