>>938387878
>>938388385
There are a couple of things you can do
- The checkpoint seems to be Illustrious-based, which is trained on booru-style tags, so you'll probably have better outputs by using tags instead of natural language prompting. If you're unfamiliar with those, head over to Danbooru, open some random images and check their tags just to get the hang of it.
- You'll probably need some base styling in your prompt. If you go to the CivitAI page of the model and open the first image, you can check the prompt that was used to generate it, and you can get some starter tags from there. To be honest, adding something like "masterpiece, best quality, newest, absurdres, highres," to the prompt can already make a difference. Once you have some familiarity with the tags, you can make up some cool styles. (Also, same goes for the negative prompt)
- Gen at a bit higher resolution - at least around 1024px on one edge. Depends on the checkpoint, but Illustrious was trained on images around or above 1024×1024