AI generated fetish images

[Like Ra]

After the disastrous price point announcement for the RTX 4000 series

Yeah, Jensen is geting hungry...

[Anne]

@Bound Whore , can you share the prompts you used? That's some amazing results. Did you Photoshop them as well?

[Bound Whore]

@Bound Whore , can you share the prompts you used? That's some amazing results. Did you Photoshop them as well?

Those are all based on the prompt I posted on the last page, with some minor adjustments for colour or lighting (studio lighting, soft lighting, natural lighting). I like the current results, so I don't mess with the art style modifiers much, but that would most certainly open a myriad of new possibilities.

The "bubble pop" image came up unprompted, I only used "bubblegum pink" as the color qualifier for the latex suit because I wanted to get different hues of pink in my images.

Same with the latex bimbodoll, if you run enough iterations and adjust the variation parameter ever so slightly, it will eventually spit out something weird/intriguing.

I tried a prompt with latex encasement, shrink wrap, etc. and it couldn't come up with anything worthwhile.

As for post-processing, most of the time I just add some color adjustments for some vibrance, sometimes I retouch an extra limb or a deformity if I like the base image a lot (see also a couple of posts back, the yellow leotard with the deformed hand), other times I just run another 30+ variations of the prompt and hope for a cleaner outcome or something I can use to replace faulty image parts with.

There's also the possibility of using img2img, where I put together the best parts of different images or roughly recolor/paint over unwanted parts, save it as a source image and feed it into the algorithm.

If you want the prompt for a specific image I posted, just quote it to me and I should be able to check for the exact prompt in my logfile - close to 10,000 prompts now (including tons of variations, of course). But most are the base prompt from last page.

[Bound Whore]

I am sooo curious if you could do this properly with bondage positions.
Or different types of gags.
Or postures.
Or anything else super kinky.
Bare minimum would be to bring the faces of your favourite bondage models into the pool.

Still requires at least 24 GB of VRAM, but luckily those can be rented for $ 0.50 - $2.00 per hour (that's what you'll probably need for one "prompt term"). I'll be out of town for a couple of days, but after that, I'll definetly try and give this a shot.

Very detailed tutorial for Textual Inversion using Dreambooth.

Source: https://www.youtube.com/watch?v=7m__xadX0z0

[Like Ra]

Still requires at least 24 GB of VRAM

🙀 ... I did not have plans for such a rig ...

Oh, AMD is not supported, only 3090+

but luckily those can be rented for $ 0.50 - $2.00 per hour

Need to investigate this...

Very detailed tutorial for Textual Inversion using Dreambooth.

Yes, great tutorial indeed!

[krinlyc]

Still waiting for low/normal prices for RX 6700XT... 12GB should be enough, I hope...

Haha - Ah sorry - I dont think, that will happen soon.

[Like Ra]

Ah sorry - I dont think, that will happen soon.

Getting better. This is the graph for XFX RX 6700XT:

https://tweakers.net/pricewatch/1715722/...aming.html

Still at least 100 euros too much 😆

Intel Arc A770? 16GB!

[Bound Whore]

There's been an update to the distribution I originally posted an installation tutorial for on the previous page, which brings some long awaited new features, among others:

• Completely new WebGUI - launch with
  
• Support for inpainting and outpainting
  
• img2img runs on all k* samplers
  
• Support for negative prompts
  

Negative prompts should help to minimize deformities and extra limbs, that would be a huge timesaver and should help cut down some chaff.

Most of these features were originally introduced by the Automatic11111 distribution, which seems to have fallen from grace with the Stable Diffusion developers for adding support for a pirated/leaked version of the upcoming 1.5 model data file (which allegedly will have stricter limitations for NSFW content, so backup your 1.4 model checkpoint files just to be safe).

There's been some drama behind the scenes, Stable Diffusion (the company) has taken over the SD subreddit and purged all information regarding Automatic11111. Rumor has it they are even trying to implement code in ther model that detects that particular distribution to block it from using the 1.5 model database.

So I'm happy that those features made it to the repo I've been using.
Still need to run the update, I have been away for a bit.

[Like Ra]

Does the new distribution support the 1.4 model?

BTW, can you post a direct link to 1.4? I will add it to Retroshare.

[Bound Whore]

The 1.4 model is the only/current model, so all repos should work with it naturally.
If you have aeady setup any SD repo, it's the ~4GB model.ckpt file, in the case of this distribution, it should be located in stable diffusion\models\ldm\stable-diffusion-v1


It's available for download here: https://huggingface.co/CompVis/stable-di...4-original (the non-EMA-version)
You will need to create a free huggingface account, I don't have a direct download link that works without proper credentials, unfortunately. I would not suggest downloading a copy from shady sources, it contains executable/interpretable code after all. So by extension, I personally would not add it to RS, just keep a local backup copy.