My Introduction to AI Image Generation
Discovering AI image generation was like opening Pandora's box for me. As an amateur photographer, it has transformed my approach to art. While I still do photoshoots I don't have to wait forever to scratch my artistic itch. Gone are the days of hiring models or planning detailed photoshoots; now, it's about the concept, the input, and the artistic output. The fusion of realistic and cartoon or rendered images offers endless creative possibilities. Often, the final result is a delightful surprise, different from my original idea but no less artistic. It's art, just evolved to a new level.
The Challenge of Age Representation in AI Art
The blending of anime styles with photorealism in AI art is fascinating but not without challenges. My personal preference, in part to my age, leans towards glamour-style images, influenced by the aesthetics of old catalogs and classic Playboy magazines. However, creating AI-generated images that accurately represent age, especially in this style, can be tricky. Even with specific prompts like "RAW, 50mm f 1.2, glamour photograph of a fit 21-year-old woman wearing intricately detailed lace lingerie," the results can vary significantly. Sometimes, the images appear much younger than intended, despite my efforts to include negative terms like 'child' or 'childish' in the prompts. My wife, with her better sense of age, often assists me in judging the appropriateness of these images. The inconsistency in results from online AI age detection tools further complicates this challenge.
Textural Inversion Techniques and Age Perception My exploration into Textural Inversions began with an attempt to recreate images of my favorite model, my wife. (Note: None of my uploaded TIs are perfect recreations of her, a decision I made both for technical reasons and to maintain privacy). Despite using reference photos of an adult, the AI-generated results often skew younger, showing the unpredictable nature of age representation in AI art. This inconsistency raises questions about the influence of the textural inversion process on perceived age.
Testing Images with Online Age Determination Services Relying on online age determination services has been an eye-opener. The age estimates for a single image can vary by as much as 7 years. This inconsistency has led me to adopt a cautious approach: if there's any doubt about an image's age appropriateness, I choose not to share it. It's a simple but effective rule to ensure that I stay within ethical boundaries. Personally I find it easier to just delete the image and move on to the next one, but when I spend a few hours cropping and captioning images for a Textural Inversion I find it hard to let go sometimes.
Navigating Legal and Ethical Boundaries
When it comes to legal and ethical boundaries in AI art, my principle is clear: if I find myself justifying an image's appropriateness, it's already a red flag. I try to adhere to the Terms of Service. While I'm not trying to generate minors I like to keep it in mind.
Concerning depictions of minors. The TOS explicitly prohibit content that promotes illegal activities, including child abuse, exploitation, and "Loli", "lolicon", "shota", "shotacon" content. Detailed TOS information can be found here.
Additionally, guidelines on depicting minors emphasize the importance of age-appropriate clothing, poses, and interactions, detailed here.
Again for me I'm not looking to create that.
Best Practices for Content Creators
In the realm of AI art, I advocate for personal responsibility and self-awareness. I don't think it is fair or appropriate for me to say what a best practice is for someone else. For me... If an image feels like it's pushing ethical or legal boundaries, it's best left unpublished. Each creator's viewpoint is unique, and it's crucial to maintain a balance between creative freedom and ethical responsibility. The subjective nature of art, coupled with the varying interpretations of age in AI-generated images, calls for a thoughtful and cautious approach. The reality is what I find fine my not be for someone else. I'd love for you to share how you approach it.
Conclusion A recent experience with a Textural Inversion that yielded younger-looking images than intended prompted me to adjust my approach, integrating images of an older model to align better with my ethical standards. In my workflow I tend to let the computer generate images of each iteration, and I usually train with outputs every 25 steps, so that means when I come back I'm looking through hundred of images trying to see what catches my eye. This last time I was really surprised to see the output looking so much younger than the reference images. If you've read to this point thanks, and I'd appreciate any insights you may have or workflow improvements.
I utilized ChatGPT 4 to assist with this post.