Make AMAZING AI Animation with AnimateLCM! // Civitai Vid2Vid Tutorial

This paragraph introduces the video tutorial by Tyler from cai.com, focusing on the Animate LCM (Latent Conditioned Mesh) video to video workflow. The workflow allows users to perform style transfer on existing videos using either text prompts or reference images from the IP adapter. It is mentioned that the process requires at least 10GB of VRAM and provides a warning for users with less. Tyler also gives a shoutout to community members Sir Spence and PES Flows for their contributions and provides Instagram handles for further exploration.

In this section, Tyler explains the video upload node, emphasizing the importance of the frame load cap, skip first frames, and select every settings. Frame load cap determines the number of frames to be rendered, skip first frames allows skipping a certain number of frames at the start, and select every dictates the rendering frequency. He also discusses the aspect ratio and resolution settings, recommending a vertical format for social media optimization. The paragraph concludes with an introduction to the Laura stacker, including model strength and clip strength settings.

This paragraph delves into the model loader settings for the animate diff motion model and the VRAM usage. Tyler highlights the need for the animate LCM motion model and provides a link for downloading it. He also discusses the use of a photon LCM specific model trained by a community member, Machine Delusions. The paragraph further explains the control net settings, including line art, soft edge, depth, and open pose options, and the necessity of having the control net models installed for their use.

Tyler introduces the IP (Image Prompt) adapter, which uses reference images to guide the animation style. Users can upload images or select them from the browser, and the system will attempt to incorporate the style into the animation. The paragraph explains the crop position selector and weight settings for the IP adapter, allowing users to focus on specific parts of the image for style transfer. It also mentions the need for the IP adapter plus stable diffusion 1.5 bin file and the sd15 pytorch model bin for the process to work.

This section covers the use of positive and negative prompts to guide the video's content and style. Tyler explains the syntax for prompt traveling, where specific prompts can be applied at different frames of the video. The paragraph details the use of the batch prompt scheduler, where the pretext and batch prompt boxes dictate the order of descriptors. It also touches on the use of embeddings and textual inversions for negative prompts to exclude unwanted elements from the render.

Tyler discusses the K sampler settings, emphasizing the need for low steps and CFG values in the LCM workflow. The paragraph explains the importance of the sampler being set to LCM and the scheduler set to sgm uniform for optimal results. The highres fix script, or upscaler, is introduced, with specific settings provided for upscaling the video to a higher resolution. The paragraph also mentions the reactor face swapper, but advises that it may be problematic to install and suggests skipping it if issues arise.

The final paragraph focuses on the output stage of the workflow, where Tyler explains the video combine node and its settings. He demonstrates the successful application of the workflow using a dancer girl video and reference images, achieving a looped video with the desired style. The paragraph concludes with a mention of the preview gallery, where users can compare different versions of their videos during the testing and iterating process. Tyler encourages viewers to share their creations and provides a link to his cai.com profile for downloading the workflow.