Stroke of Surprise: Progressive Semantic Illusions in Vector Sketching
Huai-Hsun Cheng, Siang-Ling Zhang, Yu-Lun Liu
2026-02-13
Summary
This paper introduces a new way to create visual illusions, but instead of relying on how things are positioned in space, it focuses on changing what an image *means* over time as you draw it. It's like starting to draw one thing, and then adding strokes that transform it into something completely different, but in a smooth and believable way.
What's the problem?
The main challenge is creating a drawing that works as two different images. Imagine starting a sketch that clearly looks like a duck, but then being able to add to it so it convincingly becomes a sheep. The problem isn't just making each image look good on its own, but ensuring the initial part of the drawing supports *both* interpretations. Previous methods often locked the initial drawing, making it hard to find a good foundation for the second image.
What's the solution?
The researchers developed a system called 'Stroke of Surprise' that uses a clever optimization technique. Instead of freezing the initial strokes, their method subtly adjusts them as it adds new strokes, searching for a 'common structural subspace' – a basic shape that can be seen as either a duck *or* a sheep. They also added a special 'Overlay Loss' that makes sure the new strokes blend with the existing ones instead of just covering them up, creating a seamless transformation. This system uses something called 'Score Distillation Sampling' to guide the drawing process.
Why it matters?
This work is important because it expands the idea of visual illusions beyond just how things look to *what* they represent. It moves illusions from being a single, static moment to a dynamic process of transformation. This could have applications in areas like creative tools, animation, or even understanding how our brains interpret ambiguous images, and it opens up new possibilities for generating surprising and engaging visuals.
Abstract
Visual illusions traditionally rely on spatial manipulations such as multi-view consistency. In this work, we introduce Progressive Semantic Illusions, a novel vector sketching task where a single sketch undergoes a dramatic semantic transformation through the sequential addition of strokes. We present Stroke of Surprise, a generative framework that optimizes vector strokes to satisfy distinct semantic interpretations at different drawing stages. The core challenge lies in the "dual-constraint": initial prefix strokes must form a coherent object (e.g., a duck) while simultaneously serving as the structural foundation for a second concept (e.g., a sheep) upon adding delta strokes. To address this, we propose a sequence-aware joint optimization framework driven by a dual-branch Score Distillation Sampling (SDS) mechanism. Unlike sequential approaches that freeze the initial state, our method dynamically adjusts prefix strokes to discover a "common structural subspace" valid for both targets. Furthermore, we introduce a novel Overlay Loss that enforces spatial complementarity, ensuring structural integration rather than occlusion. Extensive experiments demonstrate that our method significantly outperforms state-of-the-art baselines in recognizability and illusion strength, successfully expanding visual anagrams from the spatial to the temporal dimension. Project page: https://stroke-of-surprise.github.io/