Blendful—Styled, Custom Templates for Tailwind

Hi IH 👋,

My name is Nikk (hello! from Seattle). Around 2022 I started building Blendful (www.blendful.com) —a way for people to build Tailwind templates that were themed aesthetically according to their own design/brand preferences.

Tailwind templates on the web today are certainly aesthetically pleasing, but they all follow a single, unitary visual style. When individuals implement these templates on their website; it cheapens the brand—I'd say in a manner similar to using stock imagery. The templates look good, but they don't feel good; they're cookie-cutter, and users know that.

I played with the idea of being able to modify just the presentation layer. But I wanted to do it in a way that would treat the users with respect. Allowing too-granular stylistic control (allowing users to modify button padding, text line heights, border colors, et al) typically will get most users into a pretty bad place aesthetically. On the inverse, not allowing users enough stylistic control won't give them an aesthetic that they find meaningful.

I find that most template websites miss the mark when it comes to providing customization that is accessible, and produces good results. So I tried my hand at my own version, which you may or not like.

Part way through my journey, I started suspect that increasingly capable LLMs would eventually supersede my efforts in this area. Being skiddish and self-doubtful as I usually am, I decided to pivot to other activities that wouldn't lose ground to the seemingly endless territorial gains that were being eaten by language models.

But a few weeks ago, after a hiatus, a personal epiphany dawned. Language models are superhuman across a wide domain of activities now, but one area in which they continue to be quite weak is stylistic consistency. Even in just text, stylistic drift is a huge problem. But this problem is amplified when an additional burden is placed on the LLM, like producing HTML and CSS of a certain visual style.

Certainly, you can get an LLM to produce a good looking hero container for you. And you certainly can get the LLM to tweak it to look Victorian, Baroque, or the like. But getting the LLM to understand what a style is, and maintain it between templates it generates, doesn't seem like an area where LLMs are currently making a lot of progress. Certainly, fine tuned or specially trained LLMs could be quite capable of this, but I'm not necessarily certain that is an interesting problem space anyways.

After realizing that LLMs may not make inroads into this adventure in the near future, I decided to spend a final week polishing it up and, well, launching it as a sort of proof of concept. I'm curious as to what other community members think of this problem space, or my solution, or the intersection between this type of tool and newer technologies; so if anyone finds any interest in what I have built, or what I have said, I would love to hear from you.

Thank you so much for lending your eyeballs for the few mins.