DWPose for AnimateDiff - Tutorial - FREE Workflow Download

Olivio Sarikas
20 Jan 202417:15

TLDRThis tutorial showcases the impressive capabilities of AI video rendering with DV Pose input, highlighting a stable and smooth animation workflow. The video demonstrates the stability of clothing, hair, and facial movements, with minimal flickering and design inconsistencies. It explains the process of using a dance video as input and adjusting settings for optimal results. The tutorial also delves into the technical aspects of the workflow, including model selection, frame handling, and the use of control nets for consistency. The presenter encourages experimentation with settings and prompts to achieve the best video output, emphasizing the importance of a concise and clear prompt for better AI performance.


  • 😲 The video showcases a highly stable AI-generated animation using DV POS input, demonstrating significant improvements in AI video rendering technology.
  • 🤖 The tutorial is a collaboration with Mato, an expert in AI video rendering, whose channel offers extensive learning resources on the subject.
  • 🎥 The video input used for the demonstration is a dance video from Sweetie High, a popular content creator with over a million followers.
  • 🔧 The workflow allows for customization of video size, frame load cap, and starting frame number to optimize video processing.
  • 📏 The DV POS estimator is a crucial component of the workflow, automatically downloading necessary models for pose estimation.
  • 🧩 The workflow designed by Mato is not overly complex, but it requires careful setup and understanding to achieve the best results.
  • 🔄 The use of a 1.5 model for rendering is recommended due to the time-consuming nature of video rendering, which involves rendering each frame twice for higher quality.
  • 📝 The importance of using the correct models and checkpoints, such as the V3 SD 1.5 adapter checkpoint and the control net, is emphasized for successful animation.
  • 🔢 Experimentation with settings like strength, start percentage, end percentage, and step count is necessary to achieve consistent and high-quality results.
  • 🔍 The tutorial highlights the need for concise and clear prompts when working with AI video rendering to maintain the desired outcome.
  • 🔗 The video transcript provides links to download necessary models and checkpoints, as well as a workflow template for further experimentation.

Q & A

  • What is the main focus of the tutorial video?

    -The main focus of the tutorial video is to demonstrate the use of AI video rendering with DV Pose input to create stable and high-quality animations.

  • Who is Mato and what is his role in this tutorial?

    -Mato is a master of AI video rendering and has collaborated with the creator of the tutorial to develop the workflow for the AI video rendering process.

  • What is the purpose of using a dance video from Sweetie High in the tutorial?

    -The dance video from Sweetie High is used as a video input to demonstrate how the AI video rendering process works with actual footage.

  • What is DV Pose and how is it used in the workflow?

    -DV Pose is a tool used to estimate and apply pose information to the video frames. It is used in the workflow to ensure stability and smooth movement in the animation.

  • What is the significance of the 'frame load cap' setting in the video input?

    -The 'frame load cap' setting determines the number of frames the workflow will process. It helps in managing the workload and can be used to skip certain frames or select every nth frame.

  • How does the 'batch prompt schedule' work in the workflow?

    -The 'batch prompt schedule' allows for the input of multiple prompts, each associated with a specific frame number. This helps in creating a sequence of different prompts for the AI to follow throughout the animation.

  • What is the role of the 'V3 SD 1.5 adapter checkpoint' in the animation?

    -The 'V3 SD 1.5 adapter checkpoint' is used to control the strength of the animation. It is crucial for ensuring that the animation is smooth and consistent.

  • Why is it recommended to render the video twice in the workflow?

    -Rendering the video twice helps in improving the quality of the animation. The first rendering establishes the overall look, while the second rendering refines the details and fixes any inconsistencies.

  • What is the purpose of the 'uniform context options' in the workflow?

    -The 'uniform context options' are used when the number of frames exceeds the maximum renderable limit by 'Animate Diff'. It sets up the rendering in batches with an overlap to ensure stylistic consistency across the entire animation.

  • How does the 'K sampler' setting affect the rendering process?

    -The 'K sampler' setting determines the number of steps and the CFG scale used in the rendering process. It is important to experiment with these settings to achieve the desired quality and consistency in the animation.

  • What are the additional steps taken by Mato in the video combiner to enhance the animation?

    -Mato uses additional steps such as sharpening and interpolation in the video combiner. Interpolation adds extra frames to make the animation flow smoother, while sharpening enhances the visual clarity of the final output.



🎨 Introduction to AI Video Rendering with Stable Diffusion

The video script introduces a tutorial on creating stable AI videos using diffusion technology. The presenter collaborates with Mato, an expert in AI video rendering, and invites viewers to explore Mato's educational channel. The goal is to achieve high stability in animations, with a focus on clothing, movement, hair, facial features, and background details. The script acknowledges minor imperfections such as flickering and melting hands but emphasizes the overall improvement in video quality. A dance video from Sweetie High is used as a practical example, and the workflow setup for the AI video rendering process is discussed, including the use of a DV pose estimator and model settings.


🔧 Detailed Workflow and Technical Settings for AI Video Rendering

This paragraph delves into the technical aspects of the AI video rendering workflow. It discusses the use of the Dream Shaper 8 model for its capability in handling video rendering, and the importance of using a 1.5 model to ensure quality. The script explains the process of setting frame numbers, using batch prompts, and the significance of the V3 SD 1.5 adapter checkpoint for animation consistency. It also describes the use of the uniform context options for rendering more than 16 frames, the animated div loader, and the importance of experimenting with settings like the batch size, step count, and CFG scale for optimal results. The paragraph concludes with a mention of the second case sampler and VA decode for improving the quality of the final video.


🖼️ Applying DV POS in AI Video Workflow and Model Selection

The third paragraph focuses on the application of the DV POS estimator in the AI video workflow. It provides instructions for loading a video from a path, adjusting settings for size and starting frames, and emphasizes the importance of using the correct model, specifically the control.v1p.sd15 open pose file. The script also discusses the need for experimentation with strength and percentage values to achieve the best video results. Additionally, it guides viewers on installing missing custom nodes using the manager for comi and restarting the application for changes to take effect.


🌟 Finalizing AI Video Rendering and Experimentation Tips

The final paragraph discusses the final steps in the AI video rendering process, including bypassing certain notes and directly connecting elements to positive and negative prompts to eliminate the impact of the control net. It provides tips for experimentation, such as adjusting the strength of the model and clip, and varying the steps and CFG scale in the K sampler. The script also mentions the importance of keeping prompts simple and clear for better results. It concludes with a suggestion to download a video template from OpenArt, experiment with prompts and settings, and adapt them to the specific video content. The presenter expresses excitement about the video quality and stability and invites viewers to share their thoughts in the comments.




DWPose is a technology used in the video to estimate and apply poses to a video input. It's crucial for creating animations that are stable and smooth. In the video, DWPose is used to enhance the animation quality, making the clothing, hair, and face movements appear more natural and less flickering.

💡AI video rendering

AI video rendering refers to the process of using artificial intelligence to generate or enhance video content. In the context of the video, AI video rendering is highlighted as being 'crazy good', indicating significant advancements in the field. It's used to create stable and high-quality animations from video inputs.


AnimateDiff is a software or tool mentioned in the video that seems to be used for rendering animations. It's integral to the workflow described, as it helps in creating smooth animations with less flickering and more detail, as showcased in the video examples.

💡Video input

The term 'video input' is used to describe the source material for the animation process. In this video, a dance video from 'Sweetie High' is used as the video input. It's the base that the AI uses to create the stable and detailed animations.

💡Frame rate

Frame rate in the video refers to the number of frames displayed per second in a video. The script mentions setting a frame rate of eight in the video combiner to control the speed of the animation. A lower frame rate can make the animation appear slower.

💡Batch prompt

A batch prompt is a series of prompts used in the animation process to guide the AI in rendering the video. The video script discusses using a batch prompt schedule, which involves setting different prompts for different frames to create a sequence of animations.

💡Dream Shaper 8

Dream Shaper 8 is mentioned as a model used in the animation workflow. It's described as a '1.5 model', which might suggest it's a version or iteration of a model that is capable of handling video rendering, a process that can be time-consuming due to the need to render multiple frames.

💡Control net

The control net is a model used in the workflow to maintain consistency between different stages of the animation process. It's applied after the initial rendering to improve the quality and ensure that the animation remains stable and coherent.


A checkpoint in the context of the video refers to a saved state of a model that can be loaded for specific tasks. The script mentions several checkpoints, such as 'V3 SD 1.5 adapter checkpoint' and 'control, v1p, sd15 open post, fp16 save tenser file', which are crucial for the animation process.


Interpolation in the video script refers to a technique used to add additional frames to the animation to make it smoother. It's one of the final steps in the workflow described by the video creator, Mato, to enhance the fluidity of the animation.


Experimentation is a recurring theme in the video, emphasizing the need to adjust settings and prompts to achieve the best results in AI video rendering. The script suggests that finding the right balance of settings requires testing and iteration, as no one-size-fits-all solution exists for all video inputs.


Introduction to a stable AI video with DV POS input, showcasing improved stability and quality in AI video rendering.

Collaboration with Mato, an expert in AI video rendering, offering a wealth of learning resources on his channel.

Demonstration of a beautiful animation with stability in clothing, smooth movement, hair, face, and background details.

Achievement of no flickering in the animation, a significant improvement from past techniques.

Acknowledgment of some design changes and minor issues like hands melting into the body due to rushed processing.

Second example created by Mato showing consistent results in clothing, hair, background, and face morphing.

Explanation of the need for a video input and the use of a dance video from Sweetie High as an example.

Description of the workflow node differences and settings for video input, size, and frame load cap.

Introduction of the DV pose estimator and its setup for automatic model downloading.

Mato's workflow overview, emphasizing its complexity and the value of his hard work.

Use of Dream Shaper 8 model for video rendering and the rationale for choosing a 1.5 model.

Explanation of the batch prompt schedule and its role in the video rendering process.

Importance of the V3 SD 1.5 adapter checkpoint for animation consistency.

Details on the uniform context options for handling more than 16 frames in rendering.

Discussion on the use of the animated div loader and the significance of the V3 sd15 mm checkpoint model.

Advice on experimentation with settings like batch size, image size, and K sampler for optimal results.

Mention of the second case sampler and VA decode for improving the quality of the animation.

Explanation of the anidi control net checkpoint and its role in maintaining consistency between renders.

Final thoughts on the workflow's ability to produce high-quality, stable AI video animations.