If I want to make realistic NSFW pictures and video which software/pipeline do you recommend?
When it comes to getting high quality nsfw pictures you can pick 2 of the 3: Ease of Getting High Quality Outputs, Lower VRAM Requirements, and Lack of Censorship (usually tied to amount of LoRAs available).
I'm no expert at generating AI images but I'd say the best methods are as follows:
A) Use Flux models with tools like
You must be registered to see the links
or
You must be registered to see the links
. Flux models are a great place for beginners to start since they have better prompt adherence than Stable Diffusion models and have quite a few tools that will give you NSFW outputs without steep VRAM requirements. Flux Kontext will also let you iterate on existing images through text-prompts. The only problem is they won't give you the most high quality images, but it's an acceptable tradeoff for many since Flux outputs often have less defects than SD outputs.
B) Get good at learning how to use Stable Diffusion models like SDXL or a SD variant like Pony Realism with
You must be registered to see the links
workflows. This is the most time consuming and difficult path, but done properly it will give you excellent results that are on par with or better than SOTA image models without needing as much VRAM thanks to the extensive tools and LoRAs available for them.
C) Learn to use Wan 2.1/2.2. While technically a video model, people realized quickly that generating 1 frame with Wan models was a great way to generate high-quality, prompt adherent images. Middling requirements for VRAM as well, especially if you're going for single-frame images. Something about it being a video model gives it really good understanding of where things should belong in a picture. Unfortunately it's harder to get really nsfw outputs from it due to the lack of LoRAs
D) Use SOTA models like Gemini 2.5 or ChatGPT5 to generate good sfw starting images and then iterate on those images using their inbuilt image editing tools, then modify those pictures using local AI tools that allow for inpainting and your own editing skills.
*A note about VRAM requirements: While you will see people claim to be able to run more powerful models on lower VRAM with quants, remember that without pictures their claims of "good" outputs should be taken with a grain of salt. Lower quants will result in lower quality images and/or lower prompt adherence most of the time, and you'll have to get creative with your ComfyUI workflows to make up for that. Even if you get a good workflow going, without a powerful GPU you will be waiting a lot for these images to generate.
If you want something where you can just get hardcore nsfw images of good quality by just prompting, you'll just have to keep waiting for now. That doesn't exist yet, and there's a lot of incentive to not make generating hardcore nsfw content easy and accessible with AI. Those AI image generation websites that offer subscriptions or token-based pre-payments? Don't use them, they're ripoffs and are bound by censorship rules.
Learn ComfyUI. There's no better alternative.