Research map / Video / Video generation
Autoregressive long video generation
Generates long videos autoregressively with continuous tokens, propagation and rich perception, plus evaluation suites.
156 accepted papers at ICML, NeurIPS, ICLR, CVPR and AAAI (2016–2026) · an approach in Video generation, keyword cluster “video generation · video diffusion · text video”
State History Tried Checklist Preprints Who Export
Where it stands
Not enough benchmark data Fewer than four papers report the same benchmark, so there is no record to follow.
13 papers in 2023–24 and 130 in 2025–26 (at ICML, NeurIPS, ICLR, CVPR and AAAI); its share of the topic grew, from 5.7% to 11.6%.
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 · Diversified Dynamical Gaussian Process Latent Variable Model for Video Repair (AAAI 2016 · 1 citation)
- 2018 · MoCoGAN: Decomposing Motion and Content for Video Generation (CVPR 2018 · 1,116 citations)
- 2019 · Video Generation From Single Semantic Label Map (CVPR 2019 · 99 citations)
- 2020 · Scaling Autoregressive Video Models (ICLR 2020 · 14 citations)
- 2021 · A Good Image Generator Is What You Need for High-Resolution Video Synthesis (ICLR 2021 · 66 citations)
- 2022 · Generating Videos with Dynamics-aware Implicit Generative Adversarial Networks (ICLR 2022 · 27 citations)
- 2023 · MAGVIT: Masked Generative Video Transformer (CVPR 2023 · 118 citations)
- 2024 · VBench: Comprehensive Benchmark Suite for Video Generative Models (CVPR 2024 · 351 citations)
- 2025 · DepthCrafter: Generating Consistent Long Depth Sequences for Open-world Videos (CVPR 2025 · 51 citations)
- 2026 · GenVidBench: A 6-Million Benchmark for AI-Generated Video Detection (AAAI 2026 · 2 citations)
Earliest papers here
- Diversified Dynamical Gaussian Process Latent Variable Model for Video Repair (AAAI 2016 · 1 citation)
- MoCoGAN: Decomposing Motion and Content for Video Generation (CVPR 2018 · 1,116 citations)
- Stochastic Video Generation with a Learned Prior (ICML 2018 · 315 citations)
What has been tried
53 papers here report 549 measured changes (a component added, removed or swapped, compared with a baseline): helped 73%, no clear effect 3%, hurt 21%, 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 Stage 2 temporal-layer fine-tuning to Stage 1 | Sintel · AbsRel | 0.322 -> 0.316 | Helped | DepthCrafter: Generating Consistent Long Depth Sequences for Open-world Videos (CVPR 2025 · 51 citations) |
| Replace default StyleGAN-v motion codes with LSTM codes | Sky Time-lapse, 128 frames · FVD128 | 190.82 -> 172.71(−18.11) | Helped | On the Content Bias in Frechet Video Distance (CVPR 2024 · 28 citations) |
| Filter the Koala data using VTSS instead of using all 48M data | VBench · Total Score | 0.7140 -> 0.7156 | Helped | Koala-36M: A Large-scale Video Dataset Improving Consistency between Fine-grained Conditions and Video Content (CVPR 2025 · 17 citations) |
| Use the human appearance text-video alignment filter | Human video generation test set · Face Consistency | 0.7976 -> 0.8674 | Helped | OpenHumanVid: A Large-Scale High-Quality Dataset for Enhancing Human-Centric Video Generation (CVPR 2025 · 11 citations) |
| use frame-wise line-matching and pose conditions | Frame Gap = 23 · FVD | 536.5 -> 437.9 | Helped | Generative Inbetweening through Frame-wise Conditions-Driven Video Generation (CVPR 2025 · 10 citations) |
| Add few-shot scoring | Video quality · Average Spearman correlation | 0.561 -> 0.620 | Helped | Video-Bench: Human-Aligned Video Generation Benchmark (CVPR 2025 · 9 citations) |
| add StoryFlow Adapter | PororoSV story visualization · FID | 14.84 -> 13.61 | Helped | ContextualStory: Consistent Visual Storytelling with Spatially-Enhanced and Storyline Context (AAAI 2025 · 8 citations) |
| scale the model from 2B to 8B parameters | Text-to-video generation · visual structure quality | Helped | Goku: Flow Based Video Generative Foundation Models (CVPR 2025 · 8 citations) | |
| Train with BTT instead of pointwise regression | VideoGen-RewardBench · overall accuracy w/ ties | 58.39 -> 61.50 | Helped | Improving Video Generation with Human Feedback (NeurIPS 2025 · 6 citations) |
| Use 2D decoder regularization instead of 2D encoder regularization | Webvid-Val · PSNR | 26.0 -> 28.0 | Helped | CV-VAE: A Compatible Video VAE for Latent Generative Video Models (NeurIPS 2024 · 6 citations) |
| add the Mask Prediction Decoder (MPD) | Challenging Test Set · CLIP-T | 0.3252 -> 0.3316 | Helped | Generative Video Propagation (CVPR 2025 · 4 citations) |
| Use joint image-video training instead of video-only training | VBench · total score | −2.54 | Helped | SnapGen-V: Generating a Five-Second Video within Five Seconds on a Mobile Device (CVPR 2025 · 4 citations) |
What reviewers will expect
Datasets and baselines named in the 57 papers of this approach from 2025–2026 that have a full-text summary. Leaving out the common ones is a frequent review question.
Datasets
- UCF-101 · 5 of 57 papers
- Panda-70M · 4 of 57 papers
- Kinetics-600 · 4 of 57 papers
- Pexels · 3 of 57 papers
- JourneyDB · 3 of 57 papers
- DynamicReplica · 2 of 57 papers
- DAVIS · 2 of 57 papers
- RealEstate10K · 2 of 57 papers
Baselines
- CogVideoX-5B · 7 of 57 papers
- Emu3 · 5 of 57 papers
- CausVid · 5 of 57 papers
- HunyuanVideo · 4 of 57 papers
- Kling · 4 of 57 papers
- Self Forcing · 4 of 57 papers
- SVD · 3 of 57 papers
- OpenSora-V1.2 · 3 of 57 papers
Recent preprints
Loading recent arXiv preprints…
Who publishes on it
Institutions on the most papers of this approach since 2023 (of 143 papers; from OpenAlex and the papers' affiliations).
- The Chinese University of Hong Kong · 7 papers
- ByteDance · 7 papers
- Tsinghua University · 7 papers
- Carnegie Mellon University · 5 papers
- Tencent · 5 papers
- Fudan University · 4 papers
- Hong Kong University of Science and Technology · 4 papers
- University of Science and Technology of China · 4 papers
- The University of Hong Kong · 4 papers
- Shanghai Jiao Tong University · 4 papers
Use it
Download BibTeX of 33 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
- Efficient video diffusion models (283 papers)
- Compositional text-to-video diffusion (159 papers)
- Trajectory-guided motion control (184 papers)
- World-model video generation (110 papers)
Elsewhere in the topic
- Text-to-motion diffusion (140 papers)
- Diffusion-based human image animation (126 papers)
- Diffusion-based video editing (92 papers)
- Talking face generation (81 papers)
- Video-to-audio synthesis (76 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).