Furkan Gözükara

1 Click Auto RunPod Installer For Rerender A Video - 1 Click Video To Anime

Added 2023-10-15 22:56:10 +0000 UTC

Join discord and tell me your discord username to get a special rank : SECourses Discord

5 February 2024 Update:

A tutorial video published : https://youtu.be/cVf9Qf_pKks
Please also upvote this Reddit Thread I appreciate that
Go to your RunPod : https://runpod.io?ref=1aka98lq
Select template RunPod Pytorch 2.1.1
Customize deployment and set volume disk size to 50
After pod started connect to jupyter lab interface
Default downloaded models improved
Default selected source model set as MeinaMix V11
Download attached installer.sh and upload into the workspace folder
Run below command (copy paste into new terminal).
chmod +x installer.sh
./installer.sh

The above command will clone ReRender repo and install all of the requirements fully automatically for you. It will even download compiled ebsynth so you will not wait its compile.

It will also auto download modified to run on RunPod Rerender A Video web UI as well.

Moreover installer sh file will download the below model files with their proper sd_model_cfg py file so you can select them from web UI dropdown and use directly.

model_dict = {

'Realistic Vision V6' : 'models/Realistic_Vision_V6.safetensors',

'Flat 2D Animerge': 'models/Flat_2D_Animerge.safetensors',

'MeinaMix V11': 'models/MeinaMix_V11.safetensors'

}

If you also want to add more models you can download them with wget and and add them to sd_model_cfg py file. Look at the installer . sh file to see how to download from CivitAI or from Hugging Face.

After installation you need to run below code to start ReRender web UI.

It will give you a Gradio share link use it. You can turn off your Pod start again and directly use again with below command (copy paste into new terminal).

Don't forget chmod +x of ebsynth

export HF_HOME="/workspace"
chmod +x /workspace/Rerender_A_Video/deps/ebsynth/bin/ebsynth
cd /workspace
cd Rerender_A_Video
source venv/bin/activate
python webUI.py

Select video and click render first frame. Since it will upload video until you get correct first frame try again and again and watch the command line interface messages.

Moreover make your video FPS like 24 25 30. Do not make it like 23.976fps

You can use ffmpeg to re-encode FPS like below code

ffmpeg -i input.mp4 -vf "fps=24" -c:v libx264 -crf 7 -c:a copy larry_24fps.mp4

Make your video minimum 2 seconds. You can test example attached 2_sec_example.mp4 file

To see all process edit video_blend py file and set OPEN_EBSYNTH_LOG = True

The result will be inside /workspace/Rerender_A_Video/result/your_uploaded_video_file_name folder

The file name will be blend.mp4

The progress will be slow and it won't show messages

You can check result folder changes to follow

First I suggest you to test 2 second clip then do full clip with same settings

Watch your VRAM usage too

Control type: Canny works better than Hed option

Bigger resolution will significantly improve quality like 768px

You can also play with Canny threshold to get more accurate output

Finally it will not copy the sound of the original video

Put output render video and initial original video into the same folder and run below command.
Don't forget to change file names. Also you need to have installed (add system path) or downloaded ffmpeg into the same folder
ffmpeg -i blend.mp4 -i source_video.mp4 -c:v copy -c:a copy -map 0:v:0 -map 1:a:0 output.mp4