AI Art

[littleweewee2]

Part 2...

[loneyheart]

Wow so perfect n fresh

[MontageEvening]

I mess this one up XD
Suppose to have her stretching her legs out, give me futa LMAO



However, this one....really surprised me
looks just like that one ML from Level 3 of Orchard Towers because the face looks the same

[asiachief]

nice,angling to this style now, trying to make some that look like my ex's

this one a milf filipino i knew who had moved to London when young but i flew her to SG regularly

[ladder49]

Originally Posted by littleweewee2

Part 2...

AAAAAAnd this is why I keep coming back to this thread! Forget songsonggao jurong.. this, to me, is the best thread in this forum and pictures like these is the reason why!

[littleweewee2]

I'm gonna share how I managed to generate this babe so those of you who are also on A1111 can try it out for yourself and hopefully get something similar.

Tool: Automatic1111
Below resources all can be found on Civitai.com
Checkpoint: I use BRA(Beautiful Realistic Asians) from Civitai. I like the asian look it provides. It's now on V5 but V4 gave me better output.

Positive Prompt: Largely to your own preferential input cos environmental prompts depend on what you want to see like clothing, lighting, background, filter effects etc.

Lora: This is the important one. I use Shojovibe v1.1 and [Character / Art Style] Fashion Girl v5.4 to achieve the final look. Weighted at 0.6 and 0.2 respectively. I also use epi_noiseoffset v2 to improve on shadow and lighting contrast though it's optional depending on the environment. These are part of the positive prompt.

Negative Prompt: I have been defaulting on the following textual inversions - badhandv4, DangerHawk and Deep Negative V1.x(75T). Taking up total 150 tokens, they are more than adequate to prevent overly bad renders.

Parameters: These are very subjective and up to you to trial and error.
Sampling method - DPM++ SDE Karras
Steps - I play around 30-42 steps
CFG - I play around 7-11. Some Loras require less, some more, so test it out.
Sizing - My initial txt2image churns are always 512x768 or vice versa, then upscale and denoise in img2img. Max i can go is 1024x1024 on low CFG and steps before I run out of VRAM. You should fare better if you have a top tier GPU.

Let me know if I miss out anything. I'll share what I know.

[WILHD]

Originally Posted by littleweewee2

I'm gonna share how I managed to generate this babe so those of you who are also on A1111 can try it out for yourself and hopefully get something similar.

Tool: Automatic1111
Below resources all can be found on Civitai.com
Checkpoint: I use BRA(Beautiful Realistic Asians) from Civitai. I like the asian look it provides. It's now on V5 but V4 gave me better output.

Positive Prompt: Largely to your own preferential input cos environmental prompts depend on what you want to see like clothing, lighting, background, filter effects etc.

Lora: This is the important one. I use Shojovibe v1.1 and [Character / Art Style] Fashion Girl v5.4 to achieve the final look. Weighted at 0.6 and 0.2 respectively. I also use epi_noiseoffset v2 to improve on shadow and lighting contrast though it's optional depending on the environment. These are part of the positive prompt.

Negative Prompt: I have been defaulting on the following textual inversions - badhandv4, DangerHawk and Deep Negative V1.x(75T). Taking up total 150 tokens, they are more than adequate to prevent overly bad renders.

Parameters: These are very subjective and up to you to trial and error.
Sampling method - DPM++ SDE Karras
Steps - I play around 30-42 steps
CFG - I play around 7-11. Some Loras require less, some more, so test it out.
Sizing - My initial txt2image churns are always 512x768 or vice versa, then upscale and denoise in img2img. Max i can go is 1024x1024 on low CFG and steps before I run out of VRAM. You should fare better if you have a top tier GPU.

Let me know if I miss out anything. I'll share what I know.

Thank you for the good advice. The up in quality from Hassan's Blend to the Civitai models is incredible and it only took around 6 months

[littleweewee2]

Yea so true. I'm incredibly awed by how fast it's growing and the POWER it gives to the layman. Literally anyone can do what we have all shared here in this thread. It's really exciting.

[Stezo]

It won't be long until we can create videos of our AI Babes.

[littleweewee2]

Originally Posted by Stezo

It won't be long until we can create videos of our AI Babes.

Oh it's already happening. But corporate idiots have to go and paywall the tools.

[sony2002]

Porn stars going to lose their job to AI soon

btw, any of these AI tools can work on video? change the face of my own sex video to another person?

[littleweewee2]

I have seen text to video generators and these are paywalled mostly.

I do know automatic1111 has the capability to create GIFs though, with some scripting help.

[littleweewee2]

Beauty is never wasted in cosplay...

[Ignite]

[littleweewee2]