Meta teaches AI image model to stop generating human fingers like a drunk Picasso

Meta has announced a computer vision model that learns to recognize images while also building contextual knowledge that makes artificial intelligence less clunky and costly. "This model, the Image Joint Embedding Predictive Architecture (I-JEPA), learns by creating an internal model of the outside world, which compares …

[Arthur the cat]

Oh dear

The result is that I-JEPA is less prone to errors – like creating hands with extra fingers – when generating images.

Will nobody think of the Norfolk market?

[_Elvi_]

Boffins ..

Glad to see our old friends back in the Reg ..

Twas lonely without them.

[TheMaskedMan]

Re: Boffins ..

"Glad to see our old friends back in the Reg"

Was going to say much the same.

As for the extra-digital escapades of AI image generation, I must admit that the more recent versions of midjourney have been much better in that respect. Only rarely do I see dodgy hands, but the pesky thing did generate a dog with five feet earlier. Not legs, feet.

Impressive though it's output often is, I don't think any but the most surrealistic artists need worry too much just yet.

[FILE_ID.DIZ]

Generative architectures, Meta says, often have trouble with human hands because they try to fill in every bit of information without a conceptual basis for the scene.

Like most ChatGPT answers.

[M.V. Lipvig]

I have a finger

for Faecesbook.

https://d30y9cdsu7xlg0.cloudfront.net/png/82054-200.png