# Text-Driven

TurboEdit is a technology developed by Adobe Research, designed to tackle the challenges of precise image inversion and decoupled image editing. It achieves precise image editing in just a few steps using iterative inversion techniques and text prompt-based conditional control. This technology is not only fast but also outperforms existing multi-step diffusion model editing techniques.

HoloDreamer is a text-driven 3D scene generation framework capable of producing immersive and view-consistent fully enclosed 3D scenes. It consists of two fundamental modules: stylized rectangular panoramic generation and enhanced two-phase panoramic reconstruction. This framework first generates high-definition panoramic images as a complete initialization for the 3D scene, then quickly reconstructs the 3D scene using 3D Gaussian scattering (3D-GS) technology, resulting in view-consistent and fully enclosed 3D scene generation. HoloDreamer's main advantages include high visual consistency, harmony, and robust reconstruction quality and rendering.

MoMask is a model for text-driven 3D human motion generation. It employs a hierarchical quantization scheme to represent human motion as multi-layer discrete motion tokens with high fidelity details. MoMask utilizes two distinct bidirectional Transformer networks for generation, predicting motion tokens from text input. This model outperforms existing methods in the text-to-motion generation task and can seamlessly be applied to related tasks such as text-guided temporal repair.

Story-to-Motion is a novel task that takes a story (top green area) and generates motion and trajectories consistent with the textual description. The system utilizes modern large language models as a text-driven motion scheduler, extracting a series of (text, location) pairs from long texts. It also develops a text-driven motion retrieval scheme, combining classic motion matching with motion semantics and trajectory constraints. Furthermore, it designs a progressive masking transformer to address common problems in transition motions, such as unnatural poses and sliding. The system excels in three different subtasks: trajectory following, temporal action combination, and action mixing, outperforming previous motion synthesis methods.