Comments URL: https://news.ycombinator.com/item?id=48940303

Points: 12

# Comments: 1

Comments URL: https://news.ycombinator.com/item?id=48870381

Points: 241

# Comments: 173

In the days since I've launched Mixfont, an AI font generator, I've gotten a good amount of feedback that's been skeptical of the quality of AI-generated fonts. I've also seen some articles shared discussing how high quality AI-generated fonts are still 3-4 years away.

I wanted to challenge these claims as I personally believe that the raster->vector font generation pipeline that's being used in Mixfont today is already good enough for the vast majority of use cases for decorative, display, stylized, etc fonts. While I think that creating a legendary variable font is not quite possible, I do feel that for most projects using AI-generated fonts is already high quality enough for many designers.

To prove out this hypothesis I've put together an AI Font Quiz, where some of these fonts are licensed from real font foundries, while others in this quiz are 100% generated with AI using the Mixfont model. You can try typing and testing out all of the fonts, and when you're ready you can select whether you think the font is AI-generated or human-made.

I would love to get your take on whether or not you can tell the difference between the AI-generated vs human-made fonts. I think that if AI-generated fonts are high enough quality, things will change quickly for the font industry in general.

I would appreciate your feedback and would love to hear what you score! Eric

Comments URL: https://news.ycombinator.com/item?id=48737639

Points: 4

# Comments: 1

Comments URL: https://news.ycombinator.com/item?id=48305915

Points: 2

# Comments: 0

For the past few months I've been working up to this launch of Free Fonts - it's a collection of completely free, open source, and original fonts that can be used for any project, including commercial ones.

The collection has over 400 fonts in a variety of novel styles, from simple sans serifs to decorative handwritten and blackletter fonts. It will also continue to be updated and grow over time.

Designers and developers today have essentially two options when it comes to choosing and creating fonts. Either use a Google Font, which can be limited in terms of unique and novel selections, or pay a significant amount of money for font licenses from a few large players in the space like Monotype. The problem today is that it can be difficult to find a novel, unique font for an affordable price. Our hope is that this collection can be a starting point for solving this problem.

What makes this project different is that every font in the collection was generated completely by AI, and is a novel set of letterforms. This is what makes it possible for me to launch the project with confidence that these fonts can be freely used in any type of project.

The AI workflow behind this project is also accessible on the same site. Our model was trained on a variety of images across the internet and then runs a full pipeline to transform a consistent set of letters into a single downloadable TTF font file. It can be prompted from a text prompt, or a sample image that contains an image of some text.

Font licensing is a complex world, and I deeply believe that it is ripe for a change. My hope is that this launch can spark a new conversation on the future of typography and how it can be freely and easily used by designers and developers over time.

Hope that you enjoy, and would love your feedback about the project and also the fonts themselves.

Eric

Comments URL: https://news.ycombinator.com/item?id=48230244

Points: 13

# Comments: 1

Comments URL: https://news.ycombinator.com/item?id=48215272

Points: 6

# Comments: 1

Comments URL: https://news.ycombinator.com/item?id=47417707

Points: 35

# Comments: 7

Recognizing the closest font from an image of text accurately has always been a challenge. Even SOTA vision LLMs can detect basic characteristics like if a font is a serif or a sans-serif, but when it comes to the actual font file that has the most similar styles, vision LLMs will often be quite off for the design and style of a chosen font file.

Popular font identification solutions on the internet today also have a two major shortcomings:

1. Most of them work based off identifying particular characters and don’t look at an overall word or image, so they require a manual character segmenting step. 2. Most online tools identify fonts from on a proprietary dataset of commercial fonts in order to monetize. However, this makes it difficult when building applications because often times the recognized fonts can’t be used or downloaded easily.

To solve this, I trained the Lens model on a large collection of open-source fonts (mostly Google Fonts, but also a few other open source / freely available fonts). The Lens model is also trained on a large collection of images, each which contain a variety of font styles and weights. As far as I can tell, Lens is the most accurate model right now that maps images to fonts in a single step.

I came up with the idea to create this model when I was playing around more with vibe coding tools like Claude Code and Codex. I noticed that when I would ask AI to implement a screenshot exactly, one of the main issues was always the fonts that the AI would choose. They would never match the screenshot exactly, and then swapping out the fonts for a usable alternative was always a pain for me personally.

Under the hood this is a classification model built on top of the Resnet 18 architecture. It is trained on over 1000+ fonts and 1.5m+ images.

The entire inference stack and model weights are available on Github: https://github.com/mixfont/lens. I also created a playground for the model on my website where it can be demo’d for free: https://www.mixfont.com/models/lens

Blog post on the training process coming soon! Would love to hear what you all think and some potential applications of a model like this. Happy to answer other questions in the meantime - thank you for checking it out!

Comments URL: https://news.ycombinator.com/item?id=47405761

Points: 3

# Comments: 3

Comments URL: https://news.ycombinator.com/item?id=43577490

Points: 415

# Comments: 708

Comments URL: https://news.ycombinator.com/item?id=43563023

Points: 5

# Comments: 0

We're looking for full-stack engineers that are interested in building a polished, consumer facing product that enables many content creation use cases. This job consists of everything from building the frontend tools to making sure our backend is scalable and reliable.

We're backed by top investors including CRV, KPCB, Shasta, Village Global. We have a culture of doing a lot of random things to grow our company and we write a lot about our startup journey on our blog: https://www.kapwing.com/blog.

I'm a co-founder, feel free to reach out directly if you're interested: eric@kapwing.com; or check out our open roles at https://www.kapwing.com/careers

Comments URL: https://news.ycombinator.com/item?id=41841193

Points: 7

# Comments: 2

I've gotten pretty worried about the developments in video deepfakes, and I wanted to raise some awareness about what's possible with fake videos today. I put together this quiz around real vs fake videos and I've been sharing it to see how well folks do. It's hard for me to tell how easy it is to complete it, given that I was the one who made the quiz, but I think the tech has come a long way and it's pretty scary how good it'll get. Especially in the context of the upcoming elections. Would love to get your thoughts!

Comments URL: https://news.ycombinator.com/item?id=41744910

Points: 6

# Comments: 3

Comments URL: https://news.ycombinator.com/item?id=41571184

Points: 2

# Comments: 0

I wanted to share a side project that we've been working on for a few weeks. It's called Tess and it's a new kind of AI image generator that pays artists when generations are made in their style. Artists can use Tess to create their own model, and choose to use it privately, or publish it publicly to earn revenue when it's used. The plan is to split generation revenue with artists 50/50.

We thought something like this would be inevitable in order to have artists be able to participate in conversation around AI art. We've paid out over $15k to real artists on the platform already, and we hope to grow these partnerships over time. Obviously not every artist is excited about AI, but we feel like this is a small step towards putting control back in the hands of artists.

Here's our launch post - would love to hear your feedback! https://www.tess.design/blog/the-first-image-generator-to-pa...

Comments URL: https://news.ycombinator.com/item?id=40233973

Points: 2

# Comments: 1

Comments URL: https://news.ycombinator.com/item?id=40047617

Points: 3

# Comments: 0