I don't even know where to start but I guess a introduction to the topic: I wanted to see if AI video generation from your own PC has reached a usable state, and yes it has and it's sort of ignored by most bot AI slop dispencers and artist.
In this post I want to talk about the programs internals and all different AI all open source. Mainly for long form AI videos
The programs: A best way to start is introduce the open source programs used and what they are used for and how.
ComfyUI: Most widelly know, hated and accept mostlly as the best alternative to writing your own code. It's Scratch code for ai except 100 times more complicated breaks all the time and is slow cause it runs on Python. Their is no good tutorial no guide and because AI has become a tabu not a lot of usefull advice.
it can genrate everything except you need to instruct it manualy
Example: You set up the loaders Val- audio Val-Video The model Crop for what video clip generation. Ksamplers, Text promp Plus and minus boxes, Image refernce or a make a entire Image generator form text that transfers the Image to the Image reference that then gets transfered to the video that then uses references from the training the model you picked was trained on to make a video as accurate to the video text prompt instruction given.
This is a very very simple ComfyUI Text to Video explanation. It would take about a hour to set it up if you knew what you were doing.
Mago.Studios: A studio for the industry it uses referenes like poses videos already recorded blender animations and stuff like that. To use it you already need to be able to animate in blender it's basically a cote of paint on top of gray figures.I haven't seen impressive things be done with it cause it's hard to use set up etc. And it's not made for the common person. it shows with all the options.
And it's obviuslly a subscription service reverse engineered to be Open source
LTX Desktop: Simple to use limited just basic stuff. Comerial.
Maestro on Pinokio - It's probably the best because it dosen't melt your PC it has clear UI that you can understand. Can use all the AI models like ComfyUI but it's way way simpler.
Takes a long time. A lot of options for a lot of things complicated. it can do everything. A good middle option between comfyUi and the others
The AIs: Different AIs do different things almost all open source are chinese.
Main LLm to write with you is Qwen3.8 27B for this. Does the job best it's the most modern open source option. Gemma 4 31B.
Their is also Underdog Saluki that is a distilled Qwen 3.8 version making it faster and 7.8 GB from the 54GB it is. It makes it 20% lower quality when scored but that depends for what you will be using it for.
(I understand most people don't understand wht the B means what a distilled AI is or what I am talking about the basic is B is like procesing power. Different AIs have different processors so bigger isn't better if the AI you have with 1 Trillion Bs is made for cooking. Distilled is a guy grabbed the AI and quantified it and that is a whole rabbit hole. Best i can say bigger isn't better here. It depends how it's used)
MiniMax H3 - The big boy of open source video generation it compares to all the big boys you usually pay a subscription for. It's open source it's a good over all option.
LTX- focused on cynematics and more profesional videos. It's consistent and made for cynematics
Wan - it's alibaba's video generator and they use a shotgun aproach on Ai from tinny ais to big models. It depends best describes it.
You will be using Minimax for everything.
Images:
Flux or Moss
Moss is really good if writing is involved
Flux is the basic do stuff model.
Their is also qwen but it's really just there. For images their is a lot of models I can't really explain all of them.
Requirements: Most just need a minimum of 8GB of Vram. It's gonna take a long time no mater what you use. It's gonna break no mater what you use. It's gonna take you knowing the basics of animation writing and all that no mater who said otherwise.
A lot of the options are set up to work with reference not make stuff up.
My opinion of AI: It's alright. People who talk about it don't use it to understand it. Even the AI bros just google browser generators that existed way before all this. The bottom line is I ain't paying for your subscription.
If you want to know more. Brows Hugging face. This is a tech that is from the 80s we have been using it for a long time and their is a lot of things.