Eye Contact Solution Slime

How Eye Contact Solution Slime Actually Works

I started using Eye Contact Solution Slime after dealing with a bunch of portrait shots where the subject's gaze was pointing slightly off-lens. Not dramatically off—just enough that it made the images look disconnected from whoever was viewing them. The tool takes a source face image and a target image, then warps the eyes in the target to match the gaze direction from the source. It's essentially a ganymed-style inpainting job focused on the periocular region. The setup isn't complicated. You need a source photo where the eyes are looking at the camera, and a target photo where they're not. Load both into the Slime interface, run the alignment pass, and let it generate. Takes roughly three to five minutes per image pair on a GPU with 12GB of VRAM. On a CPU-only machine it drags out to about twenty minutes, and the quality drops noticeably around the iris edges.

Eye Contact Solution Slime Setup Walkthrough

First, pick your source image carefully. The eyes need to be clearly visible, reasonably well-lit, and facing forward. If the source has sunglasses or heavy shadows over the eyes, the warp gets messy. I learned that the hard way on a batch of about forty headshots where the lighting was inconsistent. Run the automatic face detection first. The tool will box the face and then refine the eye-region segmentation. Pay attention to the eye landmark placement—sometimes the automatic detection misses one eye, especially if the head is turned more than fifteen degrees from frontal. You can manually adjust landmarks by clicking and dragging, which is faster than re-running detection every time. Set the strength slider somewhere between 0.6 and 0.8 for most cases. Going above 0.8 tends to introduce artifacts around the sclera and creates that glossy, plastic look that screams "I was AI-edited." Below 0.5 and the correction is barely noticeable, which defeats the point. The sweet spot depends on how far off the original gaze is. If the eyes are only five degrees off-center, 0.4 is enough. If they're looking off to the side entirely, you'll need 0.7 or higher.

Generate and review. The output comes out as a new image file. Compare it side by side with the original at 100% zoom, not thumbnail size. Artifacts hide at small sizes. Check the corners of the eyes, the catchlight reflection in the iris, and the skin texture around the eyelids. If any of those look smeared or duplicated, lower the strength and regenerate. Save the result. Export as PNG if you're doing further editing in Photoshop or GIMP. JPEG introduces compression artifacts that compound when you apply other retouching steps afterward. Here's something people don't usually mention: the tool works best when the target and source images have similar lighting direction. If your source has light coming from the upper left and your target has it from directly above, the warped iris highlight will look wrong no matter what strength you use. I spent an afternoon on a project where the mismatched lighting made every attempt look artificial until I re-lit the target photograph in post. It added forty-five minutes of work but saved me from redoing the entire batch twice.

Common Pitfalls and What Actually Fails

The biggest issue is extreme head rotation. If the target face is turned more than about thirty degrees from frontal, the eye warp starts pulling the sclera into the nose bridge or creating impossible eye shapes. The tool isn't built for that scenario. There's a workaround where you rotate the face to frontal in a separate step, apply the correction, then rotate it back, but that adds complexity and often degrades image quality through double interpolation. Just use a different method for severely turned faces. Another problem area is images with low resolution. If the eyes are smaller than a hundred pixels across in the source image, the landmark detection becomes unreliable and the warp lacks the detail needed for a clean result. I had a batch of wedding photos where the guests in the background were tiny, and trying to correct eye contact on those just produced blurry blobs. The practical fix is to skip small faces and focus on the primary subjects where the eyes take up enough screen space. The tool also struggles with heavy makeup around the eyes. Thick eyeliner, dark eyeshadow, or dramatic false lashes confuse the segmentation model. The output will often smear the makeup pattern or create halos around the lash line. If your subjects wear heavy makeup, you'll need to do some manual cleanup in a photo editor afterward, which means you're spending more time fixing errors than the original correction saved you.

One edge case I ran into involved a subject wearing tinted glasses. The tool detected the eye region behind the lenses but the color and refraction from the tints threw off the inpainting. The resulting irises looked like they were underwater. The workaround was to mask out the glasses area before running the correction, process the face without the glasses, and then composite the glasses back in afterward. It added about ten minutes per image but produced results that looked natural instead of broken.

Alternatives When This Isn't the Right Tool

If you're working with video footage instead of still images, Eye Contact Solution Slime won't help you. It's designed for single frames. For video, you'd need a temporal coherence pipeline that processes frame sequences, and those are significantly more complex to set up. I've used dedicated video-aware tools for that purpose and they require substantially more GPU memory and processing time per frame. For batch processing large volumes of portraits, the per-image manual adjustment adds up. If you have hundreds of photos to correct, consider scripting the workflow or using a tool that supports batch mode with uniform settings. Running each image individually through the interface gets tedious fast. There are also open-source alternatives if you don't want to use this particular solution. Models like EyeContact-GAN and various diffusion-based approaches can achieve similar results, though they typically require more technical setup and familiarity with PyTorch environments. For most people doing occasional corrections, the paid or freemium tools are more practical. For studios processing hundreds of images weekly, the open-source route pays off after the initial configuration effort.

The bottom line is that Eye Contact Solution Slime handles the common case well—slightly off-gaze portraits where the face is mostly frontal and the lighting is reasonable. It doesn't solve every problem, and it makes mistakes in predictable ways. Knowing those failure modes before you start saves time.