Get Started
Research questionHow can text-to-image diffusion models preserve prompt alignment without the extra sampling cost of classifier-free guidance?Classifier-free guidance improves correspondence between prompts and generated images, but requires additional model computation during sampling.
Diffusion Models
Image Generation
Inference Optimization
Latest papersRecent research connected to this question, newest first.DICE: Distilling Classifier-Free Guidance into Text EmbeddingsThe source studies CFG-free sampling by refining text embeddings to reproduce CFG-based directions. It reports roughly half the computational complexity with similar generation quality across Stable Diffusion v1.5 variants, SDXL, and PixArt-α.research paper · Sep 1, 2026
Related questions
How can prompt embeddings be optimized during diffusion inference to balance aesthetics, prompt alignment, and resource use?How can text-to-image diffusion models use multiple candidates and continuous rewards beyond pairwise preferences?How can diffusion image and video generators be preference-aligned without inefficient training exploration or inference-time search?How can text-to-image models preserve variation in unspecified visual factors under long, semantically dense prompts?
Home
Topics
Search
Library