Research map / Image generation / Autoregressive and latent generation
Latent Wavelet Diffusion For Ultra High-Resolution Image Synthesis
Luigi Sigillo, Shengfeng He, Danilo Comminiello
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 LWD training (SC-VAE plus wavelet-masked loss) | URAE · HPD 2048x2048 | FID: 35.25 -> 32.88 | Improved |
| add LWD | SD3-F16 · Aesthetic-Eval 2K | FID: 43.82 -> 38.74 | Improved |
| add LWD | PixArt-Sigma-XL · Aesthetic-Eval 2K | FID: 39.13 -> 36.14 | Improved |
| add LWD | Sana-1.6B · Aesthetic-Eval 2K | FID: 32.06 -> 34.30 | Hurt |
| add LWD | Sana-1.6B · Aesthetic-Eval 4K | GLCM score: 0.39 -> 0.60 | Improved |
| add LWD | SD3-F16 · Aesthetic-Eval 2K | GLCM score: 0.75 -> 0.74 | Hurt |
| add VAE scale-consistency loss only (ablation) | SD3-Diff4k-F16 · Aesthetic-Eval 2K | FID: 40.18 -> 39.50 | Improved |
| add wavelet masking only (ablation) | SD3-Diff4k-F16 · Aesthetic-Eval 2K | FID: 40.18 -> 39.20 | Improved |
| combine both SC-VAE and wavelet masking (ablation) | SD3-Diff4k-F16 · Aesthetic-Eval 2K | FID: 39.20 -> 38.74 | Improved |
| add scale-consistency loss to VAE (ablation) | Flux-VAE · Aesthetic-4K reconstruction | rFID: 0.73 -> 0.50 | Improved |
| add scale-consistency loss to VAE (ablation) | Sana-AE · Aesthetic-4K reconstruction | LPIPS: 0.159 -> 0.163 | No clear effect |
| use FFT high-pass instead of Haar wavelet for saliency (ablation) | LWD · Aesthetic-Eval 2K | FID: 38.74 -> 39.45 | Hurt |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 2 of 2 on most shared results.
- Sana-1.6B — SANA: Efficient High-Resolution Text-to-Image Synthesis with Linear Diffusion Transformers (ICLR 2025) · better on 3 of 4 shared results
- PixArt-Sigma-XL — PixArt-Sigma: Weak-to-Strong Training of Diffusion Transformer for 4K Text-to-Image Generation (ECCV 2024) · better on 1 of 1 shared result
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
- Diffusion-4K: Ultra-High-Resolution Image Synthesis with Latent Diffusion Models (CVPR 2025 · similarity 0.92)
- ScaleDiff: Higher-Resolution Image Synthesis via Efficient and Model-Agnostic Diffusion (NeurIPS 2025 · similarity 0.90)
- UltraDiff:Transferring High-Fidelity Priors to Compressed Latent Spaces for High-resolution Image Generation (NeurIPS 2026 · similarity 0.90)
- DiffuseHigh: Training-Free Progressive High-Resolution Image Synthesis Through Structure Guidance (AAAI 2025 · similarity 0.90)
- UltraPixel: Advancing Ultra High-Resolution Image Synthesis to New Peaks (NeurIPS 2024 · similarity 0.90)
- FAM Diffusion: Frequency and Attention Modulation for High-Resolution Image Generation with Stable Diffusion (CVPR 2025 · similarity 0.90)
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.
