Researchers have developed AViTS, a novel framework designed to enhance the efficiency of Diffusion Transformers (DiTs) used for image generation. This method adaptively selects spatiotemporal tokens, prioritizing critical ones for resolution refinement while deferring less important ones. AViTS aims to reduce redundant computations and improve the quality-efficiency trade-off in dynamic-resolution sampling for DiTs. The framework has demonstrated significant FLOPs reduction, achieving up to 6.34x on FLUX and nearly 9x on Qwen Image Edit and FLUX.1-Kontext-dev, with potential for even greater gains when combined with distillation techniques. AI
IMPACT This framework could significantly reduce the computational cost of high-quality image generation, making advanced AI models more accessible and practical for wider use.
RANK_REASON The cluster contains an academic paper detailing a new technical framework for improving AI model efficiency. [lever_c_demoted from research: ic=1 ai=1.0]
AI-generated summary · Google Gemini · from 1 sources. How we write summaries →
