Generative as a Service
In 2022, robots are after the jobs of creative people. Illustrators and copywriters are now rivaled by AI: Likes of DALLE-2 and Copy.ai are fast, cheap and, believe it or not, versatile.
This post is about putting this kind of power to end users such as us rather than their technical capabilities.
Generative Art — Unbelievably Good
It can be argued that generative art was with us since a long time. Though, in my opinion, introduction of Generative Adversarial Network marks the start of the computational creativity era.
It started with blurred images, mutated shapes and “abstract” art. As data, algorithmic capabilities and computational power increased so did the quality of the images. There were really nice models such as VQGAN+CLIP and GLIDE. But something changed in the spring of 2022 and escalated quickly.
Here is a quick summary of the recent developments:
- DALL-E 2: It is introduced in April 2022 by Open AI. It was the very first algorithm that actually created the “WOW effect”. This post’s image is generated by the same algorithm.
- Google’s response was not far away. Imagen and Parti are announced shortly after DALL-E 2 in May and June.
- Meta AI took it to the next level by announcing Make a Scene in July. You are not constrained with text prompts anymore and you can assist the AI with sketches and it gives you more control over the creative process. Bonus: See the video which Refik Anadol and other artists creating stuff.
Currently, DALL-E 2 is the first commercially available, most convenient and powerful alternative. There are other examples such as Midjourney and Elbo and you can try DALLE Mini on Hugging Face.
Update: Stable diffusion is publicly available to plug in your codebase.
Copywriting - Let’s hear it from the AI itself
Copy.ai is the latest hype for generating meaningful text. Based on Open AI’s GPT-3 language model, they built a AI copywriting startup. Read their brief story in their Series A announcement.
Language models went a long way from Sokal affair to multibillion parameter translators, generators, conversation bots etc. I do not have a comprehensive list but we are using them in many applications.
But here is a short demo. I used Copy.ai’s generative model to contribute to this blog post.
Prompt to Copy.ai
Copy.ai’s Response
The algorithm generates several alternatives but I picked the first one for you.
“AI to generate art: It’s the future of the arts!
Generative art is a new way to create beautiful, original artwork. And it’s not just for artists — you can use it too!
Art is often thought of as something that requires special training and years of effort. But with AI-generated art, you don’t need any training or experience — just let the software do all the hard work for you.
You can use this software to make your own paintings, murals, or even sculptures. You’ll get high-quality output at a fraction of the cost of traditional methods. And the best part? It takes just minutes to produce each piece!”
Real good!
Bonus: Programmers Beware! Github Copilot
In June 2022, Github announced general availability of its code assistant tool Copilot after almost exactly one year of its initial technical preview announcement.
It functions very similar to generative art. You give a text prompt and the model writes functions for you. It can be argued that the end product has functional value and there will be technical limitations. Nevertheless, it is great assistance and perhaps it is even an educational tool.
Costs
Currently all those services are commercially available for quite affordable prices and they offer free tiers.
- DALL-E 2 (in Beta) costs $15 for 115 credits.
- Copy.ai has a $35/month Pro subscription package.
- Github Copilot subscription is $10/month or $100/year (no free tier)
Conclusion
The first public victims of artificial intelligence were chess players, then Go players and then professional photographers. They learned to adapt and they collaborate with the machine rather than going luddite. Same principle applies for illustrators, copywriters and programmers. It is just another tool to (super) enhance your capabilities as a human.
Even though such algorithms have a merit on their own, it is possible achieve much-MUCH more with human-machine collaboration.
ps. Check out similar services from this blog post.
https://www.linkedin.com/in/berkorbay/
