Neural networks are capable of creating amazing images. We’ve put together 5 tips that are cool for Stable Diffusion, Fooocus, and the PornWorks generator, and some for Midjourney or DALL-E.
Let’s go!
LoRA is a mini-model that allows you to more accurately hit the style or look you want. Still useful if you need a certain character or pose.
Dark atmospheres, logos or game assets – LoRA will help with that. Just don’t get carried away and use no more than 5 😁
You can find ALL LoRA on CivitAI. And they’re already available on PornWorks.
References can be applied by different ways.
The ControlNet add-on has a bunch of preprocessors to help utilize the depth map, pose, and various masks.
A very creative tool that will make generations more interesting and unique.
To avoid downloading 50+ gigs of models, we compared them all and made a miro board.
Uploading the image to CLIP Interrogator and we get the finished prompt.
You can refine the request manually or use Florence-2, it’ll help make the prompt more detailed.
For each model and neuronkey, there are settings that drastically change the result.
Try changing the following:
Realistic models often use values 1-2, anime models use 7.
A good prompt is an accurate and detailed prompt. Adding a bunch of extra words will not make the image better. Rather, you’ll get farther away from the desired result.
Also check the model on CivitAI for trigger words or function words. For example, Pony models require a bunch of score tags to work properly.
The more you work with prompts and generate, the more accurate they will be and the faster you will get to the results you want.
Even the coolest generation can be made even better with final processing.
Try to use:
Which of the methods are you already using? Share in our Discord community.
