Video specialty models

An older but still relevant technique that adds animation capabilities to existing image models. AnimateDiffAnimateDiffAn early, widely used community extension that adds motion to image models to produce short video, notable as a custom node rather than a natively supported model. was the dominant way to do AI video generation in late 2023 before the dedicated video models matured. It is still used in some workflows, particularly those that benefit from the rich LoRALoRA (Low-Rank Adaptation)A fine-tuning technique that adapts a base model to a specific style or concept using a small additional file. The standard way to customize open-source models. ecosystem of Stable Diffusion 1.5 and SDXL.

These models take a video and insert new frames between the existing ones to increase the frame rate or smooth out motion. They are not generative in the same sense as the main video models, they do not invent new content, but they are essential for converting low-frame-rate AI-generated video into smooth-looking final output. RIFE and FILM are the most widely used.

Topaz Video AI is the dominant tool here. It can upscale low-resolutionResolutionThe pixel dimensions of an image or video, for example 1024x1024. Higher resolution shows more detail but costs more memory and time. video, deinterlace old footage, increase frame rates, and reduce noise. Used heavily in archival restoration work and as a final polish step on AI-generated video.