Sticky to Image Widget

I was excited when I first stumbled upon this emoji model on Replicate. You can give it a text prompt, and it outputs an image in the style of an Apple emoji. I was so thrilled by the idea of creating a self-updating sticker gallery using generated prompts that I immediately bought a domain to build it.

UX Problems

After hacking on it for a day, I ran into several UX challenges:

Image Quality Issues

Detecting malformed images and NSFW content quickly became top priorities, with malformed images being particularly tricky to handle. Abstract prompts like "test" would generate unintelligible results, and some images would have missing corners. This is why I ended up downsizing the images - it makes these imperfections less noticeable in the demo.

Given these challenges, I realized that focusing on emoji-sized images would work better than full-sized stickers.

Limited User Customization

How much control should users have over their generated images? Services like Cloudinary offer on-the-fly image transformations that could help - like adding outlines to make images look more like stickers. But what other meaningful customization options could we offer? Right now, text-to-image still feels more like an exploratory tool rather than something you can really polish and confidently share.