Cinematic AI

AI art is a developing medium but in recent years it has made major leaps in technology. This section is to provide more context to my work in relation to AI video, but more specifically a sub-genre I’m calling “Cinematic AI”.

Animated AI has been around since the very beginning of AI generated images. Jediwolf did a great retrospective on the history of AI works minted on the Ethereum blockchain. The first animated AI work was minted by Robbie Barrat.

Jediwolf (@randomcdog) - Post from X/Twitter

“The first animated AI artwork ever minted on Ethereum (June 1, 2018) was "Latent Space of Landscape Paintings #1" by Robbie Barrat and is owned by nomadd. It's important to highlight that Nomadd has held this beauty in their collection since 2018!”

Braindrops has had animated works since its first drop with Gene Kogan’s “Brain Loops” and throughout its history. I was lucky enough to be part of their curation with my “Anthrocinematica” collection.

“Cinematic AI” is a sub genre of AI video that I think has really come into its own in the past year. It is AI video that feels more like a movie or film. It’s hard to pin point, but I’m going to focus on one technological advance to define it: image-to-video AI models.

There were many early AI works that were more image-to-image AI applied to each frame of a video that produced hints of cinematic elements. But the earliest paper that truly demonstrated image-to-video was DeepMind’s “Transframer” in March 2022.

Deepmind proved the concept, but then Runway shipped the first publicly available product that made that concept a reality with GEN-2 in July 2023. For me this was the breakthrough moment for “cinematic AI”.

Around the same time, Pika was released along with AnimateDiff from the AI open source community and many others.

The later half of 2023 saw a flurry of new features come out from AI video products. I think Runway was the most impressive with their camera control and motion brush features.

The big news of 2024 was when OpenAI announced Sora in February 2024. This was an impressive show of technology from the biggest name in AI.

Everybody wanted to get access to it, but it was only shared with a very limited number of artists, who created some spectacular short films with it.

But what happened next is something I did not except. It looked like OpenAI Sora was way ahead of everybody else in technology. But then Google showcased an impressive demo at I/O in May 2024 with Google Veo.

It would seem that it was a big company game for AI video, but in June a Chinese company called Kling AI not just announced, but launched an AI text-to-video product competitive with Sora on June 8, 2024. But it was only accessible in China.

Then four days later on June 12, 2024 Luma AI launched “Dream Machine”. And again it was not an announcement like OpenAI Sora or Google Veo, it was an actual full product launch. And Luma was available to the whole world.

And Luma was very good and competitive with OpenAI Sora, but it had one feature that the other products did not: image-to-video. This instantly transformed my AI video art practice.

Now what was unexpected was that it was not OpenAI Sora who launched first, but a small startup. Luma Labs brought the next major step in AI video and in particular for “cinematic AI”. It meant to me that AI video was still anybody’s game, not just the big players.

I hope this gave you a better perspective on the development of AI video and in particular “cinematic AI”. It is truly a unique opportunity to be living through such a major technological change and to be creator getting to play with all of these new tools.