Hello altogether,

first of all I would like to say what great works several of you have done and posted so far. Marvellous! Some of your work looks so real that you would not expect it is AI generated.

I would like to do the same, especially with women in leather clothes, but I have no idea how to do it. Therefore my questions.

1. What software is needed and what can you recommend (also for 18+ content)?

2. How detailed must your description be to generate a picture?

3. What specifications must your computer fullfil?

Thanks in advance and I am alreday curious for your replies!

Yours

henands69

Hi henands69 I'm happy to hear you are interested.


There are 2 ways:

1. Software that runs on your computer.
2. Online service.

Some of us use software called automatic1111:
https://github.com/AUTOMATIC1111/stable-diffusion-webui
It is a bit technical to set up, but it allows you generate any 18+ content with no restrictions. Some online tools do not allow for nudity. Some of the other folks on this forum use online tools which I am not familiar with.

I use the stable diffusion 1.5 model along with realism checkpoints and LoRA files which I download from https://civitai.com/ . I guess the best way to understand is to try and find some youtube videos which explain automatic1111. It is a bit technical to set up.

There is a SDXL model which some are using a software called comfyui. I have not tried this.

The online tools are nice because they remove the complex setup. They are preconfigured with what you need to generate images. They have checkpoint models and LoRAs to choose from. Some online sites do have restrictions, and some do cost money to use.

Here are some of the online tools:

https://app.prodia.com/
https://artsio.xyz/
https://dreamlike.art/
https://getimg.ai/
https://lexica.art/
https://ocriador.app/
https://sinkin.ai/
https://visuali.io/
https://www.mage.space/
https://www.seaart.ai/

I don't use any of these, so I'm not sure what is good. Some of the other folks here could make a suggestion. I have heard a lot of the folks mention https://www.seaart.ai/ so maybe that is good.


It takes as little as a few words, and as much as a few sentences to generate the image. It could also be a comma separated list of phrases.

I have one example of a prompt which I used which is here:
https://umd.net/groups/ai-wet-hair

I think some of the other folks can answer this #2 question better. They are much better at getting wetlook images. It takes a lot of experimentation. It can be very challenging to get good wetlook results.


You will need a computer with a graphics card with a lot of VRAM. The more VRAM, the better. The minimum is around 4GB. I have a card with 11GB. I have a GTX1080Ti from 2017. The github page has more details about requirements.

The online tools are nice because it eliminates the hardware requirement.

It can be overwhelming to get started. There is so much information and so many different ways of doing it. If you want to avoid the hardware requirement and software setup, online tools might be good to try. Hopefully this helps

I'm using online services, as I do not have a decent GPU which can create images in reasonable time. Most of these are also using some incarnation of Stable Diffusion which could be referred to as an "industry standard".

The problem with these online services may be less flexibility in models and LoRas, and those are the most important part for good results. I made my first steps with an Android app called "Imagine" which was helpful to understand how the AI generation works. The first good results I got were from SeaArt, I think this will be worth a try as it offers a zillion of models. Don't get frustrated too fast however, it will need a lot of tries and playing around with different base models and LoRas to get decent results. Back then, SeaArt was still in beta and offered unlimited generations for free, which they don't anymore.

Right now I'm using graydient.ai as a paid pro member for all of my creations as they also have unlimited creations (instead of the credits for generation model of many other services) and are especially designed for 18+ images, which some others are very restrictive with.

However I would not recommend this for beginners as it has about the worst and most confusing user interface I've seen so far.

Thanks for your posts so far! I hope I can work it out.

1. What software is needed and what can you recommend (also for 18+ content)?

I use Stable Diffusion 1.5 on SeaArt and haven't tried anything else yet. I've only been doing this since June and don't have the time or skill to produce images of the quality that others are doing, so I've reached a crossroads in deciding what to do going forward. SeaArt doesn't allow NSFW (18+) content.

2. How detailed must your description be to generate a picture?

You can get good results with very simple prompts, but I've found as soon as you try to be more specific about what you want, the more you end up including. It ends up becoming very addictive, as there's always a sense of excitement about what your next set of images is going to look like! It usually ends up in a disappointment! A lot of patience is required as, at least in my experience, the failure rate is probably around 95%+. I've had a major struggle trying to reliably generate long skirts for example - the AI just doesn't want to do it very easily, presumably due to a lack of suitable training images in the models.

Bear in mind too that you'll get lots of physical deformities in the images - extra legs, too many fingers and similar things. It's just the way it is at the moment!

3. What specifications must your computer fullfil?

My PC is quite old so, unless yours is very old, I don't think you'll encounter too many problems. If you decide to go with SeaArt, you could initially try their VIP Lite 30-Day card which only costs $2.99 and lets you generate up to 300 "stamina" every day. Extra stamina can be bought separately. A typical high quality generation run of 4 images costs 12 stamina and an upscale for image sharing costs something like 8 stamina.

Give it a go - there's really nothing to lose!

Thank you!