Skip to content

Method for generating structured images without using diffusion models?

2 pointshxypqrdiscuss
On HN

The computational complexity of the diffusion model is very high, which means it may not be applicable in many practical scenarios.

If our target image is two-dimensional but essentially a combination of points and lines, with either a single color or a limited number of colors, are there any other frameworks to achieve the conversion between text and image, besides stable diffusion?

I can think of the following: 1. Combinatorial optimization methods for generating interior design plans. 2. Methods based on simple neural networks for matching text and images in plane geometry math problem

Welcome to add

Comments

No comments yet.

AboutSource Built by g1lg1l

Hackerly is an independent reader for Hacker News, built on the public HN API. Not affiliated with Y Combinator.