What question did this study set out to answer?

To introduce a framework for text-to-3D content creation that enables interactive geometric control by conditioning 3D generation on hand-drawn sketches.

October 26, 2023

Control3D: Towards Controllable Text-to-3D Generation

View Full Paper

Ask AI

Bookmark

Authors

YCYang ChenSoochow University YPYingwei PanPeptidream (Japan)YLYehao LiUniversity of Science and Technology of China

Discussion

Member takes

Implication

Computational study demonstrates sketch-conditioned text-to-3D synthesis in neural radiance fields, highlighting improved spatial control in generative modeling.

Key Points

To introduce a framework for text-to-3D content creation that enables interactive geometric control by conditioning 3D generation on hand-drawn sketches.
Adapted a 2D conditioned diffusion model (ControlNet) to guide the training of a neural radiance field (NeRF) aligned with text prompts and sketches.
Utilized a pre-trained differentiable photo-to-sketch model to estimate sketches from rendered views and enforce geometric consistency with user drawings.
Synthesized faithful 3D scenes that closely adhere to both input descriptive text prompts and user-specified sketch geometries.
Achieved consistent multi-view rendering without sacrificing textual alignment or visual quality.

Cite This Study

Chen et al. (2023) studied this question.

synapsesocial.com/papers/6a09ea6ab0d552aa8b45f4de https://doi.org/10.1145/3581783.3612489

View Full Paper

Ask AI

Bookmark

Implication

Computational study demonstrates sketch-conditioned text-to-3D synthesis in neural radiance fields, highlighting improved spatial control in generative modeling.

Control3D: Towards Controllable Text-to-3D Generation

Discussion

Member takes

Implication

Key Points

Cite This Study

Control3D: Towards Controllable Text-to-3D Generation

Discussion

Member takes

Implication

Key Points

Cite This Study

Also Consider

Also Consider