🎨SketchDeco:

Training-Free Latent Composition for Precise Sketch Colourisation

CVPR 2026

Chaitat Utintu, Yi-Zhe Song
SketchX, CVSSP, University of Surrey
Paper arXiv Code
Demo Image.

SketchDeco turns arbitrary sketch, masks, and colour palettes into stunning realistic image.

Abstract

We introduce SketchDeco, a training-free approach to sketch colourisation that bridges the gap between professional design needs and intuitive, region-based control. Our method empowers artists to use simple masks and colour palettes for precise spatial and chromatic specification, avoiding both the tediousness of manual assignment and the ambiguity of text-based prompts. We reformulate this task as a novel, training-free composition problem.

Our core technical contribution is a guided latent-space blending process: we first leverage diffusion inversion to precisely "paint" user-defined colours into specified regions, and then use a custom self-attention mechanism to harmoniously blend these local edits with a globally consistent base image. This ensures both local colour fidelity and global harmony without requiring any model fine-tuning. Our system produces high-quality results in 15-20 inference steps on consumer GPUs, making professional-quality, controllable colourisation accessible.

Architecture

Model Image.

Our training-free sketch colourisation pipeline.

Results

Model Image.

Comparison with SOTA diffusion-based approaches. Our training-free colourisation approach outperforms state-of-the-art (SOTA) approaches in various aspects including local colourisation, colour vividness, colour harmonisation, fidelity to the original sketch, and overall realistic look.

Model Image.

Model Image.

Model Image.

Model Image.

Model Image.

Model Image.

Qualitative evaluation with random sketches. The local sketch colourisation pipeline showcased notable adaptability, allowing users to incorporate three conditional inputs: a variety of sketches, hand-drawn masks, and preferred colour palettes. Consequently, our approach yields precise outcomes tailored to user-defined specifications. It is non-trivial to mention that these sketch images are randomly sourced from www.freepik.com with search keywords: “black-and-white [class] sketch”.

BibTeX

@article{utintu2025sketchdeco,
  author    = {Chaitat Utintu and Yi-Zhe Song},
  title     = {SketchDeco: Training-Free Latent Composition for Precise Sketch Colourisation},
  journal   = {arXiv preprint arXiv:2405.18716},
  year      = {2026},
}