Research map / Video / Video generation
Temporally Consistent Transformers for Video Generation
Wilson Yan, Danijar Hafner, Stephen James, Pieter Abbeel
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 |
|---|---|---|---|
| Use TECO instead of Perceiver AR | video prediction · DMLab | FVD: 96 -> 48 | Improved |
| Use TECO instead of Perceiver AR | video prediction · Minecraft | FVD: 76 -> 116 | Hurt |
| Use TECO instead of Perceiver AR | video prediction · Habitat | FVD: 164 -> 73 | Improved |
| Use TECO instead of Perceiver AR without top-k sampling | video prediction · Kinetics-600 | FVD: 1022 -> 799 | Improved |
| Replace VQ posterior with OneHot posterior (ablation) | TECO · Something-Something-v2 (16 frames) | FVD: 189 -> 199 | Hurt |
| Replace VQ posterior and MaskGit prior with OneHot posterior and Block AR prior (ablation) | TECO · Something-Something-v2 (16 frames) | FVD: 189 -> 209 | Hurt |
| Replace VQ posterior and MaskGit prior with OneHot posterior and Independent prior (ablation) | TECO · Something-Something-v2 (16 frames) | FVD: 189 -> 228 | Hurt |
| Replace VQ posterior with Argmax posterior (ablation) | TECO · Something-Something-v2 (16 frames) | FVD: 189 -> 336 | Hurt |
| Replace MaskGit prior with Independent prior (ablation) | TECO · Something-Something-v2 (16 frames) | FVD: 189 -> 220 | Hurt |
| Replace MaskGit prior with Autoregressive prior (ablation) | TECO · Something-Something-v2 (16 frames) | FVD: 189 -> 207 | Hurt |
| Remove conditional encoding (ablation) | TECO · Something-Something-v2 (16 frames) | FVD: 189 -> 208 | Hurt |
| Vary codebook size from 64 to 4096 (ablation) | TECO · Something-Something-v2 (16 frames) | FVD: 64: 191; 256: 195; 1024: 186; 4096: 200 | Mixed |
Closest papers by meaning
- MilliVid: Adaptive Latents for Long-Range Consistency in Video Generation (NeurIPS 2026 · similarity 0.91)
- Snap Video: Scaled Spatiotemporal Transformers for Text-to-Video Synthesis (CVPR 2024 · similarity 0.89)
- WorldWeaver: Generating Long-Horizon Video Worlds via Rich Perception (NeurIPS 2025 · similarity 0.88)
- SV4D: Dynamic 3D Content Generation with Multi-Frame and Multi-View Consistency (ICLR 2025 · similarity 0.87)
- StreamingT2V: Consistent, Dynamic, and Extendable Long Video Generation from Text (CVPR 2025 · similarity 0.87)
- ARLON: Boosting Diffusion Transformers with Autoregressive Models for Long Video Generation (ICLR 2025 · 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.
