I’ve been really, really looking forward to Dalle-2. As of writing this, there is a waitlist (no invitations, they manually do approve people). I applied July 6, and it took over a month, being approved on August 10th. The AI was trained on millions of images. It knows photographers, painters, camera settings. The key to it is to be specific in your prompts, and from that, the AI takes over. Since AI is computationally intensive, and you only have 50 free credits, each prompt needed to count. The limit does make it somewhat daunting. Keywords like “medium shot” “close up” “long shot” seem to determine how far the camera is, and you can also specify camera settings. You can also use this clip software to *kind of* see if the AI might know what you’re talking about. For instance, I can see it does have entries for Nick Knight and paolo roversi but not Lindsey Adler, but that isn’t perfect. You also don’t want to prompt with negatives, as that can confuse the AI.

A high quality full body editorial fashion image of a redheaded woman looking slightly left wearing a red dress, key light to the left, Sigma 105mm F1.4 100 iso, white background, in the style of Nick Knight

Although it is a beautiful dress, the images that do have a face look very not realistic. I however, love the dresses and think they’re really good. The one on the right looks like it could be really in vogue.

Head and shoulder portrait of a red-headed woman wearing a red shirt in a studio, Sigma 105mm F1.4 100 ISO, Rembrandt lighting

The subjects aren’t particularly attractive, but it sure looks real. So, how can I get a full body image that includes the face? I’m wondering if the AI only saw ‘full body’ and thought it didn’t want a face. But otherwise, it’s starting to come together, and I love the lighting, particularly on the left.

A long shot studio photographic portrait of a beautiful red headed woman, Sigma 105mm F2.8 100 ISO, in the style of Paolo Roversi.

Okay, I’m kinda liking this, but I don’t see much Paolo in it. Paolo also does a lot of nudes, so maybe I should steer clear of this, as the creators were very careful about what they submitted to feed it. Also, wondering why it’s not a long shot. But I like where this is going.

A full shot studio editorial fashion of a beautiful woman wearing grunge clothing, Sigma 105mm, dramatic lighting

I am really, really into the outfits, but I am not sure where the faces went. It seems to do bodies OR faces.

Head and shoulder portrait of a beautiful woman wearing sunglasses with neon signs reflected in them, in an alley of a futuristic city, night time, Sigma 105mm f/2.8 100 ISO

I wanted to see what would happen if I changed from f2.8 to f/2.8.

These look a bit less professional but very real, but still kind of cool. It feels ‘snapshot like’ like someone took this image of their friends.

Head and shoulder studio portrait with red and blue studio split lighting of a beautiful woman in grunge clothing, blowing bubblegum, sigma 105mm F/2.8 100 iso, purple background.

I was going for Jill Greenberg here. Gotta say, I’m making some progress. I just can’t figure out how to get the full body in the shot, but I think I will need to do what is called outpainting (I’ll come back to this).

I am going to a keyword-based approach.

photo of slim girl, 20 years old, close-up, high detail, studio, smoke, sharp, pink violet light, studio, 105mm sigma art lens

I like how the fog looks, and the lighting is starting to add drama. There are nice shadows on the faces.

Photo of a slim girl, 20 years old, side profile, close up, high detail, studio, dramatic lighting, grunge fashion, smoking a cigarette, Sigma 105mm F/2.8

Okay, so this is starting to get a little more interesting. We can actually shrink the image in photoshop, put transparency around it, then Dall-e should be able to fill in the rest, this is what openAI calls what is essentially uncropping. Let’s see if it works.

I ran the same prompt again: Photo of a slim girl, 20 years old, side profile, close up, high detail, studio, dramatic lighting, grunge fashion, smoking a cigarette, Sigma 105mm F/2.8

and it gave me:

So, this got much less realistic. I’m going to try repainting one of my earlier grunge images that I liked the body of, and see if it does a better job getting a head from a body. Maybe I should have removed close up?

Photo of a slim girl, 20 years old, high detail, studio, dramatic lighting, grunge fashion, Sigma 105mm, sharp

After talking to people I learned that outpainting is better if you erase less, and don’t make the original too small, so I am doing one last try: A full shot studio editorial fashion of a beautiful woman wearing grunge clothing, Sigma 105mm, dramatic lighting

I’m done with outpainting, haha. Doesn’t matter how little I zoom out. It is awful. Need to look more into this.

Studio close up of a beautiful 20 year old woman, black and white, Sigma ART 10mm f/1.4 100iso, dramatic lighting

Not really fish eye, but definitely dramatic, and definitely interesting results. Onto part 3!

Blog

Posts tagged with AI-art

Ai Exploration part 2