HomeAI News
The fool's map generator blasts the field, and the stickman becomes Sherlock Holmes in seconds! Stability AI's latest drawing artifact makes graffiti become a blockbuster in seconds

The fool's map generator blasts the field, and the stickman becomes Sherlock Holmes in seconds! Stability AI's latest drawing artifact makes graffiti become a blockbuster in seconds

Hayo News
Hayo News
July 17th, 2023
View OriginalTranslated by Google
You can be promoted to an illustrator with zero threshold, and you can get a movie-level blockbuster with a simple sketch. Stability AI's new drawing artifact is going to put painters and designers out of work!

Kill crazy, kill crazy, the AI ​​painting tool that makes painters and designers unemployed in place, is new again!

Stability AI's new Stable Doodle is almost divine!

It can turn your graffiti into a masterpiece in a second, completely subverting the plain text prompt method in the traditional process of AI painting.

Many of my friends are "dead in language", and using precise language to describe the image they want is simply fatal.

Now, with a few scribbles, an image instantly becomes a colorful, fully rendered, blockbuster image.

As long as you give a rough outline, AI can complete the rest of the details for you!

This meme is no longer a meme, it can come true with Stable Doodle

So, a painting novice can become an art master in seconds?

A few drafts turned into a blockbuster

The demonstration given on the official website is simply cool.

Draw a few buildings at random, and draw a circle on the upper right.

Enter the prompt "Cityspace with full moon (street view under the full moon)", and then, it is the moment to witness the miracle——

Choose no style (no style), photography (photographic), analog film (analog film), fantasy art (fantasy art), comic style (manga style), origami (origami), pixel art (pixel art), neon punk (neon punk), isometric (isometric)

Draw a robot sketch, supplemented by a prompt prompt: "photo, photorealistic CG rendering of a robot head, predator, downtown city bokeh at night, dramatic lighting (photo, realistic CG rendering of the robot head, Predator, blurred city night scene , Dramatic light effect)", you get 3 robot fighters in the city night scene.

With just a few strokes, you get a portrait of a cyberpunk mech woman. (prompt: Mecha woman portrait)

Simply draw the outline of a castle, add the prompt "Castle on a hill, winter, 'Anime' Style (castle on a hill, winter, 'Anime' Style)", and immediately generate illustrations that can be used on book covers.

Draw a cat wearing clothes, the prompt is "Cat with a jeans jacket, 'Digital Art' Style (a cat wearing a denim jacket, digital art style)", and you get three Mr. Cats full of artistic styles.

For friends who are good at painting, it can help you generate more detailed sketches. For example, give it a "living room sketch", the prompt is "living room, comic-book style (living room, comic-book style)", and it will automatically draw a comic with perfect composition and coloring.

netizen live

Needless to say, netizens are already excited to play crazy.

Simple as a sketch of Medusa, Stable Doodle gives you a metallic Rose Queen of Thorns.

Draw an outline of a human body at will, without drawing facial features, without drawing a figure, and without knowing human anatomy, just a simple prompt, and you can draw a beautiful woman holding her hair.

prompt: photo, beautiful woman with her hands in her hair

Want a beautiful woman in a bikini with a perfect figure in the sun? It is enough to roughly outline the outline. As long as the prompt is in place, Stable Doodle will definitely surprise you.

prompt: photo, beautiful woman, bikini, facing the camera, standing between two big open windows, her arms up, sunlight

For a chic lady with a white hat and red lips, let the Stable Doodle let its imagination run wild.

prompt: beautiful face, fashion, white colors, red lips

When learning stick figures in elementary school art class, drawing a small house is the most basic. With Stable Doodle, you can directly advance from the elementary school version to the painting master version.

This netizen has painting skills at first glance, and wants a few muscular versions of Pikachu.

As long as you give Stable Doodle a sample, all kinds of Pikachu muscular men are at your fingertips.

Give it an abstract sketch, as long as it is equipped with a prompt, it can give you a satisfactory answer.

Draw a very freehand sketch and tell it a "gentleman walking on the street in the rain", and it will automatically fill in the appropriate street scene for you. In a second, you will walk through the streets of England with Big Ben in the 19th century.

With only a few sketches, Stable Doodle drew a mysterious temple in the jungle with a full moon in the sky and flowing water in front of the door.

prompt: secret temple in the jungle, river in front of the temple, full moon in the night sky

So, if there is no graffiti, what will be the effect if you use prompt directly?

It can be seen that with the guidance of graffiti sketches, the AI ​​drawing tool will get the meaning we want very easily. If you only enter the prompt, you may need to take a lot of detours and generate a lot of pictures before you can just "hit" the result you want.

In addition, on the mobile app, it is also very convenient to use.

You know, the current release is only the first version! In a few months, the Stable Doodle might be terrifyingly powerful.

However, Stable Doodle also "perfectly" inherited the various problems of Stable Diffusion in character generation.

For example, use the graffiti below to match "man standing woman sitting on a chair holding a baby", and the generated images can be said to be more sans than one.

The step-by-step tutorial has been released

With such a powerful tool, it's okay to not know how to use it.

No, the Stable Doodle has not been out for a few days, and there are even tutorials.

After entering the Stable Doodle, the user sees a normal canvas.

Then, generate a graph in three steps:

  1. Select the pen and use the mouse to draw a simple sketch. Just doodle freely, no need to emphasize details or aesthetics.
  2. Enter text description
  3. Choose an art style (of course, you can choose no style if you don't need a style), and click Generate.

For example:

Look, it's the kind of stick figures for kindergarten children, and it doesn't need any artistic skills and foundation at all.

Of course, Stable Doodle will provide users with 3 options, and you can choose the one that best meets your expectations to save and use.

There are a total of 14 styles at the bottom, basically covering all the mainstream art styles.

They are: photography style, imaginative art style, origami style, isometric style, digital style, comic style, anime style, movie style, analog movie style, neon punk style, pixel style, low poly style, 3D model style, lines Wind (pictures from left to right)

Take a look at the difference in the generated images of the ship above under different style conditions.

isometric wind

From left to right: neon punk style, pixel style, comic style

In addition to broadly generating images, you have special needs Stable Doodle can also help you achieve.

For example, generate a logo.

I feel that if I use any of these pictures, I can open a popular ramen restaurant.

In addition, even the interior decoration, Stable Doodle can help you. You only need to roughly draw the arrangement structure of the objects in the house in the sketch, and there is probably something, and Stable Doodle can help you with the rest.

The author only drew a bar counter and a few tables in the sketch, and Stable Doodle created a quiet and refreshing bar that can linger here every night.

You can even customize icons for companies——

The generated icons can be used for branding, marketing, presentations, or to create illustrations for user interfaces.

The equidistant wind, isometric, is selected here.

Generate a pixel-style game scene——

Note, don't forget to choose pixel art, pixel style, text prompt, don't forget to write clearly that this is a game scene.

You can even choose a comic style to generate a variety of comics.

The Spider-Man above is an example, we can also generate Thor-

(Look at the little hammer in the sketch, I didn't see it)

Finally, we can also describe some scenes in our fantasy——

This is where Fantasy Art comes in handy!

Technology behind: SDXL+T2I-Adapter

According to the official introduction, the strongest ability of Stable Doodle is sketch-to-image, and more perfect and accurate control.

That is to say, from sketches to exquisite works in one step, and even generate animations.

Currently, Stable Doodle can be tried for free on Clipdrop, and of course the latest SDXL0.9.

prompt: Cute owl, 'Origami' Style (cute owl, origami style)

Whether you are a novice or an industry leader, no matter whether you have used AI image generation software before, as long as you spend a few minutes to get started with Stable Doodle, anyone can generate high-quality image works in seconds.

Whether it is designing a logo for a client, making display materials, or various illustrations, Stable Doodle can handle it with one click, maximizing efficiency.

The reason why it is so powerful is because it uses Stability AI's most advanced SDXL image generation technology and T2I-Adapter.

The latter is a conditional control solution developed by Tencent ARC, with which users can have more precise control over AI-generated graphics.

During the generation process, T2I-Adapter will provide SDXL with an auxiliary guidance function, so that SDXL can understand the input sketch outline and generate an image according to the prompt.

Project address: https://github.com/TencentARC/T2I-Adapter

Specifically, T2I-Adapter is a simple and small network composed of about 7.7 million parameters, which can provide additional guidance for pre-trained text-to-image models while freezing the original large-scale text-to-image models, and This provides increased control over the generation process.

Among them, the T2I-Adapter aligns internal knowledge in the T2I model with external control signals. Users can train various adapters according to different conditions, and realize rich control and editing effects.

The following are some implementations of T2I-Adapter on Hugging Face.

T2I-Adapter can guide the results of Stable Diffusion. The figure below shows the key posture results predicted by MMPose.

Guided by keypose, the T2I adapter can generate animals with the same pose, such as pandas and tigers.

Sketch T2I Adapter can be plug and play, without additional training, can produce any result.

The same is true for Keypose T2I-Adapter.

All in all, in Stable Doodle, the same sketch can generate a variety of exquisite pictures with different prompts and different styles.

Why don't you try it out quickly?






Reprinted from 新智元View Original