Outline and Detail: A Semantic-Driven Framework for Layered 2D Character Generation
summary
A pipeline called Spiritus produces layered, rigged 2D characters from mixed text and sketch inputs in minutes by semantic-matching prompts to an open mask library using word embeddings and cosine similarity, building a silhouette, then generating outlines with ComfyUI and base plus LoRA models, segmenting with a mask-guided approach after SAM proves unreliable, assembling deformable attachment meshes with a unified 2D skeleton, driving motion via BVH motion capture, exporting to Spine API, and showing user-study gains in ease and animation quality while noting front-view segmentation and face-template limits.
AI-written summary · about
chapters (8)
- 0:00Introduction to framework
- 2:31Key challenges addressed
- 3:40Framework workflow design
- 6:10Character mesh system
- 7:53Evaluation and results
- 8:57Limitations and future work
- 9:40Applications and community
- 10:22Q&A session
related talks
19mSketchforme: Composing Sketched Scenes from Text Descriptions for Interactive Applications
08mAutomated Accessory Rigs for Layered 2D Character Illustrations
34m2D or not 2D
49mCharacter Modeling in CSS
19mAuthoring Illustrations of Human Movements by Iterative Physical Demonstration
57mLet's get serious: the Wonderful World of Character Design
47m3D Digital Avatars with Machine Learning
30mCharacter Modeling in CSS