Workflow Getting Started...How Do I Begin?

A

ArunSocket9

Lurker
Joined
Mar 29, 2022
Posts
106
Reaction score
0
I'm interested in creating some photos first, then maybe some videos, both replicating images and videos I already have, but this is all new to me. Not interested in signing up for a paid site yet. Starting with things like feet pics and upskirts. I'm curious to know where could I test out and create photorealistic material?
 
If you aren't interested in signing up for a paid site, the you have the free options, but if you want to create your own locally (just using your PC hardware) you NEED the hardware first, something Nvidia ideally with at least 12gb of VRAM (that isn't a 3060).

Best place to start is using Google Ai to help you along. I learned everything in a few weeks, how to create images, how to make a LORA, and how to make AI character with text to voice function, all through Google AI. You can actually use Google AI to help you build an Ai guide too using the software you're using.

It has that endless patience you need and can describe everything for you to understand, as there's a lot going on to get your head into like Temperatures and secondary Lora's to tailor detail. It can also provide you direct links for Command Prompt to download everything you need, without you having to go find, and accidentally download the wrong version, which does save time.
 
You mean the 10gb model over the 12gb 3080 ti if I remember correctly? The 3080 is fine to start off with, as there's a lot out there available to get you started off with that can fit on an 8gb VRAM GPU, so having the extra 2gb will help out a lot.

You can definitely learn a lot and have some fun with a 3080. It's actually a good place to be, as if you work within your VRAM, you will either know after the initial learning stage a 12gb GPU is suitable for your needs (like a 5070) or if you struggle hard to find what you need within a small model, that 12gb VRAM wouldn't work for you either, so you'd have to step up to a 5070 ti 16gb VRAM. Big difference in price these days sadly.

There's a HUGE amount to learn though and to get your teeth into. For instance a 7.5GB model can fit in the 8GB of VRAM a 3070 has, but the Ui and back end will also eat into the VRAM, and over filling your VRAM kills responsiveness.
 
got a unrelated question. How do you generate the same face and body for every promt like those AI profiles? do you need to build and train a LORA?
 
Yes, but sites like websites/apps like Kling can avoid having to make LORA'S simply by using reference images. I guess it's just their algorithms are highly trained to pick out facial details, but for home use, to generate an image without a reference, with a consistent face and body you'd need to build a LORA.

I'm not sure what the expression means to "train" a LORA however, as it doesn't learn to need training. You can't provide it feedback to correct itself. You just feed it images for it to work from. "Building" a LORA however is relatively simple comprising of images from as many angles as possible of your subject.
 
Hi folks, does anyone have any recommendations for software and models that can produce decent results with a 8GB 3070 Ti, and that are worth the effort for creating images and short videos like this, 3D or photorealistic ones? Should I use ComfyUI or are there simpler and good results producing new tools on GitHub?
 
ComfyUI is the most complicated, manual user interface you can use. It's not a good place to start as a beginner as you have every option available for you to tweak and therefore break, it's quite overwhelming at first.

Start with making images, using Stable Diffusion and something like Fooocus and experimenting how far you can go with that.

Especially with a 3070ti and 8gb VRAM. Experimenting will be slow, so learn the basics first and start small. The video creation is not something I've started on yet, but you'll see how how messy your images come out at first, with deformed fingers and eyes etc, videos will be the same.
 
Well the easiest I've found is with Runpod. Theres a user on civitai which has a runpod template which sets up everything ready. All you need to do is insert an image and change some settings add some loras and type in your prompt and run and you have a high quality video. The Template is called Yet another workflow. he has a full guide and also a video guide too. I have been getting decent results and have found H3 to be surprisingly prompt enhanced. I literally prompted it like grok and got okeyish results - for example with a starting picture of a womans face: instant cut to new scene, she is laying in bed fully naked holding her knees, big tits bouncing, pov a man is fucking her missionary, he is thrusting his dick deep inside her while she moans and looks at viewer.

I have been getting very good results like this although when prompting it correctly you can get way better results, especially with multiple shots
 
You must log in or register to reply here.
Back
Top