FlowSlider: Training-Free Continuous Image Editing via Fidelity-Steering Decomposition

Dimensions overall score 7.0

GitHub Code Pulse

No public code linked for this paper yet.

• FlowSlider enables continuous image editing with slider-style control of edit strength without requiring any post-training.method 95%
• Existing learning-based slider methods for continuous editing introduce additional training overhead and can reduce reliability under distribution shifts.limitation 90%
• FlowSlider decomposes the editing update into a fidelity term that preserves identity/structure and a steering term that drives semantic transition.technical 95%
• The fidelity and steering terms in FlowSlider's decomposition are approximately orthogonal.technical 85%
• The orthogonal decomposition enables stable strength control by scaling only the steering term while keeping the fidelity term unchanged.technical 90%
• FlowSlider provides smooth and reliable control for continuous editing across diverse tasks.result 85%
• Continuous image editing aims to provide slider-style control while preserving source-image fidelity and maintaining a consistent edit direction.technical 95%

Competitive landscape

Competitor map is still being generated for this paper. Enable generation or check back soon.

Startup potential card