The 2d Art Thread

[rovingjack]

I gotta find better hosting for images, as these link go dead quickly.

[SeVeNeVeS]

I gotta find better hosting for images, as these link go dead quickly.

https://imgbb.com/

[rovingjack]

[J. Wilhelm]

Not related to Valkyrie (I'll start a new thread for that).

As an experiment I asked Stable Diffusion to give me a photo of 1920s actress Louise Brooks wearing a 1999 silver dress made by Prada.

Prompt: "Photograph of 1920s actress Louise Brooks wearing a 1999 silver dress made by Prada"
Negative Prompt: "Cartoon"
Generated with: Deliberate

Reference Image: Someone who looks very different; model Hunter Schafer in the 2021 Met Gala, wearing a 1999 silver dress by Prada:
https://www.popsugar.com/fashion/hunter-schafer-prada-look-met-gala-2021-48505855

I didn't get the Prada dress, but Louise Brooks came out loud and clear in a silver dress on the first try without a portrait of her! I got 5 images, out of which three were usable. How well somebody's face is known in the media affects whether Stable Diffusion will return a face you'd recognize. The image in this case is almost perfect (the black and white picture) except for the eyebrows which were short and straight on the real person.

[SeVeNeVeS]

Ive been playing with https://dezgo.com/
Generated with:AbsoluteReality (realistic)
Prompt: nubian/auburn female on a beach
Negative prompt: Cartoon, unrealistic etc etc.

Not perfect but some results on a female face are pretty good.

Spoiler: ShowHide

If I can get the basics to look realistic then its time to experiment with more creative prompts.

[J. Wilhelm]

Ive been playing with https://dezgo.com/
Generated with:AbsoluteReality (realistic)
Prompt: nubian/auburn female on a beach
Negative prompt: Cartoon, unrealistic etc etc.

Not perfect but some results on a female face are pretty good.

Spoiler: ShowHide

If I can get the basics to look realistic then its time to experiment with more creative prompts.

It's not as easy as one might think to get something very specific, because the AI program is also training on people's stereotypes and lack of knowledge as well.

For example, I tried getting a similar result with Brooks' (slightly older) contemporary, the Mexican Flapper, artist and muse, Carmen "Nahui Ollín" Mondragón. The AI refused to give me anything other than a stereotypical Latin American person, in spite of providing a specific name and a specific decade, plus a portrait of Mondragón! Stable Diffusion gave me 20 faces, not one remotely looked like her. You see, Mondragón didn't look like the typical stereotype that people have in the First World have about Latin Americans. The AI ignored the portrait, the decade, and didn't even try to search a face with the name. Instead it looked for people who could be named Carmen according to the public submissions!! In terms of a weighted classification, I guess there aren't enough portraits of Mondragón to overcome the cultural stereotype!

Spoiler: ShowHide

Reference Photos of Carmen Mondragón

The third image did give me a very good picture of an elderly Inca priestess though! 

I'll try again. After all, the software keeps training constantly.

[SeVeNeVeS]

I saw this on BBC news this morning https://www.bbc.co.uk/news/uk-65932372

WHY has no fail safe protocol been administered to prevent such things including deep fake porn?  I really dispair sometimes............. Uuuummm, couldnt it be a little obvious that certain people will try and create such images.