LookingGlass: Generative Anamorphoses via Laplacian Pyramid Warping | Hacker News Reader

LookingGlass: Generative Anamorphoses via Laplacian Pyramid Warping

studios.disneyresearch.com

129 pointsjw122429 comments

https://www.youtube.com/watch?v=B3AZay43688

This reminds me a bit of a project I am currently doing where I swap adjacent pixels in the last image of short video sequences until it resembles the first image:

https://www.instagram.com/marekgibney/reel/DILksFYNSkE/

That's quite an extraordinary effect. There's a point where enough pixels have been swapped that you can see the original, and it jumps out, and thereafter the image is quite surreal before eventually being 'tinted'.

(Also shows the video compression in the final frame well, because it doesn't match the first frame. Just found that interesting.)

Thanks. Yes compression arteficts always become very visible. Especially in low-detail areas like the sky.

What is the algorithm you are using for replacement / swap?

I have written up a description here:

https://www.instagram.com/marekgibney/reel/DDezhSisD4J/

This is really neat. The videos give me an instant headache. It's quite exciting to feel such a palpable and unusual effect from a video. Until now I wasn't aware that a video could give me a headache. I wonder what other physical sensations are possible from video? My compliment is genuine.

This reminds me of Stand-up Maths and Steve Mould creating a puzzle that has two solutions with two different images: https://www.youtube.com/watch?v=b5nElEbbnfU

If this interests you, you might also like the below:

https://diffusionillusions.com

https://dangeng.github.io/visual_anagrams/

There are also videos with the process explained

It looks like this could be used for some form of steganography. You need the right mirror form for decoding.

laplacian pyramid warping or just anamorphosis?

'anamorphic encryption' schemes have been used for hundreds of years.

Can you give a specific example? The term “anamorphic encryption” was used in a paper from EUROCRYPT 2022 and I don’t see any prior formal study of the subject. A follow-up paper from PoPETS 2023 establishes that most of the widely-used cryptosystems introduced over the past few decades are indeed capable of anamorphism but nobody used the term “anamorphic encryption” until the last few years.

To say that they have been used for “hundreds” of years seems like a sweeping generalization in need of better context.

Da Vincis mirror writing? Sure it didn't use the exact word but the concept is there.

https://www.openculture.com/2017/11/why-did-leonardo-da-vinc...

Not trying to be argumentative, but mirror writing is neither encryption nor steganography. The concept is not there.

Using the mirror to make it difficult to transcribe seems like a similar concept. Sure it's not exact phrases you'll find doing "google research"

My PhD research is in steganography. I was familiar with those papers because I’d read them before.

When I saw the video I wondered how many layers of refraction you could achieve while maintaining coherent images. The pixelated style they should might be the easiest use to develop this from.

it's all pixelated. Some is just more pixelated.

I love how many optical illusions have been revisited with new generative techniques

Are there other examples? Definitely curious to see more.

Visual anagrams popped up last year using similar, though simpler methods to those in the posted. Flips, internal rotations, rearrangements, color negatives etc. [0]

Diffusion illusions did something similar at about the same time but with puzzles and multi-image color layer mixing. Some of the double puzzles they made are a lot of fun. They have great explainer videos [1] on that site including with Steve Mould.

Also there are diffusion double/hidden images using qrmonster, illusion diffusion, control net, or img2img that have been making the rounds. [2] for a random example. These work by using a fine-tuned diffusion model[3] to take an image and use it as a structuring element at various levels of following to a generated image. To see these illusions, squint or move the screen away. These are quite a bit more popular and easy to make than the other methods so many more examples show up around the internet.

[0] https://dangeng.github.io/visual_anagrams/

[1] https://diffusionillusions.com/

[2] https://www.reddit.com/r/StableDiffusion/s/nm9QMV6roD

[3] or a base model in the case of img2img

Happy to see that Disney is still as active as always in research

Oh great. Now I suddenly feel the urge to acquire a physical oddly shaped lens or mirror.

Disney invited me to talk about my GenAI startup and research in front of a bunch of their execs across ABC, ESPN, Pixar, Streaming, etc. All of their folks were super nice and gracious to our small startup except for one.

Steve May basically scoffed at how little my small team could accomplish. Mind you we were using mocapped skeletal animation and object animation curves to fully steer video diffusion over a year and a half ago. Before image to video modalities. He picked apart our training and engineering and gloated that they could do better.

The incident is seared into my brain.

I can't help but think of Disney as the Empire and Pixar as the Death Star.

So the classic Silicon Valley brain rape is actually real?

> gloated that they could do better.

did they? Anything good come from that meeting?

If he was upset - it wasn't how little your team could accomplish. That would bring the feeling of admiration. It's most likely projection of their inability to deliver results.

Sounds like the guy felt threatened by your business and insecure about his.

Don’t get me wrong, this is very neat. But why?

For science. Many many small and unrelated breakthroughs can amount to something you wouldn't think of, just like what happened with "AI"

Reply on news.ycombinator.com