Generative AI - Short & Sweet 04 - ?? AI Outpainting

Sign Up | GAI Course | Sponsor

[Migrating my newsletter. This is from the 23rd of August.]

We’ve talked about the stunning results of?AI Image Generation. In this episode we go a step further, wondering if this might actually be real creativity.

???The GAI topic of the week is ..

AI (Image) Outpainting. AI’s most creative effort to imagining the surrounding context of an image and generating whole worlds. It is the opposite task of the better-known inpainting.

What’s inpainting??The task of recovering spatial information by filling in cropped-out regions of an image.

And outpainting??It’s the same functionality, but first shrinking an image (via e.g. Photoshop), saving it as PNG with the transparent surrounding area, and filling the transparent area around it (inpaint) with a specified prompt.

????And, animations??Further, if you repeat the process by using the new result each time you can combine them into a coherent video. For this, you could use software like Adobe After Effects. (Many videos on Youtube.)?

One great result I almost choked on eating my toast is?this?stunning animation from?Adam Pickard.

Another great example is from?Gragland?- (Guess the outpainting prompts, and find them in the comments?here):

???What could this mean??

• Informing creative processes:?With just an intro image (perhaps synthetically designed) and a precise prompt, multiple designs can be created to further iterate on - provoking broad ideation. This could be around building architecture/car/boat/airplane/app/garden/
• furniture/interior design, and so forth. What else??
• Virtual worlds and stories:?Currently, GAI models can create a coherent picture of anything that exists or doesn't exist in seconds. Fast forward, it becomes short videos with coherent movement, then high definition, then character recurrence, then with an actual plot/story, then with the right audio, and then it runs in real-time. Along the way, we might achieve multiplayer, and 3D. Just imagine the sheer amount of possibilities.
• 2 levers for creators:?There might be 2 ways to retrieve a creator-unique outcome.?

1. Prompt engineering?will become a valuable skill to walk the balance between being specific and room for some freedom for novel points of view.
2. Specific datasets?for fine-tuning pre-trained models. Have your unique dataset provide outstanding results.

Social media comments suggest that this will cheapen the value of art. This fear repeats itself, see e.g.?this TED Talk?about music sampling.

?? There are so many steps to creating a wonderful image. You could use this for collage or montage. Fuse it with other programs. Throw it into After Effects or Photoshop. Chop, churn, slice and dice. It won’t replace artists (that are willing to utilize GAI) rather it adds to the creative flow.?

What is your take on this??

?

Subscribing to, giving feedback about, and sharing the?newsletter?as well as our?renowned online course?will be highly appreciated and helps a lot. ??

For all (other) matters, just respond to this email.

Thank you for reading,?

Martin