>>8688I use "
https://civitai.com" to find Loras. Loras are trained files that help create specific characters, poses, emotions, etc. Download a Lora and place into "Stable Diffusion Webui/models/Lora folder". Then, when inputting prompts put, "<lora:(LoraFileName)>" so it can reference it. The Lora tag is usually the downloaded file name. (Ex: <lora:AcerolaLora-10> or <AcerolaLora-10:1>.) The number after the second colon strengthens or weakens the Lora, (1, 0.9, 0.8, etc.) You usually want to weaken it so the character can be placed in different poses, but you can play around with it. You can also do this with other prompts without Loras by putting them in parenthesis like this: "(thick thighs:1.1), (flat chest:1.2)"
I personally set my generation settings to: Steps:20~27, 640x640 or 720x720. The lower you go the faster the generation, but it worsens the quality. For "Sampling Method" I use DPM ++2M Karras. Any of the other DPM++ are good too, except "fast." Once I find a good pic, click the "Re-use" button next to the "Seed" bar to save the seed/picture. You can add and remove prompts to alter the same picture, but some prompts can change the picture too much.
Then, I checkmark the Hires.fix box, set Deionizer between 53~60, and CFG Scale 7~9. You want Deionizer at least above 50 or the picture won't change much, and setting it too high will get you a different picture. CFG is the strength of your prompts, I usually don't move it too much, but you can play adjust it however you like. The picture may be altered a little.
For Upscaler use R-ESRGAN 4X+ Anime6, or use 4x Anime/UltraSharp upscaler from here:
https://civitai.com/models/78396/fix-your-blur-hires-modelsInstall these in: "stable-diffusion-webui/models/ESRGAN folder"
I set "Upscale by" to 2 but you can adjust to how you want.
Once it's upscaled, click the button "Send to img2img" Set the "Sampling Method" to the same you used before, "Sampling Steps" to the same, or more if you want to add a little more details. Set "Width and Height" to double, or more, of the original size you used before, and "CFG" and "Deionizer" as the same as before. From there, you're pretty much done.
If you have a weaker GPU, go to "stable-diffusion-webui and right click on "webui-user.bat" and click on edit using Notepad. Where it says "set COMMANDLINE ARGS" put these prompts after it, like so and save:
set COMMANDLINE_ARGS= --xformers --no-half-vae --medvram --opt-split-attention --disable-nan-check (space in-between each prompt)
xformers lowers the VRAM usage for GPUs. (Doesn't work for AMD GPUs.) No-half allows better generations by creating fp32 instead fp16 generations, this can prevent black squares appearing in pictures. (This is not always needed.) Medvram/lowvram makes generations take longer to create, and weakens the quality, but lessens the amount of VRAM required. Medvram, I found, doesn't affect quality too much, but is best for 8gb GPUs and higher. Opt-split-attention should increase generation speed without increasing VRAM usage. Disable-nan-check removes some errors you may encounter when generating. This is what I use and haven't encountered any errors, but someone else may have different settings depending on their GPUs.
Anyway, that's my essay, teacher. Have I reached the word count yet?