Generating AI images with StableDiffusion - Beginner Guide
- Thread starter Psan2022
- Start date
Psan2022
Member
- Mar 8, 2022
- 100
- 143
nicod
Newbie
- May 17, 2019
- 82
- 235
Amazing please keep ur tutorial going iam following u and downloading itYes, if you exclude your electricity bill
- Jun 19, 2019
- 9,157
- 34,244
Psan2022
Member
- Mar 8, 2022
- 100
- 143
It is quite excentric. Like Van Gogh
To answer your question I need some intel:
1. What checkpoint are you using
2. What prompt
3. Any addons like Loras or are you using plugins?
4. What resolution are you generating in?
Psan2022
Member
- Mar 8, 2022
- 100
- 143
Ah I see now. The default one is quite shit.
Download a checkpoint in the style of the image you want to generate.
SD 1.5:
For realism like this ->
For anime style ->
SDXL/PONY (More taxing on your PC, image gen resolution around 1024x1024):
Realism->
Anime Style ->
Edit: If you want to generate Hentai/XXX images Pony style Checkpoints and Loras are quite easy and straight forward to use.
Download a checkpoint in the style of the image you want to generate.
SD 1.5:
For realism like this ->
You must be registered to see the links
For anime style ->
You must be registered to see the links
SDXL/PONY (More taxing on your PC, image gen resolution around 1024x1024):
Realism->
You must be registered to see the links
Anime Style ->
You must be registered to see the links
Edit: If you want to generate Hentai/XXX images Pony style Checkpoints and Loras are quite easy and straight forward to use.
nicod
Newbie
- May 17, 2019
- 82
- 235
NansException: A tensor with NaNs was produced in Unet. This could be either because there's not enough precision to represent the picture, or because your video card does not support half type. Try setting the "Upcast cross attention layer to float32" option in Settings > Stable Diffusion or using the --no-half commandline argument to fix this. Use --disable-nan-check commandline argument to disable this check.
anyone knows why i get this
Psan2022
Member
- Mar 8, 2022
- 100
- 143
Open the webui-user.bat in the editor and put this into the command line: --xformers --no-half-vae
nicod
Newbie
- May 17, 2019
- 82
- 235
i did and got same error ..
Edited : Fixed now i after i enabled Upcast cross attention layer to float32
Edited again : nahh seems the issue still exist
Last edited:
Psan2022
Member
- Mar 8, 2022
- 100
- 143
nicod
Newbie
- May 17, 2019
- 82
- 235
it caused the programm to render one pic in like half hour
the strange is it happens sometimes when i put many promps
MY card maybe the issue it's Old GTX 1650
Psan2022
Member
- Mar 8, 2022
- 100
- 143
Well, I would not expect any miracles with that card. For image generation you need a lot of VRAM and Tensor Cores. An GTX card does not have those sadly.
Thanks for this thread. Taught me a lot.
I'm wondering how you'd go about using it for a game though. How easily can you generate repeated images with the same characters, without them taking on different appearances from one shot to the next? What about backgrounds that make narrative sense?
In terms of animations, I've seen animated AI images but I believe they're very limited, and very obviously AI-generated. How feasible would animating these images be with Live2D or Spine?
I'm wondering how you'd go about using it for a game though. How easily can you generate repeated images with the same characters, without them taking on different appearances from one shot to the next? What about backgrounds that make narrative sense?
In terms of animations, I've seen animated AI images but I believe they're very limited, and very obviously AI-generated. How feasible would animating these images be with Live2D or Spine?
Psan2022
Member
- Mar 8, 2022
- 100
- 143
I am by no means an expert in this but generating consistent characters is not easy but doable. You can achieve somewhat consistent characters by utilizing different prompting techniques and by describing the hell out of how your character looks in the prompt. the more you describe the less the AI has to imagine how that person may or may not look. There are also some Addons for Automatic1111 that may achieve what you want. You also have the option of using a Lora of a character that looks similar to what you want and change the look via prompting.
I recently dabbled into ComfyUI and this video popped up a while ago. It seems really close to what you search for.
ComfyUI has a node system. It has a steep learning curve but the control you have over your images is crazy. If I get more compfortable using ComfyUI then I will make maybe another tutorial.
Video AI models have the same issue as LLMs have or had in the past. They can link thoughts together for a few seconds until they loose the plot. For example image to video generation has come a long way but for local generation 5-10 seconds is quite the max.
A few days ago WAN2.1 has been released. You can use the WAN2.1 model in ComfyUI and make really great short AI videos from a single image and a few lines of prompts or only from text.
FallenHeroes
Newbie
- Aug 23, 2018
- 54
- 20
Psan2022
Member
- Mar 8, 2022
- 100
- 143
I used to use the
You must be registered to see the links
But Automatic1111 has its own thing now that works quite well as well.
If you are using Automatic1111 with version 1.6 or newer you should have this box under the Generate button:
It should be there without any addons. You click on the Brush symbol and type in your positive and negative prompts. After saving you can select your preset from the list. If it is selected those prompts will be applied everytime you click on generate.
ADDITONALLY I recommend this addon. This saves your changes and prompts. So that If you close the server or close the Webui your changes are kept:
You must be registered to see the links
Under Settings/State you have to check the boxes what you want to be saved. Be it your last prompt, certain settings etc. Dont forget to save and restart the UI after that.
FallenHeroes
Newbie
- Aug 23, 2018
- 54
- 20
Thank you very much for your help. Also,I tried to create something similar to what you made in this photo (Example 2)
But I couldn't manage to.Did you use this model ? I wasn't able to see XL version.
You must be registered to see the links
also what tags and loras did you use? Can you give me more details, please?
Last edited:
Psan2022
Member
- Mar 8, 2022
- 100
- 143
This is the right checkpoint for HassakuXL:
You must be registered to see the links
The prompts are directly under the example picture.
If you look closely there is a Lora prompt in there.
So the only Lora I used in this picture was
You must be registered to see the links
with a weight/strengh of 1