Adidas x Japan (concept series II) ????

Today, we'll dissect Chris Branch ?? 's (the writer of this newsletter) Generative AI workflow in his Series II conceptual designs for Adidas x Japan.

This series originally drew inspiration from Rafael Serra 's remarkable logo illustration shown below and was used as a reference image in Midjourney to create the artworks that have now been used to drive video.

As Matthieu Lorrain perfectly puts it, content is liquid.

What began as a simple logo concept shared on Instagram has blossomed into hundreds of Adidas garment designs. These designs were then integrated into an image-to-video workflow, leveraging platforms like MiniMax, Kling, Runway, and Sora to generate videos of models in a studio photoshoot environment - turning the logo into a full runway experience!

This journey started with Midjourney's Version 6 Alpha, which introduced style and character references around 10 months ago, enabling us to seamlessly integrate colour schemes, pattern consistency, and character continuity across our generations.

Generative video tools have advanced to the point where just a single reference image, whether through MiniMax's new "subject-to-image" workflow or Kling's AI Character builder, suffices for generating videos with remarkable coherence and continuity.

What's most striking is the tools' ability to predict the appearance of garments on the other side with no initial reference points.

It's eerie how accurate they've become - while some generations still lack coherence, occasional outputs are so realistic they're truly shocking, as seen in the example below.

A single reference image was given:

It is not currently possible to add an end frame image for the 1.6 Kling model, but that really doesnt matter too much...

Because via prompting you can have a very coherant video built regardless of being able to currently add an end frame to our generations, which is remarkable!

Generated video (converted to GIF):

With platforms like MiniMax excelling at style matching and using intelligent prediction to anticipate how the other side of garments might look, the need to designate an end frame is becoming obsolete, however it is still very useful for full art direction controlability but no doubt it will be here within the next few weeks!

The truly astonishing part of this evolution lies in the progress made by smaller teams.

MiniMax, originally a team of six researchers in China, has achieved in just nine-months what major competitors like OpenAI with 100x the resources are only just achieving too.

The results are nothing short of jaw-dropping.

With just one reference image (from the back):

Generated video (converted to GIF):

An interesting quirk about Sora is that it will not allow you to upload a person as an image as it will return the message "Error running this generation, media contain people"... however, if you upload a reference image from behind so that the face is not visible (which again is the only option right now, you can't add an end frame to Sora either), then you can generate video with it as seen in this video converted to a GIF below:

What do these advancements in AI workflows mean for us marketers?

In short, it means you can achieve a lot with a very small inspirational starting point.

AI has become incredibly sophisticated, capable of filling in the blanks where limitations existed not long ago. Don’t let the lack of a few features in the latest models hold you back from creating amazing content. The progress is astonishing - seriously, it’s game-changing how far this technology has come in such a short time since we did series I just 9 months ago!

Ready to explore further? Join one of our workshops

If you're not quite ready to take one of our AI workshops yet, instead why not subscribe to our Substack newsletter to stay updated on the weekly (sometimes even daily) breakthroughs in AI workflows to give you a head-start?