WordPress / Tumblr Sells Your Data, Playground Drops with Open Weights, Agencies with AI

The world of creative work is getting better every day.

Feb 28, 2024

Morning y’all!

Today I’m sharing a handful of image and video projects as well as a few more newsworthy pieces that might impact you. Have a great hump day and stay productive in your generative AI explorations!

✌(-‿-)✌

— Summer

The first thing that caught my eye was news that Automattic, the parent company of the well-known WordPress publishing platform (I still have an installation or two around the web) and Tumblr are looking to sell their data to OpenAI and Midjourney.

Naturally, people are not happy about this development (and neither am I) and it’s a powerful reminder that anything you post on hosted platforms has a high probability (closer to 100% to be honest) of being sold as part of their underlying business.

Not cool, at all. But, sadly not surprising. 😡

Using krea.ai one user shared went from sketch to 3D. The time-savings on something like this is mind-blowing!

Being reminded that some of the more compelling and useful applications of artificial intelligence is when it’s combined with robotics is neat. The Esper hand being one of them (as seen above).

I’m also grateful to know that I still have the use of my two hands and that it’s entirely possible that I’ll need augmentation at some point.

Photoroom has been circulating the news articles because of their gigantic venture capital fundraise, $43m at a $500m valuation. Uniquely they are building their own foundational models. I covered their product previously so you can give it a go if this is something you’d like to try. I didn’t get great results at first blush, to be honest.

Alibaba demonstrated some new text-to-video technology called EMO, Emote Portrait Live, but I couldn’t find a real open source library or tool out there (anyone?). I did find the research doc though. Alibaba has a habit of sharing compelling demos but doesn’t release them so I’m not holding my breath in any way.

But the results are not terrible.

The founder of Playground shared the above video with some serious improvements:

Playground v2.5 improves dramatically on color & contrast, multi aspect ratio, and aesthetics to push image quality as high possible while not changing the model arch that the community has built tools on.

I’ve been tracking this for some time and the fact it has “open weights” is really great. It’s also beaten a lot of other models re: performance. This is one powerful image editor, no question about it.

FuseAI is an open source library that combines a number of LLMs into a single, pre-trained LLM for a better outcome:

While training large language models (LLMs) from scratch can generate models with distinct functionalities and strengths, it comes at significant costs and may result in redundant capabilities. Alternatively, a cost-effective and compelling approach is to merge existing pre-trained LLMs into a more potent model. However, due to the varying architectures of these LLMs, directly blending their weights is impractical. In this paper, we introduce the notion of knowledge fusion for LLMs, aimed at combining the capabilities of existing LLMs and transferring them into a single LLM.

Makes complete sense.

Here’s some interesting thoughts on how design agencies are using generative AI — the high-level ideas are as-follows:

For explaining ideas — Agencies are increasingly using AI as part of their creative process that leads up to that deliverable. One use, for example, is to explain ideas visually, both to fellow designers and to clients.
For art direction — Another common use of generative AI in agencies is for art direction. For example, before Poppins commission a 3D artist, they'll sometimes generate an AI image to give the client an idea of what the end result might be. This saves the client from spending a lot of money on something before they know it's going to work.
For being reactive — One thing that AI art has going for it is that it's very quick and easy to tweak it. And that makes it useful if you're creating client work on the fly that's based on evolving events and needs to be delivered quickly.
For storyboarding — It's helping us with the generation of storyboard images and it's adding that extra layer of slickness to visual generation, three-dimensional space creation for experiential projects and background image extensions for differing asset formats.
For research and summaries — It's been a great tool to summarize, pull insights and fill knowledge gaps. This ensures our strategy team and designers working on projects have a wide breadth of information to bounce off in order to achieve the best possible response.
For consistency and coding — AI has helped with expediting the development of AE scripts for multiple use cases. For instance, developing an internal script to help with content management and file structures. The script scans the project to ensure all imported files are placed neatly in the right folder on our drive.