Remember those laughable TV episodes in which someone asks, “Can you enhance that?”

Well, laugh no more. Google AI has mastered “high fidelity image generation.”

You can just about hear it: “HAL, unlock the enhancing algorithm.”

Google explains their new method:

“Diffusion models work by corrupting the training data by progressively adding Gaussian noise, slowly wiping out details in the data until it becomes pure noise, and then training a neural network to reverse this corruption process. Running this reversed corruption process synthesizes data from pure noise by gradually denoising it until a clean sample is produced.”

Add noise to the picture, and then denoise it?

Here is the Super-Resolution via Repeated Refinement paper.

And the Cascaded Diffusion Models for High Fidelity Image Generation paper.

My take: It was Arthur C. Clarke who said, “Any sufficiently advanced technology is indistinguishable from magic.” Google has just given us more magic. And we so smugly said those enhancing programs can’t add resolution back into a pixilated picture. Looks like we were wrong, yet again.