Early-Signal Detector

model · first seen 5 Oct 2026

video diffusion model

Also: VDM, video diffusion, Stable Video Diffusion, video generation model, diffusion video generation

Generative diffusion models that synthesise video frames conditioned on text or other inputs.

Video diffusion models generate or edit video by iteratively denoising spatiotemporal latents, usually conditioned on text, images or camera signals. The sources show the concept serving mainly as a pretrained backbone that other work builds on. Papers repurpose it for robot world models and action prediction, dense 3D tracking, exocentric-to-egocentric view synthesis, avatars and plant reconstruction. Others work on the generator itself: sparse attention for faster inference, training-free resolution extension, and mixing serial and parallel denoising so physical events stay consistent. GitHub has research code releases, and a Show HN post describes compressing a 31B video-and-audio diffusion model to 14B parameters with INT8 weights.

Video diffusion models have been a standard tool since about 2022. The current activity reflects their use as general priors for 3D vision and robotics, not a new idea, and the weekly counts are flat to slightly falling.

Analyst note (established), 17 Nov 2026.

Mentions per week

01419 May11 Aug10 NovPapersCodeDiscussion

PapersCodeDiscussion

Rank 82 on 14 Nov 2026 · 40 mentions in 28 days · 36 independent groups · papers → code → discussion · first seen 6 weeks ago · acceleration -0.49

Earliest sources

Latest sources

Mentioned together with

low-rank adaptationworld modelworld action modelsparse attentionnovel view synthesisGaussian splatting4D reconstructionvision-language-action modelsim-to-real transferknowledge distillation