The AI Filter Trend That Keeps Popping Up On Every Timeline
You've seen the photos. Your boyfriend gets turned into a cartoon character, a cartoon cat, a medieval knight, whatever the current filter is. People post them everywhere. It's mostly harmless fun and it goes viral every few months. The problem is that most people try to force it and end up with weird warped faces, mismatched lighting, or something that looks nothing like either of them. I've done this with my partner probably two dozen times across different apps and services over the last couple years. Some work fine. Most don't. Here's how to actually get it right.
How To Turn Your Boyfriend Using AI Image Generators
The core process is straightforward: take a clear photo of him, feed it into an AI image tool that supports image-to-image generation, and add a text prompt describing what you want him to become. The quality of your source photo matters way more than anything else. A blurry or poorly lit picture will produce garbage results no matter how good your prompt is. Use a photo with even lighting, his face clearly visible, and ideally looking toward the camera. Side profiles and heavily shadowed faces are going to make the AI guess too much and produce inconsistent results. I run this through Midjourney most often now. The --style raw flag helps keep it from over-stylizing everything, and the --s (stylize) value controls how artistic the output gets. I usually keep it between 50 and 100. Anything higher and the result starts looking like everyone else's output. The image-to-image strength parameter, controlled with --iw, is where most people mess up. If you set --iw 2 or higher, the AI sticks very closely to your source photo's composition and features. Lower values like --iw 0.5 give the AI more freedom to reinterpret, which can actually help when your source photo isn't ideal. I once tried converting a photo of him taken at a dimly lit restaurant and the results were unrecognizable. Switching to --iw 0.3 and adding well-lit portrait to the prompt actually fixed it by forcing the AI to recreate the lighting rather than trying to salvage bad input. Stable Diffusion is the free alternative if you have a decent GPU or use a cloud service. The workflow is similar but you get more control over the denoising strength. Anything above 0.75 and the AI is basically redrawing from scratch. Below 0.4 and you'll still see his face but it won't transform convincingly. The sweet spot is usually 0.5 to 0.65 for a balance between recognition and transformation. ControlNet is optional but extremely helpful if you have it set up. The depth or Canny models preserve the pose and structure of the original photo while the text prompt drives the style change.
For people who just want a quick result without any configuration, there are web-based services like Lensa, Voila, and various TikTok or Instagram effect creators. These are fine for one-off fun but they're locked into whatever presets the platform offers. You can't tweak parameters, you can't refine prompts, and the output quality is usually lower than what you'd get from Midjourney or Stable Diffusion. I use them only when I'm in a hurry and don't care about the result being polished. The prompt itself should be specific but not overwhelming. Something like a photo of a man transformed into a golden retriever puppy, wearing a tiny sweater, photorealistic, soft lighting, 4k works better than a wall of disconnected tags. AI models respond more naturally to descriptive sentences than to keyword soup these days. Mention the style you want explicitly. Without it, the model picks its own default and it's usually something generic. One thing nobody mentions: the aspect ratio of your output photo should match or exceed your source photo. Generating in a smaller resolution and then upscaling usually introduces artifacts around the face area. If you need higher resolution, run the output through an upscaler like Real-ESRGAN or the built-in upscale in Midjourney rather than generating small and stretching it later.
Get the Full Details
These tools don't work well in a few situations. Photos with multiple people in them tend to distort everyone rather than just the subject. Extreme angles, especially looking straight down or up, produce weird facial geometry. And if your boyfriend has facial hair that changes frequently, the AI sometimes struggles to maintain consistency across different transformations of the same person. That last one is a real limitation, not just a quality issue. I've had prompts that worked perfectly for one session and then completely fail the next week because he trimmed his beard and the reference photo no longer matched the mental model the AI built. If you're doing this repeatedly with the same person, save a few high-quality reference shots at different angles and lighting conditions. Having a small library instead of relying on a single photo makes a noticeable difference in consistency.
