Mastering Parody and Meme AI Animation: Step by Step ( パロディとミームAIアニメーションのマスタリング：ステップ・バイ・ステップ )

Hello everyone today. I'll show you how to make Cool A I animations like the ones on my channel including Klee Loli Kami and others. There are six steps but you can skip some for now. I won't talk too much. But remember you can download all the detailed instructions for free in the video description. So let's begin our fun journey into the world of A I animation ready. Let's start before we start the installation, I want to talk about your GPU. If your GPU has less than eight GB of VRAM, it migh

t not handle this tutorial well. (However, there's a cloud GPU option, so don't lose hope!) I recommend a GPU with at least 12 GB VRAM for better results. Even with my powerful RTX 4070 TI with 12 GB VRAM. It takes over 1.5 minutes to generate each image for a 10 seconds video. I generate around 200 to 300 images taking 6 to 15 hours for a 10 to 20 seconds clip. If you don't have a powerful GPU, you have options. You can use a cloud GPU like Google Collab, but it's a bit complicated. Another opt

ion is RunDiffusion, a user friendly cloud service. You can set it up in 10 to 15 minutes. The starting package is 99 cents per hour and it's 1.5 to 3 times faster than my RTX 4070 TI. If you're serious, the monthly Creator's Club package is $35.99. But here's the exciting part. I have a reference link with a discount coupon for 15% off. So it will be $35.99 minus 15% which is $30.59. This package gives you more storage and features and you get a $10 bonus, reducing the cost to $20.59. By the wa

y, if you use my affiliate link to purchase this monthly package, I'll also receive a small commission. And if you plan to buy a new GPU, you can support our channel by using my Amazon affiliate link. It helps our channel grow and doesn't cost you anything extra. Thanks for your support. Now, let's start the installation guide. Welcome to the stable diffusion and extensions. Installation guide. In this brief overview, we'll cover the essential steps for installing stable diffusion and its extens

ions as well as configuring some settings that we'll use in this tutorial. If you're already experienced with these installations, feel free to skip ahead. Step one, install stable diffusion by following the instruction on their github. Step 1.1 optional if you followed the instruction and failed, try downloading the source code and extract it manually. Step 1.2 optional run webui.bat Wait until the installation is finished. If you've successfully installed it, you should run Webui-user.bat next

time. These are my settings for NVIDIA cards to optimize performance. Step two, install all the extensions that required for this tutorial. You can download the list of my extensions in the workflow and files, free download item on my Ko-fi page step three. Download A I models for control net extension. You should download all the PTH files except for the ones containing 'XL'. These are for stable diffusion XL which is not needed in this tutorial step 3.1 download TemporalNet model for control

net extension step four download based model for image generation checkpoint step 4.1 download specialized model for a specific character, image generation. LoRA step 4.2 optional download, CivitAI extension for LoRA thumbnail and more. Step 4.3 optional download and beating for hand generation improvement step five, set multi control net number set at 5 to 7 step 5.1 optional my other settings to set noise multiplier for img2img to 0. You need to modify some files. I'll explain in the future vi

deo. Great job you finished installing Stable diffusion and its extensions. That's it for our basic guide. If you have more questions, feel free to leave a comment before we dive into the tutorial, allow me to provide a quick explanation on how to watch it effectively. One, download my workflow in the video description two at the bottom of each scene, there is a step that we are in indicator with two color yellow is an important part that you cannot miss white is an optional part that you can wa

tch later. If you want to know the detail of how I generate each different type of image, you can come back and watch later after you understand the big picture of the workflow. These are the extensions that you need for this tutorial. This is how you should set the control net unit number. OK. Now that you understand how to watch my tutorial, let's begin A I animation guide step zero, optional remove background from the reference video step one, extract all video frames, Ebsynth stage one, you

don't need the video_mask for this tutorial. You can skip it by editing the Python script. Although it's a bit complex to explain alternatively, you can close the stable diffusion terminal after it has extracted all the video frames and then rerun the app step two extract key frames, sync stage two, I usually extract at least half of the video frames but in some cases, one third is sufficient for my PC generating 200 -300 images for a 10 to 20 seconds video takes about 6 to 10 hours which is acc

eptable to me. Step 2.1 optional, please review all the key frames and identify any potential images that might lead to distorted or poorly generated visuals. Step 2.2 optional, separate a bad reference images from a video_key folder for retouching. Step 2.3 optional organize a folder of reference images. For images need to be fixed. Step 2.4 optional photo edit redraw the potential problem reference images step 2.5 optional finish categorize all potential problem images category step 2.6 find a

general prompt that works with majority of the image poses. Step 2.7 optional test, the prompt with a sampler of each unique pose. This step involves testing different prompts to find the best one for various poses. Reducing the number of poorly generated images in batch image generation. Step 2.8 check all the setting and place the first generated keyframe that you like in TemporalNet. Step three, start batch image generation. This step is crucial you need to ensure that all ControlNet setting

s are in batch mode except for TemporalNet which uses single image mode with the loop back option. It's important to keep the seed fixed after clicking generate, make sure to wait until at least two or three images have been generated before leaving your PC. Any mistakes could result in wasting 6 to 7 hours of work or even more step 4.1 optional, manually edit the bad generated images or generate the separate difficult to generate key frames that have not been generated yet. This is an optional

step but it can significantly improve the video's quality by addressing poorly generated frames. Many A I animation videos skip this step for the Klee Loli Kami character. It took me more than 12 hours to complete. It involves photo editing, adjusting ControlNet parameters and manually adjusting Openpose bones step 4.1.1 optional, try photo edit the ref image by adding an accessory. That's important for the character. You want to make it easier to generate step 4.1.2 optional try in-paint genera

te some small part that you think it's important such as a hand step 4.1.3 optional upscale img2img_key and video_frames to get an upscale video step five, create interpolation frames, Ebsynth stage five, download app from https://ebsynth.com ,drag and drop each .EBS file into the app and click run all. If you haven't extracted the video_mask, please turn off the mask option step six, generate video by combining interpolation frames. Ebsynth stage seven, simply input the project directory and th

e path to the original video. Select Ebsynth stage seven and then click the generate button. I want to mention that I don't generate income from TikTok videos or YouTube Shorts. If you appreciate the content and wish to support my channel. You can do so by watching my longer videos like this tutorial on my youtube channel. You can also show your support by making a donation on my ko-fi page. Additionally, I accept various cryptocurrencies like BTC, ETH including layer two solutions like OP, ARB,

ZK or others Doge BNB XRP or any other cryptocurrencies. You prefer, your support goes a long way in helping me create more valuable content. Step seven optional, apply flow frames to increase videos, FPS, drag and drop the video file into the app, select RIFE(NCNN), set the FPS and speed and click interpolate. Thank you so much for watching my first video on A I animation tutorial. I'm sorry if I couldn't explain everything perfectly this time. It's my first try after all, but I promise to get

better in the future if you have any questions or need more help. Just leave a comment below. I'll try to answer your questions or make a new video to help you out. Let's keep learning and improving together. Thanks for being here with me. See you in the next video.