Research map / Image generation / Autoregressive and latent generation
Generative latent space tooling
Reuse and analyze latent spaces and tokenizers of image generators.
129 accepted papers at ICML, NeurIPS, ICLR, CVPR and AAAI (2016–2026) · an approach in Autoregressive and latent generation, keyword cluster “latent space · modeling · vae”
State History Benchmarks Tried Checklist Preprints Who Export
Where it stands
Still moving The best reported result is still improving on 1 of 1 tracked benchmark.
33 papers in 2023–24 and 61 in 2025–26 (at ICML, NeurIPS, ICLR, CVPR and AAAI); its share of the topic held steady, from 7.0% to 6.8%.
Open questions the data points to
- ImageNet 256×256 (FID): best went from 2.79 (up to 2024) to 1.46 in 2025 — Language-Guided Image Tokenization for Generation (CVPR 2025 · 3 citations)
How this is decided
A benchmark is tracked when at least four papers of this approach report their own method on the same dataset and metric in at least two different years. Its record is the best number reported up to 2024 compared with the best reported in 2025–26: still improving means at least 1% better, flat means at most 0.2% better with at least two papers from 2025–26 trying. Numbers on a different scale from the rest (a fraction among percentages) and far outliers are left out. Results come from different papers with different setups, so they are signals, not head-to-head comparisons. The share of papers is counted at ICML, NeurIPS, ICLR, CVPR and AAAI.
How it got here
Most cited paper of each year
- 2016 · Conditional Image Generation with PixelCNN Decoders (NeurIPS 2016 · 1,973 citations)
- 2017 · Plug & Play Generative Networks: Conditional Iterative Generation of Images in Latent Space (CVPR 2017 · 504 citations)
- 2018 · Towards Text Generation with Adversarially Learned Neural Outlines (NeurIPS 2018 · 37 citations)
- 2019 · From voxels to pixels and back: Self-supervision in natural-image reconstruction from fMRI (NeurIPS 2019 · 62 citations)
- 2020 · Prior Guided GAN Based Semantic Inpainting (CVPR 2020 · 102 citations)
- 2021 · Taming Transformers for High-Resolution Image Synthesis (CVPR 2021 · 2,386 citations)
- 2022 · MaskGIT: Masked Generative Image Transformer (CVPR 2022 · 424 citations)
- 2023 · Latent-NeRF for Shape-Guided Generation of 3D Shapes and Textures (CVPR 2023 · 326 citations)
- 2024 · DemoFusion: Democratising High-Resolution Image Generation With No $$$ (CVPR 2024 · 42 citations)
- 2025 · 3DTopia-XL: Scaling High-quality 3D Asset Generation via Primitive Diffusion (CVPR 2025 · 27 citations)
- 2026 · StrokeFusion: Vector Sketch Generation via Joint Stroke-UDF Encoding and Latent Sequence Diffusion (AAAI 2026 · 1 citation)
Earliest papers here
- Conditional Image Generation with PixelCNN Decoders (NeurIPS 2016 · 1,973 citations)
- Pixel Recurrent Neural Networks (ICML 2016 · 1,571 citations)
- Comparative Deep Learning of Hybrid Representations for Image Recommendations (CVPR 2016 · 70 citations)
Benchmark records
The best number each year that beat every earlier one, among papers of this approach reporting their own method. Numbers are extracted by AI from each paper's text; check the paper before citing.
ImageNet 256×256 · FID (lower is better)
4 papers, 2022–2025 · still improving
| Year | Best so far | Set by |
|---|---|---|
| 2022 | 6.18 | MaskGIT: Masked Generative Image Transformer (CVPR 2022 · 424 citations) |
| 2024 | 2.79 | ENAT: Rethinking Spatial-temporal Interactions in Token-based Image Synthesis (NeurIPS 2024 · 1 citation) |
| 2025 | 1.46 | Language-Guided Image Tokenization for Generation (CVPR 2025 · 3 citations) |
What has been tried
92 papers here report 1,000 measured changes (a component added, removed or swapped, compared with a baseline): helped 71%, no clear effect 6%, hurt 20%, mixed 3%. Papers mostly publish changes that helped, so "helped" is over-represented.
Ablations in recent papers
| What was changed | Measured on | Result | Effect | Paper |
|---|---|---|---|---|
| Add progressive upscaling | High-resolution generation at 3072² · global semantic coherence | Helped | DemoFusion: Democratising High-Resolution Image Generation With No $$$ (CVPR 2024 · 42 citations) | |
| Use N=64 primitives with a=32 resolution instead of N=2048, a=8 | PrimX fitting · PSNR-F SDF | 62.52 -> 61.05 | Hurt | 3DTopia-XL: Scaling High-quality 3D Asset Generation via Primitive Diffusion (CVPR 2025 · 27 citations) |
| predict absolute coordinates instead of grid cell indices plus relative displacements | Missing glyph generation, full test set · L1 | 0.014 -> 0.020 | Hurt | VecFusion: Vector Font Generation with Diffusion (CVPR 2024 · 25 citations) |
| Use mask-and-replace diffusion instead of mask-only Mask-Predict | RWTH-PHOENIX-WEATHER-2014T · back-translation performance | Helped | G2P-DDM: Generating Sign Pose Sequence from Gloss Sequence with Discrete Diffusion Model (AAAI 2024 · 22 citations) | |
| increase the codebook size | ImageNet-1K reconstruction · rFID | 6.59 -> 5.85 | Helped | An Image is Worth 32 Tokens for Reconstruction and Generation (NeurIPS 2024 · 18 citations) |
| add LLM-generated SVG templates instead of using randomly initialized shapes | Text-to-SVG evaluation · Image FID | 47.21 -> 33.31 | Helped | Chat2SVG: Vector Graphics Generation with Large Language Models and Image Diffusion Models (CVPR 2025 · 15 citations) |
| Add a decoder conditioned on unmasked image pixels | MISATO · Gradient@edge | 63.142 -> 48.588 | Helped | Towards Enhanced Image Inpainting: Mitigating Unwanted Object Insertion and Preserving Color Consistency (CVPR 2025 · 14 citations) |
| Use DiffVG with DBSCAN initialization instead of random initialization | All datasets · MSE | Helped | Optimize & Reduce: A Top-Down Approach for Image Vectorization (AAAI 2024 · 14 citations) | |
| Use superpixel segmentation instead of vectorizing the whole image without segmentation | ImageNet ablation, 1000 paths · MSE | 0.0083 -> 0.0032 | Helped | SuperSVG: Superpixel-based Scalable Vector Graphics Synthesis (CVPR 2024 · 9 citations) |
| add sketch retrieval instead of using contour extraction | DeepCAD test set · Valid (%) | 95.88 -> 99.65 | Helped | Revisiting CAD Model Generation by Learning Raster Sketch (AAAI 2025 · 4 citations) |
| Add camera pose pruning | Synthetic-NeRF Lego scene · generation quality | Helped | CAD: Photorealistic 3D Generation via Adversarial Distillation (CVPR 2024 · 4 citations) | |
| Use 75-word captions instead of no text conditioning | ImageNet 256×256 reconstruction · rFID | 1.49 -> 1.04 | Helped | Language-Guided Image Tokenization for Generation (CVPR 2025 · 3 citations) |
What reviewers will expect
Datasets and baselines named in the 32 papers of this approach from 2025–2026 that have a full-text summary. Leaving out the common ones is a frequent review question.
Datasets
- ImageNet · 5 of 32 papers
- ImageNet-1K · 5 of 32 papers
- COCO · 3 of 32 papers
- Objaverse · 2 of 32 papers
- COCO 2014 · 2 of 32 papers
- CC12M · 2 of 32 papers
- JourneyDB · 2 of 32 papers
Baselines
- VAR · 5 of 32 papers
- DiT-XL/2 · 4 of 32 papers
- MaskGIT · 4 of 32 papers
- MAR · 3 of 32 papers
- LlamaGen-3B · 3 of 32 papers
- DiT · 3 of 32 papers
- LlamaGen · 3 of 32 papers
- LDM · 3 of 32 papers
Recent preprints
Loading recent arXiv preprints…
Who publishes on it
Institutions on the most papers of this approach since 2023 (of 94 papers; from OpenAlex and the papers' affiliations).
- Peking University · 5 papers
- Tencent · 4 papers
- Google · 4 papers
- Carnegie Mellon University · 3 papers
- Tel Aviv University · 3 papers
- National University of Singapore · 3 papers
- Tsinghua University · 3 papers
- Fudan University · 3 papers
- Beihang University · 2 papers
- University of Chinese Academy of Sciences · 2 papers
Use it
Download BibTeX of 39 key papersOpen the topic in the interactive map
Get alerts. Describe your idea in scime atlas and save it (free scime account): you get a weekly email when new preprints land close to it.
Related approaches
Same keyword cluster
- High-resolution diffusion synthesis (140 papers)
- Latent diffusion with learned tokenizers (114 papers)
- Hierarchical VQ-VAE generation (62 papers)
- Next-scale and token autoregressive images (90 papers)
- Accelerating autoregressive image generation (22 papers)
Elsewhere in the topic
- Classical generative image synthesis (147 papers)
- Retrieval and matching with generative features (140 papers)
- Generalizable fake image detection (119 papers)
- Synthetic training data from generators (111 papers)
- Compositional scene generation (105 papers)
- Robust diffusion watermarking (79 papers)
- Neural text-to-speech synthesis (72 papers)
- Equivariant diffusion for 3D molecules (71 papers)
Benchmark numbers, ablations and baseline numbers are extracted by AI from each paper's full text and can be wrong; the paper is the authority. Data: results index of 2026-11-02, citation counts from OpenAlex (2026-10-09).