Kinky AI

The photo heading this post is of me, and yet also it isn't. In real-life I don't generally dress terribly androgynously. The occasional secretive flirt with dark nail polish is as close as I ever get to transvestitism. I don't do weird clubs. Generally I'm rather staid, boring actually.

However, the beauty of AI image generation, particularly when combined with photo editing software, is that anyone with a little experience can re-invent themselves fabulously, with an alter-ego that they would never dare to indulge in normally.

The pink outfit was created with Stable Diffusion, using a front-end called Fooocus. After pasting a few lines of Python code into a Google Colab GPU (a service designed to run machine-learning models), within a minute or two a link is offered to a webpage GUI in which you can enter any prompt, and create AI imagery. My prompt asked for a bald man wearing a pink leather jacket, fishnets and stiletto heels.

To superimpose my head, I saved the image, and opened it in GIMP, an open-source app that can run on Windows, Mac or Linux. Using a warp-transform tool, I shrunk the head slightly, and pasted in my own, which I'd selected from a recent selfie, having removed the background using Adobe Express. I added a touch of virtual make-up using free-select, hue-saturation and paint, and by modifying a png of some false-eyelashes I found online.

To create a convincing join, I saved the image, and restarted the Stable Diffusion instance, uploading my image to the browser. Choosing the inpainting option, I carefully indicated the neck areas that needed to be joined. I click generate, and hey-presto, any mismatch is ironed out.

The intelligence of this free-to-use AI software is astounding, and with practice first-rate image editing is easy-as-pie. The only drawback is it can be a bit time-consuming, particularly when allowing the Colab instance to boot in or the ML-models to load.