Research map / Image generation / Fast diffusion sampling
Diffusion Mental Averages
Phonphrm Thawatdamrongkit, Sukit Seripanitkarn, Supasorn Suwajanakorn
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 |
|---|---|---|---|
| align denoising trajectories of 1000 noise latents in h-space instead of averaging VAE latents or pixels | Realistic Vision v5.1 · Concept prototypes | ImageReward: 0.823 (D4M) -> 1.002 | Improved |
| DMA instead of D4M/MGD3 for run-to-run consistency | Realistic Vision v5.1 · Concept prototypes | Consistency CLIP dist (lower better): 0.168 -> 0.031 | Improved |
| DMA instead of Avg VAE/GANgealing for representativeness | Realistic Vision v5.1 · Concept prototypes | Representativeness CLIP dist (lower better): 0.386 (GANgealing) -> 0.179 | Improved |
| Avg VAE and GANgealing versus DMA on consistency | baselines · Concept prototypes | Consistency: 0 vs 0.031 (CLIP) | Hurt |
| add LoRA or Textual Inversion cluster-specific conditioning to DMA for modes (ablation) | DMA mode averages · Cluster prototypes (qualitative) | semantic faithfulness | Improved |
| cluster modes before averaging instead of single overall average (ablation) | DMA · Unsupervised modes (qualitative) | mode coverage | Improved |
Closest papers by meaning
- The Hidden Language of Diffusion Models (ICLR 2024 · similarity 0.88)
- Understanding diffusion models requires rethinking (again) generalization (NeurIPS 2026 · similarity 0.88)
- Diffusion PID: Interpreting Diffusion via Partial Information Decomposition (NeurIPS 2024 · similarity 0.88)
- CASL: Concept-Aligned Sparse Latents for Interpreting Diffusion Models (NeurIPS 2026 · similarity 0.87)
- I Think, Therefore I Diffuse: Enabling Multimodal In-Context Reasoning in Diffusion Models (ICML 2025 · similarity 0.87)
- Diffuse Everything: Multimodal Diffusion Models on Arbitrary State Spaces (ICML 2025 · similarity 0.87)
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.
