Artificial Intelligence Training with Photographs

Artificial Intelligence Training (AI) comes down to either coding your own AI or becoming really good at AI prompts.

While prompts sound like the best way to go, if a combination of quality and speed is your priority, selecting one of the many AIs available is a challenge. Once you’ve chosen the AIs you plan to use in your workflow, learning the tricks of getting the exact image you want is your next challenge.

For images, my current focus is on MidJourney AI. To use this AI requires having a Discord account and basic knowledge about how Discord works.

Artificial Intelligence Training with Images

Suppose you’re interested in writing your own AI program, a project I’m working on. In that case, some good classes include Andrew Ng’s DeepLearning. 

On the other hand, if you want to get started faster, then learning everything there is to know about AI prompts is essential (AI Prompt Engineer).

Note 1: AI prompts tell the program how to generate an image. While prompts are based on conversational English, simply typing a paragraph of information could result in an interesting picture. Still, it may be far away from what you envisioned. For now, prompts require some limited coding syntax to get accurate results. Here’s an elementary example of a basic prompt in MidJourney.

Note 2: AIs are all inherently “trained” based on their initial training data. Just as kids, we learned the basics of life during our first few years. And these experiences went on to shape who we are. AIs are similar in the sense that the training data shape their output. AIs do iterate, but what you receive as the final image with prompts will be based on their early experience.

Concept: City at Sunset

Here’s an example of a MidJourney-generated image based on the following prompt:

Full body beauty shot of a 30-year-old female model, skin of a baby, no water drops, blond hair, blue eyes, wearing a black dress, downtown Cleveland waterfront city background at sunset, fog on the water, cinematic lighting, contoured shading, shimmer pearly color, gauze latex, depth of field, photorealistic, 8k, –ar 16:9 –v 5

Unpacking what some of the less obvious parts of this prompt are doing:

1. Skin of a baby:  There’s a glitch in MidJourney.  As a matter of fact, glitches are all over any AI and finding workarounds is part of the process.  In most cases, MidJourney ads percerpation and acne to faces.  Why?  Don’t know.  But, this text usually solves this glitch.
2. No water drops: This helps to alleviate the same glitch as adressed in skin of a baby.  Unless you want your final image to include persperation and acne.
3. Downtown Cleveland waterfront city background at sunset:  Much like a script, this tells the AI the background setting for the final image.
4. Cinematic lighting, contoured shading, shimmer pearly color, gauze latex, depth of field, photorealistic:  These help the final image to look more like a photograph if that’s your goal.  On the otherhand, if you’re trying to create a cartoon, painting, or sculpture, don’t use the majority of these prompts.
5. 8k:  Just like in video, 8k tells the AI to create a high resulution image.
6. –ar 16:9:  This is the aspect ratio of the final image (landscape).  You can set this to anything that you want.  For portriat the prompt would be –9:16.
7. –v 5:  This is presently the latest version of MidJourney.  There’s plenty of glitches in this version, so you may want to try –v 4 for the previous version if you’re running into rendering issues.

Here’s one of the resulting images:

Concept:  Glamour Goth

City at Sunset’s resulting image could use some Lightroom and Photoshop adjustments. Still, even directly from the AI, it’s a helpful image in various settings.
To take realism to a higher level and have more control over the resulting images, I’ve trained the AI using photographs I’ve taken. This results in considerable improvement.

Here’s the MidJourney prompt:

Close-up beauty shot of a female model, cinematic lighting, contoured shading, shimmer pearly color, gauze latex, depth of field, photorealistic, 8k, –ar 16:9  –v 5

Training image for the foreground

Training image for the background

Here’s one of the resulting images:

Artificial Intelligence and Artists

Where does this leave the artists who are usually involved in the making of a photograph? In my earlier article, I talked about the Group of 5 who will do well.

Currently, if you want or need maximum control over the final image, then photographers, lighting designers, models, wardrobe, costume, and HMUA (hair and makeup artists) will continue to have a place in the creative process.

But, if clients are ok with giving AI Prompt Engineers some leeway in the final image in return for a much faster delivery time and lower cost, then the need for traditional artists is significantly lower. Someone good at this new position will also need to know how to develop a concept and use the various programs required to produce a final image.

I forecast that unique-looking models, similar to my music analogy about singers, will continue to have a future.  

With that said, there will continue to be a need for photographers, wardrobe, costume, and HMUA in industries that cater to capturing a client’s experience.  These include:

• Birthday parties
• Corporate headshots and team photographs
• Vacation journaling
• Weddings
  • Note:  This industry continues to grow

 

Photography credits  |

Concept, photography, AI Prompt Engineer | Tom Libertiny (Cover)
Image | MidJourney AI