A toolbelt form image, video and audio manipulation

This tool is a command line client and a server to generate images, videos and music in your local computer. It provides a server that runs background tasks and a cli called editorium to manage the tasks on that server.

WIP (Work In Progress)

TODOS
[x] Open the sources
[x] Add CogVideoX basic commands
[x] Add Pyramid Flow basic commands
[x] Add Flux
[x] Add Stable Diffusion 1.5
[x] Add SDXL [x] Add SD 3.5 [x] Add Omnigen [ ] Replace ffmpeg command by builtin scripts that manipulates the video.
[ ] Create documentation explaining each command

Requirements

You need to install the following components at your ubuntu operating system:

• Docker
• Nvidia runtime for docker
• ffmpeg (only if you want to use editorium ffmpeg commands)

Installation

sudo apt-get update && apt-get install pipx
pipx install --force git+https://github.com/rodjjo/editorium.git

Server

Before running the cli, it's necessary to start the server. It's required to define the following paths when we start the server:

• --models-dir - Path where the models are stored. In this directory will be the olfline models, not managed by hugging face.
• --path= - Path of the root directory of the server. This is going to be the output directory and also where the files the server uses will be.
• --cache-dir - It's where the cache directory will be located. It's detault to ~/.cache and it's necessary for hugginface being able to cache the files.

Starting the server:

editorium-cli server run --docker-image --models-dir=~/my-offline-models --path=~/my-awesome-server

Prompt file format

To start you can create a prompt file anywhere,
however the image paths should be relative to the root directory of the server
for example: ~/my-awesome-server (the root path you defined to start the server).

The file format for Cogvideo:

#start
#cogvideo.steps=50
#cogvideo.seed=-1
#cogvideo.num_videos_per_prompt=1
#cogvideo.loop=false
#cogvideo.generate_type=i2v
#cogvideo.should_upscale=false
#cogvideo.stoponthis=false
#cogvideo.strength=55
#cogvideo.count=1
#cogvideo.quant=false
#cogvideo.use_pyramid=false
Your prompt goes here.
#image
#comment The path bellow should allways tart with /app/output_dir and it's the root of our service 
/app/output_dir/<relative_path_to_the_image_or_Video
#end

The file format for Pyramid Flow:

#start
#pyramid.generate_type=i2v
#pyramid.num_inference_steps=20,20,20
#pyramid.video_num_inference_steps=10,10,10
#pyramid.height=768
#pyramid.width=1280
#pyramid.temp=16
#pyramid.guidance_scale=9.0
#pyramid.video_guidance_scale=5.0
#pyramid.seed=-1
#pyramid.use768p_model=true
Your prompt goes here.
#image
#comment The path bellow should allways tart with /app/output_dir and it's the root of our service 
/app/output_dir/<relative_path_to_the_image_or_Video
#end

You can add multiples entries starting with #start and finishing with #end, each block is going to be a prompt.

The you can start the cli on the terminal

editorium cogvideo generate-from-file --prompts-path=prompts.editorium

Help

The cli provides help for each command

editorium --help
editorium server --help
editorium server run --help
editorium cogvideo --help
editorium cogvideo generate-from-file --help

References

THUDM/CogVideo
ComfyUI-CogVideoXWrapper
diffusers
Pyramid Flow