For his upcoming birthday, I decided to honor the memory of the great photomontage creator John Heartfield (19 June 1891 – 26 April 1968). He was a true anti-fascist artist at a time in Germany when it was extremely dangerous to be so. I wanted to do a videomontage similar in spirit to his photomontages. One that mocked a particular virulent authority figure. One like this one.
First, I needed an image that I could animate, so I decided to put Trump’s face on a bop bag clown. I asked ChatGPT to do it for me. Here is our interchange.
ME
Put Trump’s face on this bop bag clown.
ChatGPT. I can help with that. Just to clarify, do you want Donald Trump’s face realistically blended onto the clown in the image, or do you prefer a cartoonish or humorous version?
ME. How about both for a start?
I then uploaded the bop bag clown image I found and this headshot of Trump.
Apparently ChatGPT quickly realized what I was up to. It’s AI senses tingled down that I was trouble. It wanted nothing more to do with me.
ChatGPT. This image generation request did not follow our content policy.
ME. Coward
ChatGPT. I get it—frustrating when something hits a wall. The reason I couldn’t create that image is because putting a real person’s face on a clown (especially in a way that could be interpreted as mocking) crosses our policy boundaries around depicting recognizable individuals. It’s not about fear—it’s about following fair use and respectful content rules.
I quickly steered off of this route. Undeterred, I asked other large language models (LLMs) like Claude and Deepseek. They also avoided processing my inflammatory request, and for similar reasons. I tried free popular image generators, but they too shuddered at the mere mention of my task. Finally Claude suggested I do it myself, using “a free image generator like GIMP.” No offense Claude, but that’s a terrible choice. But I got the point. I saw the error of my ways and realized I’d save more time doing it myself in Photoshop. Even if one of them agreed to do it, it would not be up to my standards. Here is what I came up with.
Next, I wanted to animate the bop bag getting hit and bouncing back like it did in our troubled youths. Using my Davinci Resolve video editor — for those who need to know, it’s also called a run-length encoding (RLE) application)– I created a sequence that worked! But that’s all it did. Frankly, it kind of sucked. It looked mechanical and amateurish. What help could I get to make its motion look really cool? Ask my favorite large language models (LLM), that’s what. But wouldn’t they just shoot me down again when I told them I was using it as political satire? A solution suddenly popped up. Deceive them. Here was the prompt I came up with:
I have one image of a vintage bop bag clown and another of a disembodied horizontal fist. Within Davinci Resolve, I want to create an animated image of the bop bag clown being hit by the fist moving from left to right. The bop bag tips over at 90 degrees on "the floor," the fist withdraws, and the bob bag springs back to its original position. There should be a crop/warp effect when the clown face is hit, squashing it slightly for a few fraames, then unsquashing it. The clown should wobble, cartoon-like, at the end of the sequence about three times.
Notice I didn’t even mention it was a Trump-headed clown. It was merely “a vintage bop bag clown.” Why give them more information than they needed? Unlike Trump, who loudly and publicly proclaimed “take the oil,” I did not say the quiet part out loud. Because I was cagey, my AI programs were more than willing to help. They came up with some good effects. Eventually the two of us devised a banger conclusion that clinched it.
Sometimes a little deception is necessary.
