I have been spending too much time at Stable Diffusion which is an AI interface to a system that takes text prompts and makes images. It seems to automatically filter out nudity (so "nude" gets "Error" right away with no image) but it can be coaxed with words like "unclothed" or "undressing". It often generates gibberish and faces are mangled, but sometimes it makes real good images. It's also only 512x512 images, but hey, it's free. (Note that the processor has a queue limit of 20 (I think) and if you hit it at the wrong time, it'll just report "Error".)
For ASFR purposes, I found you can make a lot of content with keywords like "sex robot" or "animatronic". "Android" unfortunately gets lumped in with the Google brand ( ) so it doesn't work as well. An example prompt is "full view of an undressed sexy robot version of cute short-hair, young Michelle Williams laying at an electronic recycler". Sometimes adding more details gets you into territory that it doesn't understand, but in others, it works pretty good.
Also, it seems to draw from a corpus of images, so celebrities with more images work better. Unfortunately, some don't work so well (e.g. Margot Robbie is an angular-faced mess) while others are quite reliable (e.g. Rachel Leigh Cook) and others need qualifiers like "young adult" to avoid the too-young versions (e.g. Emma Watson).
Oh, and parts and panels don't seem to work at all. The AI has no idea how to interpret it.
AI Text-to-Picture site(s)
- jolshefsky
- Posts: 496
- Joined: Mon Jul 14, 2003 12:26 pm
- Technosexuality: Built and Transformation
- Identification: Human
- Gender: Male
- Location: Rochester, NY
- x 9
- x 25
- Contact:
AI Text-to-Picture site(s)
May your deeds return to you tenfold,
--- Jason Olshefsky
--- Jason Olshefsky
-
- Posts: 33
- Joined: Wed Jul 16, 2003 9:25 am
- Contact:
Re: AI Text-to-Picture site(s)
Stable diffusion works well on a Nvidia GPU with 8Gb, I can make image with 704x512 pixels.
Best, it's open source and it was fork.
Waifu-diffusion was fine tuned on danbooru and the fork does not have the filters.
https://huggingface.co/hakurei/waifu-diffusion
There is also the robo-diffusion fork, the name is pretty clear
https://huggingface.co/nousr/robo-diffusion
Best, it's open source and it was fork.
Waifu-diffusion was fine tuned on danbooru and the fork does not have the filters.
https://huggingface.co/hakurei/waifu-diffusion
There is also the robo-diffusion fork, the name is pretty clear
https://huggingface.co/nousr/robo-diffusion
-
- Posts: 66
- Joined: Fri Dec 09, 2011 5:44 pm
- Technosexuality: Built
- Identification: Human
- Gender: Male
- x 1
- x 7
- Contact:
Re: AI Text-to-Picture site(s)
I have been creating fembot art for a number of years using Poser. I switched to VAM (Virt A Mate) about a year ago and in the past several weeks I have downloaded Stable Diffusion and started working with it. So, as of lately, I have no free time and I might go broke from not working when I should (I am self-employed and work from home). After a LOT of trial and error, I finally found some prompts that can create fairly nice fembot images. I have posted a few on my Deviantart page, the link below will take you to that page and the last 13 images I posted were created in Stable Diffusion.
I will definitely look into the Robot model and see if that helps.
https://www.deviantart.com/mermaidhunte ... nd-sexbots
I will definitely look into the Robot model and see if that helps.
https://www.deviantart.com/mermaidhunte ... nd-sexbots
Users browsing this forum: No registered users and 20 guests