Research map / Image generation / Text-to-image diffusion
Make a Cheap Scaling: A Self-Cascade Diffusion Model for Higher-Resolution Adaptation
Lanqing Guo, Yingqing HE, Haoxin Chen, Menghan Xia, Xiaodong Cun, Yufei Wang, Siyu Huang, Yong Zhang, Xintao Wang, Qifeng Chen, Ying Shan, Bihan Wen
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 pivot-guided self-cascade (tuning-free) to SD 2.1 at 1024^2 | SD 2.1 · Laion-5B 1024^2 | FID_r: 29.89 -> 12.25 | Improved |
| add pivot-guided self-cascade (tuning-free) to SD 2.1 at 2048^2 | SD 2.1 · Laion-5B 2048^2 | FID_r: 104.70 -> 38.99 | Improved |
| tune lightweight upsampler modules (0.002M params) instead of full fine-tuning | SD 2.1 · Laion-5B 2048^2 | FID_r: 43.55 -> 18.46 | Improved |
| tune upsampler modules instead of LoRA rank 4 | SD 2.1 · Laion-5B 1024^2 | FID_r: 14.74 -> 12.40 | Improved |
| add tuning of upsampler modules to the tuning-free cascade (ablation) | Ours-TF · Laion-5B 1024^2 | FID_r: 12.25 -> 12.40 | No clear effect |
| add tuning of upsampler modules to the tuning-free cascade (ablation) | Ours-TF · Laion-5B 1024^2 | pFID_r: 19.59 -> 15.35 | Improved |
| tuning-free cascade on extreme examples with fine-grained repeated objects | Ours-TF · qualitative failures | repetition artifacts | Hurt |
| reduce training from 50k to 10k steps for full fine-tuning on video (ablation) | Full Fine-tuning · Webvid-10M | FVD_r: 531.57 -> 721.32 | Hurt |
| reduce LoRA training from 50k to 10k steps on video (ablation) | LORA-R4 · Webvid-10M | FVD_r: 623.72 -> 1221.46 | Hurt |
| use self-cascade tuning instead of SD+SR post-processing | SD 2.1 · Laion-5B 1024^2 | pFID_r: 17.21 -> 15.35 | Improved |
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.
- ScaleCrafter — ScaleCrafter: Tuning-free Higher-Resolution Visual Generation with Diffusion Models (ICLR 2024) · better on 2 of 2 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
- ScaleCrafter: Tuning-free Higher-Resolution Visual Generation with Diffusion Models (ICLR 2024 · similarity 0.91)
- DiffuseHigh: Training-Free Progressive High-Resolution Image Synthesis Through Structure Guidance (AAAI 2025 · similarity 0.91)
- Matryoshka Diffusion Models (ICLR 2024 · similarity 0.90)
- Ultra-Resolution Adaptation with Ease (ICML 2025 · similarity 0.89)
- Hierarchical Patch Diffusion Models for High-Resolution Video Generation (CVPR 2024 · similarity 0.89)
- ResMaster: Mastering High-Resolution Image Generation via Structural and Fine-Grained Guidance (AAAI 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.
