Research map / Image generation / Text-to-image diffusion
Dynamic Classifier-Free Diffusion Guidance via Online Feedback
Pinelopi Papalampidi, Olivia Wiles, Ira Ktena, Aleksandar Shtedritski, Emanuele Bugliarello, Ivana Kajic, Isabela Albuquerque, Aida Nematzadeh
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| use adaptive timestep-dependent weighting instead of linear weighting of alignment and VQ evaluators (ablation) | Dynamic CFG search (LDM-large) · Gecko prompts | Gecko score: 45.0 -> 47.2 | Improved |
| use adaptive weighting instead of linear weighting (ablation) | Dynamic CFG search (LDM-large) · MS COCO prompts | FID: 25.4 -> 24.8 | Improved |
| replace per-prompt dynamic schedule with static mean schedule of dynamic CFG (ablation) | Dynamic CFG search (LDM-large) · Gecko prompts | Gecko score: 47.2 -> 46.5 | Hurt |
| replace per-prompt dynamic schedule with static mean schedule (ablation) | Dynamic CFG search (LDM-large) · MS COCO prompts | FID: 24.8 -> 26.8 | Hurt |
| add dynamic CFG search with adaptive Alignment + VQ evaluators to default fixed CFG | LDM-large · Gecko prompts | Gecko score: 43.8 -> 47.2 | Improved |
| add dynamic CFG search with adaptive Alignment + VQ evaluators to default fixed CFG | LDM-large · MS COCO prompts | FID: 25.6 -> 24.8 | Improved |
| use dynamic CFG with only the alignment evaluator vs only the VQ evaluator (ablation) | Dynamic CFG search (LDM-large) · Gecko prompts and MS COCO | Gecko score / FID: 45.5 / 26.4 (alignment) vs 44.0 / 24.8 (VQ) | Mixed |
| add gradient guidance combining alignment and VQ models to default CFG | LDM-large · Gecko prompts | Gecko score: 43.8 -> 45.3 | Improved |
| apply annealing CFG schedule instead of fixed CFG | LDM-large · MS COCO prompts | FID: 25.6 -> 28.9 | Hurt |
| filter with pixel-space CLIP instead of latent CLIP at 50% of denoising (ablation) | LDM-small · Gecko prompts | Gecko score: 41.4 -> 44.6 | Improved |
| filter samples (best of 4) with latent VQ discriminator at 25% of denoising | LDM-large · MS COCO | FID: 29.2 -> 27.6 | Improved |
| add alignment and reward evaluators with adaptive weighting to Imagen 3 sampling | Imagen 3 · Gecko | human win rate (%): 53.6 | Improved |
Closest papers by meaning
- CFG++: Manifold-constrained Classifier Free Guidance for Diffusion Models (ICLR 2025 · similarity 0.93)
- Towards a Golden Classifier-Free Guidance Path via Foresight Fixed Point Iterations (NeurIPS 2025 · similarity 0.92)
- Rethinking the Spatial Inconsistency in Classifier-Free Diffusion Guidance (CVPR 2024 · similarity 0.92)
- Adaptive Guidance: Training-free Acceleration of Conditional Diffusion Models (AAAI 2025 · similarity 0.91)
- Eliminating Oversaturation and Artifacts of High Guidance Scales in Diffusion Models (ICLR 2025 · similarity 0.91)
- Elucidating the design space of classifier-guided diffusion generation (ICLR 2024 · similarity 0.90)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
