SketchDirector: agentic layout reasoning generates complex scene sketches
GVCLab · hf · 2026-11-02
SketchDirector, an arXiv paper from GVC Lab at Great Bay University (Yiqi Liang, Xiaodong Cun), presents an agentic framework for complex scene sketching:
- Decomposes text prompts into draft vector assets
- Arranges layout via iterative reasoning with self-evolving memory
- Refines details using diffusion priors
Demo sketches include a mountain temple amid bamboo groves, a cat sleeping on a sofa, and a man reading a newspaper on a bench. Project page: yiki77.github.io/SketchDirector.
More from Multimodal
- 2D drawings turned into 3D models in stunning demo — azed_ai · 2026-11-02
- PRISM-DMAD-PDMD merge nails realistic skin in 8 steps, no Turbo LoRA — theOliviaRossi · 2026-11-02
- REC Frame Overlay: ComfyUI Node Adds Viewfinder and Y2K Digicam HUDs to Image/Video Batches — Hour-Shower-4028 · 2026-11-02
- Pika speeds up video generation 60% with pure inference engineering in H3 Flex — ghumare64 · 2026-11-02
- AI Prompt Turns Ordinary Product Shots into Premium Commercial Ads — anthara_ai · 2026-11-02
- Photo Restoration Prompt: Deblur Old Photos Without Inventing Details — anthara_ai · 2026-11-02