>>17139805So about this problem,
>>17138056Have you learnt about high resolution fix checkbox and prompt emphasization?
Resolutions beyond 512x512 and depending on model, 768x768, can cause generational abnormalities, the high resolution check box in the text2image generation tab will upscale smaller resolutions and try to have a second guess of what it saw and refines it.
Prompt emphasization, has two/three possible ways to work in the input.
Parentheses around a prompt/word like: (1girl, solo), ((1girl, solo)) or (((1girl, solo))) will try/force it to be included in the final output.
The other syntax is to use a colon like this
1girl:1.3, solo:1.2, when a prompt is written, the standard weight of it will be 1.0, you can also go lower like :0.6.
Another way to emphasize an output is to use multiple times similar, synonymous prompts/tags/words.
If you are somewhat of an artist and you can actually doodle a bit or use an external Photoshop program, try to use img2img more often than text2img doe, it's basically the next level of tool for the digital artist, well after digital art programs were made. In img2img, you have these two buttons called interrogate CLIP and interrogate DEEPBOORU, when you click them for the first time after inputing an image, nothing will happen, but wait, check your CMD, it downloads a file for each about 2 GBs in size.
When the files are installed, after clicking the CLIP button, Stable Diffusion tries to tell you what it sees on the image. DEEPBOORU will make Stable Diffusion try to tag the image with Danbooru tags. This is not really accurate however do it as guideline. You don't even have to explain the image, just tell it some nonsense, this is how you do meme images like Basedjak lolis.
Like all images I posted yet here are img2img.
Artists recognized by Stable Diffusion default model on which most are trained on need most often to be emphasized.
Real manga and hentai artist style can be added with a LORA from Civitai or /e/.
