AnimateDiff + Instant Lora: ultimate method for video animations ComfyUI (img2img, vid2vid, txt2vid)

Koala Nation
24 Oct 202311:03

TLDRThis tutorial showcases the powerful combination of AnimateDiff and Instant Lora for creating stunning video animations using ComfyUI. The video guides users through the process of setting up and utilizing custom nodes and models, including the IPA adapter nodes and models, to achieve seamless animations. The workflow involves downloading poses, installing necessary models, and using the AnimateDiff evolved model for high-quality results. The Instant Lora method is highlighted for its ability to generate animations without training, offering endless creative possibilities. The video demonstrates the creation of an animation with detailed facial features using the face detailer tool and concludes with tips for further post-processing to refine the final video.


  • 🎨 Use ComfyUI with custom nodes and models manager for animation creation.
  • 📚 Install additional nodes and models through the ComfyUI manager for both AnimateDiff and Instant Lora methods.
  • 🔍 Download and prepare poses from a provided link, and save them in the input folder for later use.
  • 🖼️ Save your Instant Lora image in the input folder and ensure it uses the same model as the one in the video.
  • 🔄 Install AnimateDiff Evolved and additional models for animation, such as the stabilized High model.
  • 🎥 For the Instant Lora method, use the IP adapter nodes and models, including the IP adapter plus sd15 bin model.
  • 🤖 Install the Control Net model, specifically the open pose model, to run the poses.
  • 📈 Adjust the workflow by connecting nodes correctly and setting the right model and sampler parameters.
  • 📹 Use the Face Detailer to enhance the facial details of the animation.
  • 🔗 Convert the batch of images to a list for processing with Face Detailer and then back to a batch if needed.
  • 🎉 Combine all elements to generate a new animation with improved details and character transformation.

Q & A

  • What are the two main methods discussed in the video for creating video animations?

    -The two main methods discussed are AnimateDiff and Instant Lora, which are used to create video animations using ComfyUI.

  • What is required to be installed in ComfyUI for the instant Lora method?

    -For the instant Lora method, you need the IPA adapter nodes and models. The easiest way to install these is by using the ComfyUI manager.

  • What is the purpose of the 'control net' in generating poses for the animation?

    -The control net is used to generate poses, depth maps, line art, or other control net methods. It is essential for creating the initial poses that are used in the animation workflow.

  • Which model is used for the animation in the video?

    -The video uses the 'geminix mix model' for the animation. This model should be downloaded and copied into the checkpoint folder within ComfyUI.

  • How can one improve the general definition of the animation?

    -To improve the general definition of the animation, the output of the animated fifth loader is connected to the input of the FREU node, and the output of FREU is connected to the case sampler.

  • What is the role of the 'face detailer' in the animation process?

    -The face detailer is used to enhance the face details of the animation. It requires converting the batch of images to a list of images before it can be applied effectively.

  • How can one convert the original Runner to a new character using AnimateDiff and the instant Lora method?

    -By following the steps outlined in the video, which include installing necessary models and nodes, setting up the workflow in ComfyUI, and using the instant Lora method to apply the new character's features to the animation.

  • What additional effects can be introduced to the animation using optional models?

    -Optional models such as Luras and Anime diff can introduce camera effects like zoom and pan to the animation, adding more depth and dynamism to the final result.

  • How can one postprocess the video to fine-tune the animation?

    -After generating the initial animation, one can postprocess the video by adjusting various parameters and using additional tools to refine the animation and achieve even more polished results.

  • What is the frame rate used for the final GIF or video generated with AnimateDiff?

    -The frame rate for the final GIF or video is set to 12, as the original video had 25 frames per second, and poses were extracted every two frames.

  • How can one ensure that the images in the input folder are used for the animation?

    -By using a load image node to load the reference image and refreshing the input folder to make sure that the images can be accessed and used during the animation process.

  • What is the importance of using the same model as used in the Lora image?

    -Using the same model as in the Lora image ensures consistency and compatibility between the Lora method and the animation, leading to a more seamless and accurate transformation.



🎨 Introduction to Animation with Stable Diffusion and Instant Laura

This paragraph introduces the viewer to the tutorial on creating animations using stable diffusion and the Instant Laura method. It emphasizes the need for a Comfy UI with custom nodes and models manager, and outlines the basic requirements, which are listed in the description for viewers' reference. The paragraph also explains the necessity of the IPA adapter nodes and models, and guides on how to install them using the Comfy UI manager. The process of creating animations with animate diff is introduced, and the viewer is informed about the installation of additional models and nodes for the animation workflow. The paragraph concludes with instructions on downloading poses and saving the Instant Laura image in the input folder for later use in the Comfy UI workflow.


🚀 Setting Up and Testing the Animation Workflow

The second paragraph delves into the setup process for the animation workflow. It guides the viewer on how to start with a template from the animate diff GitHub, ensuring the load image upload node points to the correct directory and the anime diff model is selected. The viewer is then instructed to run a test prompt with specific sampler settings to verify that the models are loaded and the sampler is operational. The paragraph continues with steps to improve the animation's general definition by connecting nodes and adjusting settings, such as control length and context overlap. It also introduces the Instant Laura method, detailing how to load the reference image and connect it to the IP adapter and other nodes for processing. The viewer is then shown how to generate a new animation with an increased number of frames for better detail and how to use face detailer to enhance facial features. Finally, the paragraph concludes with instructions on converting the batch of images to a list for further processing and generating a GIF or video with animate diff.


🌟 Completing the Animation and Post-Processing

The final paragraph focuses on completing the animation process and suggests post-processing techniques for fine-tuning the video. It guides the viewer on how to process all the poses by setting the image load cap to zero and running the prompt, which will convert the original Runner into a new character using animate diff and the instant Laura method. The paragraph encourages viewers to use their imagination to explore the creative potential of these methods. It concludes by inviting the viewer to check the description for more information on the method and hints at future content.




AnimateDiff is a tool used for creating animations with stable diffusion models. It enhances the capabilities of these models by allowing users to generate animated sequences. In the video, AnimateDiff is used to create a video animation, showcasing how it can transform static images into dynamic sequences, which is a central theme of the tutorial.

💡Instant Lora

Instant Lora refers to a method that enables the use of a Lora model without any training. Lora models are typically used in machine learning to adapt a pre-trained model to a new task. The Instant Lora method, as described in the video, allows for quick and efficient adaptation, which is crucial for the animation process demonstrated.


ComfyUI is a user interface that simplifies the process of working with complex machine learning models and tools. In the context of the video, ComfyUI is used with custom nodes and a models manager to facilitate the animation creation process. It serves as the primary interface through which the user interacts with the various tools and models to achieve the desired animations.

💡Custom Nodes

Custom nodes are user-defined components in ComfyUI that can be used to extend the functionality of the interface. They are essential for the workflow described in the video, as they allow for the integration of specific models and tools needed for animation creation. Custom nodes provide a way to tailor the ComfyUI to the user's specific needs.

💡IP Adapter Nodes

IP Adapter Nodes are specific custom nodes used in the video for adapting the input to the requirements of the Lora model. They play a crucial role in the Instant Lora method by ensuring that the input data is correctly formatted for the model to process. In the context of the video, they are used to load a reference image and connect it to the Lora model.

💡Control Net

Control Net is a tool used for generating poses, depth maps, line art, or other control methods. It is mentioned in the video as a way to create custom poses for the animation. Control Net is important because it allows for the creation of specific and detailed poses that can then be used in the animation workflow.

💡Anime Diff

Anime Diff is a model used for generating animations. It is part of the animation creation process described in the video. Anime Diff is used in conjunction with AnimateDiff to produce the final animated video. It is an optional component that introduces camera effects such as zoom and pan to enhance the animation.

💡Face Detailer

Face Detailer is a tool used to enhance the details of faces in images. In the video, it is used to improve the facial details of the animation. By converting a batch of images to a list and then applying Face Detailer, the video's character faces become more detailed, which is important for creating a high-quality and realistic animation.

💡Video Combine

Video Combine is a node or tool used to combine individual images into a video format. In the context of the video, it is used after applying Face Detailer to compile the enhanced images into a GIF or video. This is a crucial step in the animation process, as it finalizes the animation by creating a连贯的 video sequence.

💡Frame Rate

Frame rate refers to the number of frames displayed per second in a video. The video script mentions changing the frame rate to 12 to match the original video's timing. This is an important aspect of video production, as it affects the smoothness and speed of the animation. In the video, adjusting the frame rate is part of the final steps to ensure the animation's timing is correct.

💡Frame Interpolation

Frame interpolation is a technique used to increase the frame rate of a video by inserting new frames between existing ones. In the video, it is mentioned as a method to return to the original 25 frames per second after extracting poses every two frames. This technique helps to maintain the fluidity and smoothness of the animation.


AnimateDiff and Instant Lora are combined for video animations using ComfyUI.

ComfyUI with custom nodes and models manager is required for the process.

IP adapter nodes and models are needed for the Instant Lora method.

AnimateDiff allows creating animations with stable diffusion.

Instant Lora method enables the use of Lora without any training.

The combination of these two methods opens up endless possibilities for video animation.

Download poses from the provided link and place them in the ComfyUI input folder.

Save your Instant Lora image in the input folder for the workflow.

Use the same model as used in the Lora image for consistency.

Install all requirements for AnimateDiff and Instant Lora through the ComfyUI manager.

Additional models for AnimateDiff need to be downloaded for different results.

Install custom nodes for the workflow including advanced control net nodes and video helper suite.

Download the Control Net model and the model for AnimateDiff for the animation.

Optional models for camera effects like zoom and pan can be used with Anime Diff.

IP adapter model is necessary for the Instant Lora method in the video.

Clip Vision model for SD 1.5 is also required.

Follow the workflow for text to image with initial control net input using open pose images.

Use the same VAE as the checkpoint loader and connect it to the decoder.

Adjust the prompts and sampler settings to match the reference image.

Use the FREU node to improve the general definition of the animation.

Add a load image node to load your reference image for the Instant Lora method.

Connect the IP adapter and Clip Vision input to the respective nodes for integration.

Use the face detailer to enhance the facial details of the animation.

Convert the batch of images to a list for processing with face detailer.

Revert the image list back to image batch for video combination.

Change the frame rate in the video combine node to match the original video's frame rate.

Post-process the video for fine-tuning and achieving better results.