discussion Which AI image generator do you use?

[rifticc]

I have Stable Diffusion installed locally, but I've never been very good at generating images with it, it's mostly for games that can take advantage of it.

i think you need custom huggingface models to be able to generate good nsfw results

 

[HTYYHRK]

i think you need custom huggingface models to be able to generate good nsfw results

Oh yeah, I have models and LORAs, something like Free Cities can generate decent results with it. I just don't have the patience to tinker with prompts to get the same quality of results

 

[IxChewx]

I have Stable Diffusion installed locally, but I've never been very good at generating images with it, it's mostly for games that can take advantage of it.

Same. I use Adetailer and pass the image through img2img several times to finally get a good result.

 

[rifticc]

Oh yeah, I have models and LORAs, something like Free Cities can generate decent results with it. I just don't have the patience to tinker with prompts to get the same quality of results

thats great. do you have recommendation on which model performs better ?

 

[SilentHunter]

I don't use them, for a very simple reason - I don't need them. There's an idiom in my native language that says you need it like a straw in your ass. This will be more accurate.

 

[Sobmania]

I use

Spoiler

You must be registered to see links

to generate images, Gemini’s Nana Banana to refine them, and I make some adjustments by hand. It’s not ideal, making something more pornographic is difficult because of Nana Banana, but it’s not impossible.

thats a nice site. thanks for sharing

 

[HTYYHRK]

thats great. do you have recommendation on which model performs better ?

The two I've used much are galenaREDUX and I think hassakuXL, but it's been awhile since I've messed with it.

 

[Feetoff]

NovelAI is phenomenal for anime style images. It's unlimited if you pay monthly for it and has a ton of tools to enhance your work. It even has a character reference mode that'll generate images based off of any character you upload.

 

[heroahoy]

I use

Spoiler

You must be registered to see links

to generate images, Gemini’s Nana Banana to refine them, and I make some adjustments by hand. It’s not ideal, making something more pornographic is difficult because of Nana Banana, but it’s not impossible.

I'm just starting out and got my feet wet using Perchance. You just type what you want the image to be in the text box and select an image style and click on 'generate'. It's online and free.

You can see the stuff I've done in the public and m+ daily theme art hosted by Mime.

Here's a recent example:

You must be registered to see attachments

I have been messing around with Perchance and it's pretty nice and simple but is there any way to up the resolution?

 

[fuzzi_weezil]

I have been messing around with Perchance and it's pretty nice and simple but is there any way to up the resolution?

Not that I'm aware of, but if you find out, let me know!

 

[Smogman]

I've haven't seen an online one that I have liked. Plus I like the privacy aspect of running local image models. ComfyUI, although not the simplest, has been the best tool for me

 

[mahname2]

I've tried most of the popular open source models (SDXL, Flux 1.dev/2.dev/2.klein, Z-Image Turbo/Base, Qwen Image, and even Hunyuan Image 3) through ComfyUI, but in my experience AI image gen just isn't good enough yet.

The smaller models with LoRAs can generate good lucking imaging, but they suck at prompt following and have very limited vocabularies. That makes them kind of useless for anything even slightly complex; if you want to simply make a good looking picture they can do it, but if you want a good looking picture of something specific then small models are a no go.

Larger models have a lot better prompt following, but since training LoRAs is so much harder (and required more expensive hardware) there aren't really any NSFW LoRAs. Most base models are super censored, so this makes them unusable (for example, Flux 2.dev can generate very high quality with pretty good prompt following, but any nudes will have characters covering themselves with their hands and even if you get around that they get barbie doll genitals). It might be possible to fix censorship with image-to-image or inpainting after generation, though I haven't tried that.

Hunyuan Image 3 is a standout as am absolutely massive model without any censorship, so it'll generate NSFW even without LoRAs. It's probably my favourite in terms of quality and prompt following, but even on a 5090 with 64 GBs of RAM it's only barely possible to get it to work because it's so big (and since I need to swap a ton it takes ~15 minutes to generate an image). For decent speeds it really needs an RTX PRO 6000, making it way out of reach for 99% of consumers. Also, it's not good at generating realistic genitals even if they aren't censored.

 

[LewisCarroll]

I use Gemini Pro tools for some work, but I use ComfiUI to do more serious work (audio, video and fx).

 

[Cratosstratos]

Stability Matrix with ComfiU runs great on windows and nvidia and i use modell/checkpoint JuggernautXL.