Research map / Video / Video generation
Accelerating Diffusion Sampling via Exploiting Local Transition Coherence
Shangwen Zhu, Han Zhang, Zhantao Yang, Qianyu Peng, Zhao Pu, Huangji Wang, Fan Cheng
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 |
|---|---|---|---|
| add LTC-Accel to DeepCache (SD v2, 10 to 8 steps) | DeepCache on SD v2 · MS-COCO 2017 (1000 prompts, DDIM) | ImageReward: -0.2246 -> -0.2739 | Hurt |
| add LTC-Accel to DeepCache (20 to 16 steps) | DeepCache on SD v2 · MS-COCO 2017 (1000 prompts, DDIM) | ImageReward: 0.2445 -> 0.2456 | No clear effect |
| add LTC-Accel to DeepCache (50 to 38 steps) | DeepCache on SD v2 · MS-COCO 2017 (1000 prompts, DDIM) | ImageReward: 0.4039 -> 0.4096 | Improved |
| add LTC-Accel to DeepCache (100 to 75 steps) | DeepCache on SD v2 · MS-COCO 2017 (1000 prompts, DDIM) | ImageReward: 0.4242 -> 0.4246 | No clear effect |
| add LTC-Accel to Align Your Steps (10 to 8 steps) | Align Your Steps on SD v1.5 · MS-COCO 2017 (1000 prompts, DPM-Solver++) | ImageReward: 0.1332 -> 0.1309 | No clear effect |
| add LTC-Accel to Align Your Steps (20 to 15 steps) | Align Your Steps on SD v1.5 · MS-COCO 2017 (1000 prompts, DPM-Solver++) | ImageReward: 0.1632 -> 0.1510 | Hurt |
| add LTC-Accel to Align Your Steps (30 to 20 steps) | Align Your Steps on SD v1.5 · MS-COCO 2017 (1000 prompts, DPM-Solver++) | ImageReward: 0.1901 -> 0.2131 | Improved |
| add LTC-Accel to Animated-Diff (30 to 19 steps) | Animated-Diff · MS-COCO 2017 (100 prompts, text-to-video) | Textual Faithfulness: 0.3662 -> 0.3465 | Hurt |
| add LTC-Accel to Animated-Diff (10 to 7 steps) | Animated-Diff · MS-COCO 2017 (100 prompts, text-to-video) | Textual Faithfulness: 0.2439 -> 0.2426 | No clear effect |
| add LTC-Accel to CogVideoX 2B (40 to 26 steps) | CogVideoX 2B · MS-COCO 2017 (100 prompts, text-to-video) | Textual Faithfulness: 0.3918 -> 0.3775 | Hurt |
| add LTC-Accel to CogVideoX 2B (40 to 26 steps) | CogVideoX 2B · MS-COCO 2017 (100 prompts, text-to-video) | Frame Consistency: 0.9514 -> 0.9511 | No clear effect |
| add LTC-Accel to distilled Animated-Diff-Lightning (4 to 3 steps) | Animated-Diff-Lightning · MS-COCO 2017 (100 prompts, distilled video) | Textual Faithfulness: 0.3662 -> 0.3550 | Hurt |
Closest papers by meaning
- A Simple Early Exiting Framework for Accelerated Sampling in Diffusion Models (ICML 2024 · similarity 0.91)
- Faster Diffusion: Rethinking the Role of the Encoder for Diffusion Model Inference (NeurIPS 2024 · similarity 0.91)
- Morse: Dual-Sampling for Lossless Acceleration of Diffusion Models (ICML 2025 · similarity 0.90)
- AdaDiff: Adaptive Step Selection for Fast Diffusion Models (AAAI 2025 · similarity 0.89)
- Dual-Rate Diffusion: Accelerating diffusion models with an interleaved heavy-light network (NeurIPS 2026 · similarity 0.89)
- Spectral Progressive Diffusion for Efficient Image and Video Generation (NeurIPS 2026 · 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.
