>>3022738Are you doing local gens?
Most modern models are SDXL based. SDXL was originally trained on natural language, like "a picture of an apple." However, the porn models have been heavily trained on porn datasets, largely Danbooru for anime models though the best of the new models are NoobAI based and that was trained on Danbooru and e621 so you can use tags from both.
Sometime natural language can still be helpful, but it's largely drowned out by tag training.
For extensions, locally the most important ones are probably:
ADetailer - an automated inpainter that can improve faces, hands and more.
https://huggingface.co/Bingsu/adetailerRegional Prompter - splits an image into regions and lets you prompt each region, good for multi-character scenes.
https://github.com/hako-mikan/sd-webui-regional-prompterYou could also look at ControlNet and IPAdapter extensions once you get the hang of things, those are both powerful image input extensions but require more configuration and their own models.
A typical workflow is doing txt2img at close to 1024x1024, inpainting/editing if you want, then upscaling to higher resolution preferably with denoise to blend in the edits (like Hi-res fix does).
Gradio based UIs like A1111/Forge/Reforge are still the easiest and most popular UIs. ComfyUI is good for new tech support like video gens and Flux models.
There are thousands of different model merges, but anything based on NoobAI is going to be more or less the best you can get with SDXL. Flux is the next generation, but currently only NovelAI has a good porn-capable Flux model and it's paid only.
A /g/ anon did a poll for the most popular AI video cards and the most common where 16GB cards, with the 3090 standing out as the most popular 24GB card. VRAM is the most important for AI gens, you really only need 8GB but more is better with higher resolutions and faster gen times.
Hope that helps! Check out other gens' metadata and that should teach a lot too.
