recently,ByteDanceReleased a model called AnimateDiff-LightningAnimateDiff-Lightning's model, this model has impressive performance in video generation. With only 4-8 steps of reasoning, it is possible to generate very good quality videos, which is undoubtedly a major technological breakthrough for the video production industry.
Paper address:https://arxiv.org/html/2403.12706v1
The AnimateDiff-Lightning model also works very well with ControlNet, which means thatVideo retouchingthe workflow needs to be upgraded. At the same time, Byte has also introduced the corresponding Comfyui workflow, an open source workflow implementation that makes the AnimateDiff-Lightning model application more appropriate.
It is understood that the AnimateDiff-Lightning model is refined from AnimateDiff SD1.5v2, which contains 1-step, 2-step, 4-step and 8-step refined model versions. Among them, the generation quality of 2-step, 4-step and 8-step models is very good, providing more choices and possibilities for video producers.
When using the AnimateDiff-Lightning model, Byte also recommends using the motionLoRA.LoRA, because motionLoRA produces a stronger motion effect. In order to avoid watermarks, it is recommended to keep the intensity of motionLoRA use between 0.7 and 0.8.
Overall, the AnimateDiff-Lightning model released by Byte brings new possibilities to the video production industry with its powerful video generation capabilities, and also provides more choices and convenience for video producers.