Stability AI released without _drawing tool! Imitation of different perspectives in one second, editable details

Source: Qubit (public account QbitAI)

Author: Cressy

Among the four pictures below, one is second-hand by the famous Japanese artist Yume Takehisa. Can you tell which one it is?

The answer is in the upper left corner, but that doesn't matter.

The important thing is that the other three works with very or even nine similar styles are all generated by AI.

This is Stable Diffusion Reimagine XL, the latest image reconstruction tool released by Clipdrop under Stability AI.

(The name is too long, so we will simply call it RXL below)

With it, you don't need any professional knowledge, and you don't need to, just upload one picture and you can get three similar works.

Of course, the scope of application is not limited to painting, any picture can be operated in this way.

For example, if you want to change the layout of the room on a whim, you can also let it generate a few renderings for you.

△The upper left is the original picture, the same below

Netizens have also posted their gameplay in social media topics.

For example, generate more angled "photos" for Tesla vehicles.

Or change the character's expression, movement...

Scenery photos and illustrations are good projects

We also actually experienced it.

Landscape photography, spatial layout, and illustrations are all RXL's forte projects.

If the classification of landscape photos is further refined, it can roughly be divided into urban scenery, natural scenery and other types.

In terms of urban scenery, we used a photo of Victoria Harbor in Hong Kong as the original image and obtained three urban landscapes.

After the building is changed, of course it will be different from the real city, but it can be used to generate a virtual city.

The reconstruction of natural scenery photos seems to be what RXL is best at.

For example, the imitation of this photo of Tianchi Lake in Changbai Mountain has reached the level of being fake.

There is also Chaka Salt Lake.

RXL can also be applied in life, such as organizing your desktop.

Take a photo of your desktop, upload it to RXL, and you can get a new arrangement.

And if you are an illustrator, RXL can also improve your work efficiency.

In terms of quantity, RXL can directly help you generate more paintings of the same style.

In terms of quality, if you are not satisfied with your own work but don't know where to modify it, you can also try to ask RXL for help.

For example, the above work, while the consistency of the style is perfectly maintained, it also has a good visual effect.

For artwork, we also tested it with paintings of celebrities.

RXL and Van Gogh seem to have a tacit understanding, and both Xingyueye and Sunflower imitate so much alike.

However, similarity cannot be used as the only criterion for evaluating RXL performance.

For paintings, it is more important to imitate the style of famous artists and create works with other contents.

Therefore, due to its nature, it is not so suitable to use RXL to imitate the landscape painting of famous people.

Details are flawed, but editable

RXL still has some flaws in the depiction of some details.

For example, when imitating Monet's Sunrise Impression, two suns appeared in the sky, and the colors were different...

The fake Mona Lisa's facial skeleton also looks a little strange.

In terms of real people, the common problem of AI painting tools in hand processing seems to be unavoidable-the imitation of Cook's photo has only four fingers.

However, if you are not satisfied with the pictures generated by RXL, you can regenerate them with one click.

And if you are still not satisfied with the local effect, you can also clean it up in a targeted manner.

But the cleaning method seems a bit violent...

In addition, if you don't want the background, you can also erase it with one click.

In general, RXL can be helpful in the work of photographers, illustrators, and in the design of life scenes.

In addition to what we have thought of, there will be more interesting application scenarios waiting to be discovered.

In terms of working principle, unlike traditional AI photo editing tools, RXL directly draws new pictures from scratch.

RXL encodes the pictures uploaded by users to replace the text prompts in the Vincent diagram tool.

According to the official introduction, none of the pixels in the original image are used in the works generated by RXL.

Having said all that, can't wait to try it?

Free users can try it three times, and the generated pictures will be downloaded with watermarks.

The price of the paid version is $7 per month, and there is also an API available in addition to the web version.

However, you don't need to log in to use the free version, so...

In addition, according to the official introduction, the source code of RXL will also be released in GitHub later.

Portal:

Reference link:

[1]

[2]

[3]

• over-