Uploaded May 2024 | Updated September 2026, 1 week ago
In this video I'll explain how to use generative A.I. to create optical illusion images (also known as ambigrams, or anamorphic images), which can be made to resemble different subjects depending on how they are viewed.
All software is freely-available and can be run via a Python script on your own desktop PC - no licence or server subscription required!
I'll be using Stability AI's Deepfloyd IF model (published at huggingface.co/DeepFloyd/IF-I-XL-v1.0) to diffuse multiple views of the same image, based on different prompts, and different specified transformations, in parallel.
For example, you could generate an image that looked like a pigeon but at the same time, when flipped upside-down, looked like a rabbit and, when the colours were negated, looked like a map of Europe (maybe!).
All code used in this video is available here: github.com/playfultechnology/visual_anagrams
Based on the method described in the research paper published here: arxiv.org/abs/2311.17919 and associated code at github.com/dangeng/visual_anagrams
In this video I'll explain how to use generative A.I. to create optical illusion images (also known as ambigrams, or anamorphic images), which can be made to resemble different subjects depending on how they are viewed.
All software is freely-available and can be run via a Python script on your own desktop PC - no licence or server subscription required!
I'll be using Stability AI's Deepfloyd IF model (published at huggingface.co/DeepFloyd/IF-I-XL-v1.0) to diffuse multiple views of the same image, based on different prompts, and different specified transformations, in parallel.
For example, you could generate an image that looked like a pigeon but at the same time, when flipped upside-down, looked like a rabbit and, when the colours were negated, looked like a map of Europe (maybe!).
All code used in this video is available here: github.com/playfultechnology/visual_anagrams
Based on the method described in the research paper published here: arxiv.org/abs/2311.17919 and associated code at github.com/dangeng/visual_anagrams
