Text Slider Framework Enables Efficient Continuous Concept Control for AI Image and Video Synthesis

A new framework called Text Slider addresses limitations in current diffusion model control methods by providing lightweight, plug-and-play continuous concept manipulation. Recent advances have significantly improved image and video synthesis capabilities, but existing approaches require intensive training time and GPU memory to learn sliders or embeddings. These methods also need retraining for different diffusion backbones, limiting their scalability and adaptability. Text Slider identifies low-rank directions within pre-trained text encoders, enabling continuous control of visual concepts while dramatically reducing training time, GPU memory consumption, and trainable parameters. The framework supports multi-concept composition and continuous control, allowing for fine-grained and flexible manipulation in both image and video synthesis. By reducing computational requirements, Text Slider makes advanced concept control more accessible and practical for various applications. The approach represents arXiv:2509.18831v2, announced as a replace-cross type publication. This development comes as diffusion models continue to advance rapidly in visual synthesis capabilities. The framework's efficiency improvements could accelerate experimentation and implementation of concept-controlled AI art generation.