- Jul 12, 2017
- 1,161
- 1,109
Tried CivitAI for quite a bit...
You don't have permission to view the spoiler content.
Log in or register now.
- Jul 12, 2017
- 1,161
- 1,109
I think it'd be best to list out the necessities when people wanted to use their own PC's capabilities rather than CivitAI so that we don't need to go to other pages on this thread (if this gets significant traffic in the future).
- Jun 9, 2019
- 369
- 1,409
You must be registered to see the links
This is the same UI than automatic1111, which is the most used UI for Stable Diffusion. So if you need some tutorial, you can search for automatic1111 on youtube.
So why Forge? Because it generates render way faster compared to automatic1111 and have essential extension already installed.
You don't have permission to view the spoiler content.
Log in or register now.
- Sep 18, 2018
- 10
- 47
Basically, Lora is good at drawing characters from their other settings:
You don't have permission to view the spoiler content.
Log in or register now.
Some news:
1) I changed GPU, now i have RTX 4080 Super 16Gb (i'm not rich, now i'm very poor
);
2) And added short tutorial in first post of this thread, to show how you can generate images on your home PC with ForgeUI.
Note that english is not my main language, so suggestions for changes in tutorial are welcome, i suppose)
"4080 Super" is faster then "3070", but the difference is not so big, compared to what i expected. Biggest change is in comfort, my PC now significantly less laggy while i generate something) Also with more VRAM it is easier to work with higher resolution images.
I think i deleted more then posted)
1) I changed GPU, now i have RTX 4080 Super 16Gb (i'm not rich, now i'm very poor
);2) And added short tutorial in first post of this thread, to show how you can generate images on your home PC with ForgeUI.
Note that english is not my main language, so suggestions for changes in tutorial are welcome, i suppose)
"4080 Super" is faster then "3070", but the difference is not so big, compared to what i expected. Biggest change is in comfort, my PC now significantly less laggy while i generate something) Also with more VRAM it is easier to work with higher resolution images.
You don't have permission to view the spoiler content.
Log in or register now.
I think i deleted more then posted)
I was realised that with ADetailer and inpaint we can generate series of images with different expressions! So we can generate more coherent art)
Also i tryed to fix "wrong" hands, but failed(
You don't have permission to view the spoiler content.
Log in or register now.
Also i tryed to fix "wrong" hands, but failed(
- Jul 12, 2017
- 1,161
- 1,109
I got the Forge working but changing checkpoint and VAE results in a "connection error"
Any ideas on what I should do?
Any ideas on what I should do?
I used my old C++ utility for ffmpeg, and make small video, to see how it looks) Not good.
Also i tryed deforum, but result is very unpleasant(
Also i tryed deforum, but result is very unpleasant(
You don't have permission to view the spoiler content.
Log in or register now.
Maybe if you share prompt used on each generations is easier to create more coherent pictures too. more people can try run same prompt and adjust
VAE is not so necessery, just change checkpoint.
Maybe not enough VRAM. By default ForgeUI loads this model: realisticVisionV51_v51VAE 1,98 Gb.
This model can be fully loaded in 4Gb VRAM, but PDXL6 isn't, because of 6,46 Gb size.
Looks like you can only use SD1.5 models. If that so, then try this checkpoint and model:
You must be registered to see the links
You must be registered to see the links
I already upload original image in this post: https://f95zone.to/threads/kingdom-...-future-plans-healthcare.211283/post-13961304
To generate other images i just used inpainting (orc's hand and sabia's breast) and 8 different seeds.
But i don't think that we can generate decent animations with this method for a reasoble amount of wasted time. So series of pictures is better alternative.
The house cleaning in the main threat was probably only a matter of time. So i assume this would be the right place for the ai pictures, discussions and other things that are not directly tied to the game itself, right?
I still can't do creative work to save my life and i would need quite a long time to understand this ai-generation method, but watching all these pictures makes me wonder:
Even i could write a bad subtitle to any of the pictures here in paint, so that is technically easy to do.
Because of this i guess it still isn't too difficult to pick a few pictures and make a few panel worth of short comic either.
But i imagine that for a short story or a scenario the method would have to be reversed and instead of stringing together random pictures, the story would have to be written and then the pictures would need to be generated for that.
So to those of you, with experience in this: am i wrong? and how realistic would it be to generate a picture to illustrate a scene and maybe make a short comic out of it as a proof of concept?
I still can't do creative work to save my life and i would need quite a long time to understand this ai-generation method, but watching all these pictures makes me wonder:
Even i could write a bad subtitle to any of the pictures here in paint, so that is technically easy to do.
Because of this i guess it still isn't too difficult to pick a few pictures and make a few panel worth of short comic either.
But i imagine that for a short story or a scenario the method would have to be reversed and instead of stringing together random pictures, the story would have to be written and then the pictures would need to be generated for that.
So to those of you, with experience in this: am i wrong? and how realistic would it be to generate a picture to illustrate a scene and maybe make a short comic out of it as a proof of concept?
- Jul 12, 2017
- 1,161
- 1,109
Maybe if the current SDXL and KOD PDXL models are trained to create a scene based on (at least) ChatGPT prompts, I think it's possible -- though I'm not sure how long it takes for one to achieve that milestone.
I'm not particularly well-versed in Generative AI content or AI in general, but that's what I get when thinking about core AI procedure when it comes to content creation: Have both generative written and visual prompts learn each other's strengths and weaknesses, then find the sweet spot
- Jul 12, 2017
- 1,161
- 1,109
With this, there's an implication the size of the model corresponds to the size of VRAM needed to properly run it?
Correct me if I'm wrong.
- Nov 4, 2019
- 573
- 731
It's a good start. Or try this.. Try generating a series of images that would go together in some kind of timeline for a scene. I could draft up a story for the scene. Or we could do that in reverse.. I draft up a scene, then you folks see if you can generate images that fit that scene.
- Jun 9, 2019
- 369
- 1,409
Give it a try. Frankly, the AI is doing the job for you, a few prompts can give you a good result. You hit generate a few time, and when you got a good picture, you use the same seed for consistency. If you want a more complicated, specific pose, this is where img2img and controlnet come into account. Basically, it will copy the image style (depending on your noise setting) to get what you want.
For the best image, respect the resolution of the model.
Here's an example of a specific pose replicated using img2img and controlnet with few words.
- Jul 12, 2017
- 1,161
- 1,109
- Jul 12, 2017
- 1,161
- 1,109
Here are some more with my favorite girl(s)
You don't have permission to view the spoiler content.
Log in or register now.