The new model of stability AI is a bit better at generating hands.

[ad_1]

Stability AI, which supports various generative AI experiments, has released a new version of Stable Diffusion, which was among the first to rival OpenAI’s DALL-E 2.

The new system, called Stable Diffusion XL or SDXL — Stability AI’s generative art tool, available in beta through DreamStudio — improves on the original in key ways. Tom Mason, Stability AI’s CTO, says that it brings “richness” to image generation that the old model (Stable Diffusion 2.1) lacked, with notable improvements in applications such as graphic design and architecture.

“We are thrilled to announce the latest iteration in our Stable Diffusion series of imaging solutions,” he said in a press release. “[It’s] A game-changer across multiple industries… the results are being seen right before our eyes.

Putting the hyperbole aside, the SDXL looks similar to – and perhaps better than – the latest MidJourney model responsible for the “Balenciaga Pope” (among other memorabilia).

While the previous version of Stable Diffusion and many other text-to-image systems struggled to reproduce certain body parts like hands, SDXL has no problem. Hands are always… well, not tangible. But they are often ahead of the fantasy fuel SDXL’s predecessor.

Stable distribution

The SDXL handles the hands better, but obviously isn’t perfect.

Stable distribution

Stable distribution 2.1 is clearly worse on the hand, on the hands. (I see myself)

SDXL is supposed to be better at generating text, a task that has historically been thrown for a loop by generative AI art models. But if my brief investigation is any indication, it still has a ways to go;

Stable distribution on the side of the road

Above, results from Stable Diffusion 2.1. Below, results from SDXL.

In the press release, Stability AI also said that SDXL features “improved image composition and face generation” and unlike its predecessors, it “doesn’t require long and detailed queries to create expressive images.” In addition to text-to-image querying, SDXL also has image-to-image querying (inserting an image to find a variation of that image), inpainting (reconstructing missing parts of an image), and inpainting (including seamless reconstruction). Extending an existing image).

As a wildcard, I tried to create as short a quick “Balenciaga Pope” meme as possible. The difference in results was stronger than I expected, I have to say, with the SDXL featuring runway models that could pass for designer wear versus the downright religious-looking clothes that the old stable distribution had.

Once out of beta, SDXL will be open source, says Stability AI, just like previous iterations of Stable Diffusion. In addition to DreamStudio, SDXL is currently available through Stable’s API as well as Early Access.

As generative AI technology advances, tools like SDXL have landed companies in hot water as they build and market. Stability AI is on the cross Legal matter This company infringed the rights of millions of artists by developing its tools using copyrighted images scraped from the web. Stock Image provider Getty Images reportedly used images from the site without permission to create the first stable distribution.

The open source release of Stable Diffusion has also been the subject of controversy due to its relatively light usage restrictions. Some communities around the web have tapped pornographic celebrities to create deep fakes and violent images. So far, at least one U.S. lawmaker has called for models like Stable Diffusion to be phased out because it “doesn’t fix the content enough.”

In response to the lawsuits, Stability AI recently pledged to honor artists’ requests to remove their art from the Stable Diffusion training dataset, but that doesn’t apply to SDXL — only the next-generation Stable Diffusion models, code-named “Stable Diffusion 3.0,” according to Spawning Artists, the company leading the opt-out effort. They have removed more than 78 million artworks from the training data set to date.

Legal challenges be damned, Stability AI is under pressure to monetize its pervasive AI efforts, which run from art and animation to biomed and creative audio. Stability AI CEO Imad Mostak has hinted at plans for an IPO, but Semaphore recently reported that Stability AI — which raised more than $100 million in venture capital last October at a valuation of more than $1 billion — is burning through cash and slow. Monetization”

[ad_2]

Source link

Leave a Reply

Your email address will not be published. Required fields are marked *