Creating Pennywise Reveal Images: A Practical Guide
If you've been scrolling through social media lately, you've probably seen those images showing Pennywise the Dancing Clown slowly pulling off his makeup to reveal something monstrous underneath. These have become incredibly popular, and a lot of people are trying to recreate them. Here is how it actually works, what tools to use, and where most people mess it up. The effect you're seeing is almost always generated using AI image tools, specifically ones that support image-to-image processing or inpainting. The workflow takes a reference image of the clown makeup and then uses a prompt to gradually replace areas of the face with something more grotesque. The "unmasking" effect is achieved by using masks or region prompts that target the face area while leaving the costume and background mostly intact. The most common tools people use for this are Stable Diffusion with ControlNet, Midjourney with its vary (region) feature, and certain web-based AI art generators that have a dedicated "unmasking" or "reveal" filter mode. Each approach has different strengths. Stable Diffusion gives you the most control but requires a decent GPU or access to a cloud service. Midjourney is faster for quick results but gives you less precision over exactly which part of the face gets revealed. Web generators are the easiest to use but often produce generic results that look the same across every prompt you throw at them.
I spent probably two weekends last fall going down this rabbit hole after seeing the trend blow up. The first thing I learned is that the source image matters more than anything else. If you start with a low-resolution still from the 2017 or 2019 films, the output will look muddy and indistinct. Grab a high-quality promotional still or a frame where the lighting hits Bill Skarsgård's face clearly, and you get significantly better results. I also found that adding the word "detailed skin texture" to your prompt makes a noticeable difference, because AI tends to smooth everything out unless you explicitly ask for roughness. One specific problem I ran into that almost made me quit the whole thing was that the AI would consistently distort the teeth when generating the reveal. Every single attempt produced either fused teeth, extra rows, or just a blank void where the mouth should be. The workaround was to use a two-step process. First, generate the unmasked face without focusing on the mouth area. Then, in a separate pass, use inpainting to add the teeth separately with a very specific prompt like "row of yellowed human teeth, slight gaps, horror aesthetic." Combining the two outputs gave me something that actually looked right instead of the usual AI nightmare mouth. Here is the step-by-step process I ended up using regularly:
Step 1: Find or create a base image of Pennywise's clown face. Use an original photo if possible, not a fan edit, because the AI models are trained on real photography and respond better to it. Step 2: Load the image into your chosen tool. If you're using Stable Diffusion, set up an inpainting mask that covers the lower half of the face area. Leave the eyes and forehead unmasked if you want a partial reveal, or mask the entire face for a full transformation. Step 3: Write your prompt. Something like "underneath the clown makeup is a terrifying pale creature with wet skin, sharp teeth, and glowing eyes, horror movie still, photorealistic, detailed." Keep it descriptive but not overly long. AI models tend to get confused when you pile on too many descriptors.
Get the Full Details
Step 4: Set your denoising strength. This is the number that controls how much the AI changes the original image. For a partial unmasking effect, aim for 0.5 to 0.65. Anything above 0.7 and you'll lose the resemblance to Pennywise entirely. Below 0.4 and the reveal won't be dramatic enough. Step 5: Generate and iterate. You will probably need 5 to 10 attempts before you get one that looks good. Save the seed numbers from the ones that work so you can tweak them later instead of starting over. There are some things you should know before you dive in. The biggest limitation is that AI generators struggle with symmetry and consistency across multiple attempts. If you generate ten versions, they will all look slightly different, and sometimes wildly different. There is no guarantee you can reproduce a specific look. Also, these tools don't actually understand horror in any meaningful way. They're combining visual patterns they've seen before, which is why the results often lean toward generic "scary monster" rather than anything that captures the specific dread of King's creature. The entity in the book is described in ways that no image generator can truly capture, and that's fine, it's just a fact.
Another issue is that some of the free web generators have become so oversaturated with this trend that their models are now producing very similar outputs. If you've noticed a certain house style in these images, that's because multiple services are using the same base models and the same training data. To stand out, you need to put in the effort to customize masks and prompts rather than relying on a one-click filter. If you want a download link for the tools, Stable Diffusion is free and open source at stability.ai or can be run locally through interfaces like Automatic1111 or ComfyUI. Midjourney requires a subscription starting at about ten dollars a month. For the completely free web-based route, there are several options available through searches, but I'd caution against putting personal photos or sensitive work into random online generators, since you generally have no idea where that data goes. The whole process from finding a source image to getting a usable result usually takes me about twenty to thirty minutes if I know what I'm doing, or about an hour if I'm trying something new. That's realistic. Anyone telling you it takes five minutes is either using a preset filter that looks generic or they're lying.
I should also mention that some of these tools have started implementing content filters that block horror-themed prompts. If yours keeps rejecting your requests, try rewording. "Eldritch horror entity" sometimes gets through where "scary monster face" does not. It is annoying and inconsistent, but it is the current state of things with most commercial platforms. The underlying model behind most of these generators is fundamentally the same technology regardless of which website you use. Understanding that helps you troubleshoot when something goes wrong. If the image looks blurry, it's usually a resolution or denoising issue. If it looks weirdly symmetrical or plastic, it's a model limitation. If parts of the image are missing or warped, it's a masking problem. Knowing which category your issue falls into saves a lot of time. There is also a growing community around this specific type of content on forums and Discord servers where people share their seeds, masks, and prompt formulations. Joining one of those can speed things up considerably because you can start from someone else's working configuration and modify it rather than building everything from scratch.
Just remember that these are approximations. No AI image will ever match the actual description of the creature from the novel, and that is not a flaw in the technology, it is a limitation of the medium. The book describes something that is fundamentally incomprehensible to human perception, and putting that into a JPEG is always going to be an interpretation at best.
