General
Can I post gens made with non-SD-based models and generators (Midjourney, Dall-E and others)?
Yes, as long as your pic is /e/ and AI-generated, you are welcome to post it.
Thread highlights? What are these?
It is a selection of gens from the previous thread that stood out, whether for their style, or pose, or composition, but most importantly - their baseline quality and/or attention to detail. It started as an initiative to encourage quality posting and stayed as a tradition. Chosen and compiled by the thread baker the selections are obviously very subjective, so don't be sad if your gen(s) didn't make it.
What is the Request Anchor for?
Anchors are a special recurring type of posts that usually take the second or third spot in every thread. Request anchor on /edg/ is used specifically for requesting specific gens or gen edits. Got a request? Reply directly to the Request Anchor with what you have in mind and some kind anon might pick it up and fulfill it. The best practice whenever your request gets fulfilled is to provide your opinion and feedback to the anon who made a gen for you.
What is thread theme/challenge and what is the Theme Anchor for?
Thread theme is a topical prompt for the current thread, usually in the form of a certain prompt idea or a thread-wide challenge. Themes change with every new thread and can vary from something vague and simple to something more specific and challenging. Note that it is neither a substitute for your usual random gen posting nor it is a contest, but a "side activity".
There's just one basic rule: make your gens in the spirit of the theme, taking the theme prompt as the base (not word for word, just the general idea). And If it is a challenge with a defined set of rules, try following the instructions as close as possible.
Theme anchor exists mostly for convenience, with it we announce the thread theme/challenge and anons who participate can reply to it with their theme gens.
Previous themes/challenges list (latest at the bottom): https://rentry.org/edg-theme/
We also have a form for your theme ideas and suggestions: https://forms.gle/K3sTVmW2T4qnXkq49 (no login required)
I have so many gens, should I chain-post them all?
You shouldn't. /edg/ isn't a fast-moving general, so one-man spam can be pretty disruptive. Quality over quantity is the preferred way. If you still want to post more than just a few gens at once, a good practice is to compile them into a mini-collage/set using your image editor of choice. A Catbox gallery is another option.
What about fringe content that is in the spirit of the thread (monster girls, mild yuri, 1girls that may be a bit on a lewder side, etc.)?
We allow it, in moderation. Be reasonable, don't go too deep into fringe territory, don't spam and you'll be fine. Probably.
What's Catbox?
https://catbox.moe/ is a file/image hosting for imageboards. 4chan strips metadata containing prompt info, but Catbox preserves it, so anons use it to share the prompts embedded in their gens. Prompt information can be loaded from the PNG Info tab of the WebUI. Check the filename to see if it has a catbox link before asking. | 4chanx Catbox + NAI prompt script: https://rentry.org/hdgcb
Some anons may be using Stealth png info extension, which stores prompt info in the channel that doesn't get stripped by 4chan: https://github.com/neggles/sd-webui-stealth-pnginfo | For temporary storage and slightly higher size limit use https://litterbox.catbox.moe/
Anywhere else I can share my 1girls other than /edg/?
https://aibooru.online/ | https://www.pixiv.net/en/ (make sure to properly tag your uploads as AI art and strip all the metadata to avoid getting reported and banned for using artist styles) | https://rule34.xxx/ | Twitter | For non-/e/ posting look through the list of related generals in the OP (/b/ also has some regular AI threads with way more lax rules, if you're into that).
/edg/ logo?
Main: https://files.catbox.moe/uc2ntj.png | Pink/black: https://files.catbox.moe/qezz6l.png | Vaporwave: https://files.catbox.moe/jpln9l.png | b&w: https://files.catbox.moe/7h6z33.png | Retro: https://files.catbox.moe/1m9tjl.png
Request anchor logos: https://files.catbox.moe/bx1lo2.png | https://files.catbox.moe/281vzj.png | https://files.catbox.moe/tu230e.png
OP template?
Technical
Guides/tutorials?
Completely new to this? First, install one of the local UIs (see OP for links), then grab one of the recommended models (see below) and a VAE (for SDXL-based models it's sdxl_vae), then watch the following videos, and start prompting. In the meantime, browse the "Beginner" section at https://stable-diffusion-art.com/tutorials and start looking through "Resources and guides" part of the OP.
Upscaling, danbooru tags, xyz plots, prompt editing: https://youtu.be/Jya6qcokqZQ
Img2img and inpainting basics: https://youtu.be/NJwg7HWiZQw
More detailed inpainting tutorial: https://youtu.be/DgphyDiD4TE
Hand fixing: https://youtu.be/g4Oggft64dI
Alternative beginner-friendly img2img/upscaling method for improving your raw gens: https://civitai.com/articles/4560/upscaling-images-using-multidiffusion
Recommended models for /e/?
Free: AutismMix SDXL (PonyXL base), AnimeConfettiTuneXL (PonyXL base), AnimagineXL 3.0/3.1 (SDXL 1.0 base), Based68 (SD 1.5), Olympus v4 (SD 1.5) - all available at https://civitai.com/
Paid: NAI v3 - https://novelai.net/ (more about the model: https://rentry.org/hdg-nai-v3)
Recommended WebUI extensions?
a1111-sd-webui-tagcomplete
adetailer
multidiffusion-upscaler-for-automatic1111 | ultimate-upscale-for-automatic1111
sd-dynamic-prompts | sd-webui-ranbooru
sd-webui-stealth-pnginfo
sd-webui-regional-prompter
sd-webui-depth-lib
For NAI users: sdwebui-nai-api
Recommended upscalers?
4xAnimeSharp | 4xfoolhardy_Remacri | 4xNomos8kDAT. Note that depending on the image style and use case one might work better than the others, so experiment.
Best tool for simple gen retouching?
IOPaint (formerly LamaCleaner): https://huggingface.co/spaces/Sanster/iopaint-lama | https://www.iopaint.com/install
How to fix bad hands/eyes?
Watch the noob guide
Try sd-webui-depth-lib and this guide: https://rentry.org/inpainting-guide-SD
Additional hands library: https://civitai.com/models/67174/900-hands-library-for-depth-library-or-controlnet
Depth lib tutorials: https://files.catbox.moe/8r4y83.webm and https://files.catbox.moe/j46t3b.webm
Try the bad-hands-5 and negative_hand embeddings.
To fix bad eyes, use Adetailer with face/eyes model.
My gens on SDXL-based models look somewhat desaturated or lacking contrast, how can I achieve better colors/contrast?
There are a number of scripts and extensions that can help with that: FreeU (included by default in Forge) | https://github.com/thomaseding/sd-webui-color-enhance | https://github.com/Seshelle/diffusion-noise-alternatives-webui (note: hires fix doesn't work with it) | https://github.com/Haoming02/sd-webui-vectorscope-cc | https://github.com/city96/ComfyUI_ColorMod (for Comfy). Alternatively, try one of these VAEs: https://civitai.com/models/152040 (some versions may fry your gens more than others). Artist loras also affect the colors/contrast a lot. None of these solutions are perfect, cause it's a fundamental issue with either SDXL itself or the SDXL VAE (opinions differ on which one, or maybe it's even both?).
The lighting on my gens is too bright, I want to make darker gens with SDXL-based models
Just like with colors/contrast, you can use either to control the lighting/noise in your gens: diffusion-noise-alternatives-webui | sd-webui-vectorscope-cc
What does BREAK in a prompt field do?
The prompt you write is broken down into tokens, and normally SD comes with a hard 75 token limit. The WebUI allows going above it in steps of further 75 tokens by interpreting each 75 token set in parallel, and merging the results for the final image. Check out the top right of a prompt field to see how many tokens you have left. If it starts saying "X/150" you have gone above the 75 limit, and now filling up the second token set. Doing so causes some tokens to lose their impact, especially if a tag is being split into both the first and the second 75 token set.
Using BREAK in your prompt enforces everything following it to be part of a new token set; the previous one will be padded with empty tokens so it reaches the 75 limit prematurely. Some anons use BREAK to make a new token set for each distinct feature of the image.
BREAK is also used in Regional Prompter extension in a different way, so make sure not to mix it up whenever you have it on and read the RP documentation.
A certain Danbooru tag I used in my prompt didn't work at all
Try applying higher weight to the tag or readjusting the tag's position in your prompt. Look for an alternative tag one e621 and use it instead: https://e621.net/tags Also note that some tags, even the fairly common ones, simply don't work on certain models, depending on how they've been trained.
How can I separate two or more characters with their own distinct features without getting all the details mixing up and "bleeding"?
Use the Attention Couple extension (Forge UI only): https://github.com/Haoming02/sd-forge-couple (works best for interactions between 2 characters)
Or use Regional prompter extension: https://github.com/hako-mikan/sd-webui-regional-prompter | https://stable-diffusion-art.com/regional-prompter/ Prompt Cutoff extension might help too: https://github.com/hnmr293/sd-webui-cutoff | Read this guide: https://civitai.com/models/339604/how-to-generate-multiple-different-characters-mix-characters-andor-minimize-color-contamination-or-regional-prompt-adetailer-and-inpaint-or-my-workflow Don't expect perfect results and prepare to still have to use inpainting, cause all current models are bad at this and no extension can fully "fix" it. There is cautious optimism about the upcoming SD3 having much better prompt comprehension, though.
The SD 1.5 Controlnet models don't work on SDXL, where can I download the models for SDXL?
https://civitai.com/models/136070 | https://huggingface.co/lllyasviel/sd_control_collection/tree/main
I can't make large images, I'm running out of VRAM
Consider using WebUI Forge, it's more vram efficient on XL. Use --medvram or --lowvram, which saves VRAM but takes more time. Highres fix is VRAM hungry, skip it if you're running out. Ultimate Upscale and Tiled Diffusion use tiles to upscale, which reduces total VRAM usage.
https://github.com/AUTOMATIC1111/stable-diffusion-webui/wiki/Troubleshooting#low-vram-video-cards
What's with the three-letter prompting on Pony-based models?
https://lite.framacalc.org/4ttgzvd0rx-a6jf All you need to know is that you can use these 3-4 letters from the left column like any other tag in your prompt to get the corresponding artist style or character (add more weight if it's too weak). It's like magic. The brony magic.
How do I gen in anime style on a furry model?