Research map / Image generation / Text-to-image diffusion
SketchEvo: Leveraging Drawing Dynamics for Enhanced Image Synthesis
Zhixin Feng, Runan Yin, LAN YANG, Kaiyue Pang, Ke Li, Honggang Zhang, Yi-Zhe Song
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| add sequence-guided rollback (SGR) at inference on top of SGPO (ablation) | SGPO · Sketchy | ImageReward: 1.03 -> 1.18 | Improved |
| add sequence-guided rollback (SGR) at inference on top of SGPO (ablation) | SGPO · Sketchy | HPS v2: 28.87 -> 30.08 | Improved |
| add sequence-guided rollback (SGR) at inference on top of SGPO (ablation) | SGPO · Sketchy | LPIPS-sketch: 0.20 -> 0.15 | Improved |
| use sketch-sequence-guided preference optimization instead of SPO | ControlNet-SPO · Sketchy | ImageReward: 0.61 -> 1.03 | Improved |
| use SketchEvo instead of ControlNet-SPO on unseen AnimeDiffusion | ControlNet-SPO · AnimeDiffusion | ImageReward: 0.27 -> 1.32 | Improved |
| use SketchEvo instead of ControlNet-SPO on edge maps with pseudo-sequence | ControlNet-SPO · COCO edge maps | HPS v2: 28.05 -> 30.53 | Improved |
| guide rollback with the first stroke s1 instead of later sketch stages (ablation) | SGR · Sketchy | multi-metric | Improved |
| raise candidate pool size k from 5 to 7 (ablation) | SGPO · Sketchy | generation quality | No clear effect |
| switch scoring model to PickScore (ablation) | SGPO · Sketchy | HPS v2, PickScore, LPIPS, CLIP | No clear effect |
| add SketchEvo versus ControlNet in human study overall preference | ControlNet-SPO · 25 sketch+text pairs | overall preference: 16.53% -> 82.67% | Improved |
| use SketchEvo over ControlNet-SPO on structure match in human study | ControlNet-SPO · 25 sketch+text pairs | Structure Match: 4.28 -> 4.56 | Improved |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 4 of 4 on most shared results.
- ControlNet — ControlNet++: Improving Conditional Controls with Efficient Consistency Feedback (ECCV 2024) · better on 6 of 7 shared results
- T2i-Adapter — T2I-Adapter: Learning Adapters to Dig Out More Controllable Ability for Text-to-Image Diffusion Models (AAAI 2024) · better on 2 of 2 shared results
- AnimateDiff — AnimateDiff: Animate Your Personalized Text-to-Image Diffusion Models without Specific Tuning (ICLR 2024) · better on 2 of 2 shared results
- UniControl — UniControl: A Unified Diffusion Model for Controllable Visual Generation In the Wild (NeurIPS 2023) · better on 2 of 2 shared results
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- It's All About Your Sketch: Democratising Sketch Control in Diffusion Models (CVPR 2024 · similarity 0.89)
- SketchAgent: Language-Driven Sequential Sketch Generation (CVPR 2025 · similarity 0.89)
- Latent Representation Matters: Human-like Sketches in One-shot Drawing Tasks (NeurIPS 2024 · similarity 0.88)
- One-Shot Reference-based Structure-Aware Image to Sketch Synthesis (AAAI 2025 · similarity 0.88)
- Can Vision–Language Models Assess Graphic Design Aesthetics? A Benchmark, Evaluation, and Dataset Perspective. (ICLR 2026 · similarity 0.87)
- Text to Sketch Generation with Multi-Styles (NeurIPS 2025 · similarity 0.87)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
