Research map / Image generation / Autoregressive and latent generation
AccDiffusion v2: Toward More Accurate Higher-Resolution Diffusion Extrapolation
Zhihang Lin, Mingbao Lin, Wengyi Zhan, Rongrong Ji
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 |
|---|---|---|---|
| upgrade AccDiffusion to AccDiffusion v2 (patch-aware prompts, dilated sampling with interaction, ControlNet) | AccDiffusion · 4096x4096 (16x) | FID_r: 63.89 -> 60.88 | Improved |
| upgrade AccDiffusion to AccDiffusion v2 | AccDiffusion · 2048x2048 (4x) | IS_r: 16.48 -> 18.62 | Improved |
| upgrade AccDiffusion to AccDiffusion v2 | AccDiffusion · 3072x3072 (9x) | FID_c: 46.46 -> 44.90 | Improved |
| upgrade AccDiffusion to AccDiffusion v2 | AccDiffusion · 4096x4096 (16x) | CLIP score: 33.79 -> 32.83 | Hurt |
| upgrade AccDiffusion to AccDiffusion v2 | AccDiffusion · 4096x4096 (16x) | inference time (min): 26 min -> 35 min | Hurt |
| remove patch-content-aware prompts (ablation) | AccDiffusion v2 · 4096x4096 (16x) | qualitative | Improved |
| remove dilated sampling with window interaction (ablation) | AccDiffusion v2 · 4096x4096 (16x) | qualitative | Improved |
| remove ControlNet-assisted generation (ablation) | AccDiffusion v2 · 4096x4096 (16x) | qualitative | Improved |
| increase ControlNet steps (ablation) | AccDiffusion v2 · 4096x4096 (16x) | image quality vs time | Mixed |
| use very small patch prompt threshold c (ablation) | AccDiffusion v2 · 4096x4096 (16x) | qualitative | Hurt |
| use very large patch prompt threshold c (ablation) | AccDiffusion v2 · 4096x4096 (16x) | qualitative | Hurt |
| run on H100 instead of RTX 3090 | AccDiffusion v2 · 4096x4096 (16x) | inference time (min): 35 min -> 11 min | Improved |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 3 of 3 on most shared results.
- AccDiffusion — AccDiffusion: An Accurate Method for Higher-Resolution Image Generation (ECCV 2024) · better on 9 of 11 shared results
- DiffuseHigh — DiffuseHigh: Training-Free Progressive High-Resolution Image Synthesis Through Structure Guidance (AAAI 2025) · better on 6 of 7 shared results
- DemoFusion — DemoFusion: Democratising High-Resolution Image Generation With No $$$ (CVPR 2024) · better on 6 of 6 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
- Faster Diffusion: Rethinking the Role of the Encoder for Diffusion Model Inference (NeurIPS 2024 · similarity 0.89)
- FAM Diffusion: Frequency and Attention Modulation for High-Resolution Image Generation with Stable Diffusion (CVPR 2025 · similarity 0.88)
- UltraDiff:Transferring High-Fidelity Priors to Compressed Latent Spaces for High-resolution Image Generation (NeurIPS 2026 · similarity 0.88)
- ResMaster: Mastering High-Resolution Image Generation via Structural and Fine-Grained Guidance (AAAI 2025 · similarity 0.88)
- RepLDM: Reprogramming Pretrained Latent Diffusion Models for High-Quality, High-Efficiency, High-Resolution Image Generation (NeurIPS 2025 · similarity 0.87)
- ElasticDiffusion: Training-free Arbitrary Size Image Generation through Global-Local Content Separation (CVPR 2024 · 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.
