Research map / Image generation / Text-to-image diffusion
Temporal Concept Dynamics in Diffusion Models via Prompt-Conditioned Interventions
Ada Görgün, Fawaz Sammani, Nikos Deligiannis, Bernt Schiele, Jonas Fischer
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 |
|---|---|---|---|
| intervene later (tau30 instead of tau90) in the denoising trajectory (ablation) | PCI editing · Text-driven editing | CLIP_img (content preservation): 0.7679 -> 0.9343 | Improved |
| intervene earlier (tau90 instead of tau30) in the denoising trajectory (ablation) | PCI editing · Text-driven editing | CLIP_dir (concept insertion): 0.1014 -> 0.1963 | Improved |
| use PCI at tau70 instead of Stable Flow | PCI editing · Text-driven editing | CLIP_dir: 0.0631 -> 0.1678 | Improved |
| use PCI at tau50 instead of NTI+P2P | PCI editing · Text-driven editing | CLIP_img and CLIP_dir: 0.8666 -> 0.8885 CLIP_img; 0.0979 -> 0.1387 CLIP_dir | Improved |
| use rectified-flow models SD 3.5 and FLUX instead of diffusion models | concept insertion · CIS crossing times | band width W: sharper transition, earlier crossing | Mixed |
| swap Qwen-VL 3B scorer for Qwen 7B, LLaVA-OneVision-7B or SmolVLM-2B (ablation) | CIS scorer · animals, SD 2.1 | CIS curves | No clear effect |
| use paraphrased base prompts instead of original prompt (ablation) | CIS · animals, handmade objects, common actions | CIS curves | No clear effect |
| increase seed budget beyond k=100 (ablation) | CIS estimation · SD 2.1 | concept coverage stability | No clear effect |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.
- Stable Flow — Stable Flow: Vital Layers for Training-Free Image Editing (CVPR 2025) · better on 3 of 3 shared results
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- How Diffusion Models Memorize (NeurIPS 2026 · similarity 0.90)
- The Hidden Language of Diffusion Models (ICLR 2024 · similarity 0.89)
- Concept Reachability in Diffusion Models: Beyond Dataset Constraints (ICML 2025 · similarity 0.89)
- Adjusting Initial Noise to Mitigate Memorization in Text-to-Image Diffusion Models (NeurIPS 2025 · similarity 0.89)
- Tracing the Roots: Leveraging Temporal Dynamics in Diffusion Trajectories for Origin Attribution (NeurIPS 2025 · similarity 0.89)
- Unveiling Concept Attribution in Diffusion Models (NeurIPS 2025 · similarity 0.89)
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.
