Zephyr Cressy from the temple of Wafi
量子位 | 公众号 QbitAI
2024's most woow AI mapping tool has appeared!
Start with the "Youth Memorial Handbook", and you will say that this taste is not authentic:
If you don't believe it, don't say it.
Don't worry, the next second "snap":
...... What's the situation?
This is the AI retouching artifact that many people in the photography industry have been playing recently-Super Canvas.
It is said that you can fix a creative blockbuster in minutes with just one prompt word, and you don't need to upload multiple pictures in advance to train an exclusive model.
Recently, it has also rapidly iteratively upgraded a wave of products, and the circle has become more lively. Some netizens called themselves experts after trying it, after all, it is simply not too cost-effective to use it to deal with the strange needs of customers.
For customers such as serving Aunt Silk Scarf, the raw picture effect of the super canvas is more than enough:
Such a magical AI tool qubit is naturally not to be missed (doge).
In-depth experience, it turns out that this AI can not only repair the whole picture but also part of it, and it is also possible to change the hair color with one click:
Even if you wear a small floral casually:
can also become a desert princess and a Tibetan girl in minutes, the key is that people look the same.
In the future, wouldn't it be possible to realize that people can wear pajamas at home, and the nine-square grid of the circle of friends can be made up with one click.
How exactly is this done?
One-stop service of AI "million retouchers".
After inquiring, as mentioned above, the R&D team of the super canvas comes from Baidu Netdisk, focusing on professional photography and retouching one-stop service.
Traditional professional portrait photography needs to change makeup, change clothes, and set up scenery, and it is a matter of one sentence to produce a blockbuster on a super canvas.
In summary, there are three workflows:
- AI Photo: The focus is on preserving the posture and face, and changing everything from clothing to background.
- Partial Redraw-Redraw: The focus is on the local processing of the original image and the addition/deletion of elements.
- Partial Repaint-Change Background: The focus is on existing costumes and the scene needs to be built.
Let's take a look at what superpowers this AI tool has.
Ahead of time AI photo part.
Bring a photo of my best friend Zhongguancun Jun Zhixian and upload it. The next scene immediately caught our eyes, and the movements and outlines of the characters in the picture were drawn with one click.
Looking down, it turns out that this is the conditional control function of the super canvas: motion capture, contour control.
In addition to these two, there are also depth of field, style reference, and enhancement details, and these five conditional controls can upload control charts and set the intensity to control the generation, which is the original image control by default. Up to three conditions can be combined for each map.
Next, there are two types of basic models available on the right side of the interface: real photography model and film photography model. The former is more realistic and detail-oriented, while the latter includes a depth-of-field effect and focuses on atmosphere.
The prompt words support Chinese and English, and the weights can be adjusted, and there are inspiration associations for users to choose directly. There are a range of controllable parameters and inverted prompts in the advanced settings, and the generation settings modify the canvas size and the number of images generated.
One thing to say, compared with most similar AI repair/life drawing tools in China, there are a lot of details that the Super Canvas can be set by the user himself.
This set of settings can be DIY by yourself and saved as "My Recipe" if you have professional needs.
If there is no inspiration, the official also provides a ready-made "inspiration formula", and the matching between the formula and the actual generation scene is also very high:
We chose the desert wind camel bell girl inspiration formula, and then directly clicked to generate, and the original Zhongguancun fenxian photo was transformed with one click.
That's it, my mother told the truth when she saw it. Consistency in face and body posture:
Let's take a look at the partial redraw plates. Click "Edit Selection", you can see that there are two ways to use local element change and one-click background change.
Among them, the change of local elements means where it is difficult to change, and the elements can be added and deleted.
The one-click background change function is suitable for those who already have mature clothing and only need to build a scene.
For more details of how to play, the official provides a Zhihu tutorial, which can be viewed by clicking on the "Help Document".
It not only explains in detail the effect of each controllable component, but also puts forward some tips:
- Try to choose pictures with light colors and simple backgrounds, and avoid pictures with too small characters.
- It is recommended to use the scale of the original image, and when the scale of the generated image does not match the original image, the position of the element will deviate.
- Positive prompt word reference formula: character subject + background + quality style + lighting + color + composition scene.
A large model that understands aesthetics
In terms of technology, behind the super canvas is the aesthetic AI image generation model built by Baidu's network disk R&D team based on Baidu's intelligent cloud Qianfan large model platform.
With the blessing of the large model, the characteristics of the super canvas are quite prominent.
First of all, the main function of this model is the generation of high-quality portrait photos. By using the reference image as the control condition, you can control the character posture, image structure, depth of field, and even the light and shadow of the picture and facial makeup of the generated results.
Maintaining the similarity between the finished film and the original image, the super canvas also has a killer feature: intelligent face fusion algorithm.
The algorithm can participate in the image generation process with the user's facial information as a control condition.
It analyzes the facial lighting, skin tone, texture and other features of the canvas generated results, combined with the facial features and face shape in the user's input image, to achieve "seamless" retention of the user's facial image in the generated image.
△ Comparison between the fusion diagram and the original material
In addition to "true", when creating a super canvas, the Baidu Netdisk team also gave it an understanding of "beauty" - an aesthetic evaluation model.
In order to align the model with human aesthetic cognition, the team built a model optimization link based on user feedback, so that the generated images are more in line with the user's aesthetic preferences.
It can be said that through the above three models, Super Canvas provides a one-stop portrait photography retouching solution for the photographer community.
Of course, behind the "superpowers" of these models, the strong underlying model and computing power support provided by our familiar old friend, Baidu Intelligent Cloud Qianfan Large Model Platform, also contributed a lot.
Qianfan large model platform is a one-stop large model development and service operation platform for enterprise developers, providing a variety of large model services, including a variety of AI development tools and a complete set of development environments, allowing developers to complete application development in one stop.
"Photography ChatGPT" is coming?
Speaking of the super canvas, the new experience it brings has made netizens who have tried it sigh: The ChatGPT moment of photography is coming.
This also leads us to think, what is the "ChatGPT moment of photography"?
A year ago, the domestic large-scale model arena launched a language model. Nowadays, image generation seems to have become a battleground for domestic large-scale model manufacturers.
The emergence of the super canvas can be said to have entered this battle from a new perspective - AI retouching/AI retouching is very common, but "GenAI retouching" seems to be a blue ocean in China.
To be more specific, the super canvas is aimed at the pain points of domestic photographers.
Non-AI industries such as photographers may not be able to use tools such as SD and MJ, while the Super Canvas is a domestic application, and everyone understands the gold content of "domestic", making it more accessible to users.
When photographers encounter customers who don't follow the routine, Super Canvas can help them save most of the P-picture work with one click, and easily meet the needs of users such as cos photos, guest films, and commercial videos.
It is understood that with the optimization and upgrading of products, the super canvas can also cope with the needs of secondary sales of guest films and repurchase of old customers.
For example, taking a photo and giving away an AI photo is really a good way to attract customers?
What's more convenient is that the super canvas does not need to be trained, deployed, or downloaded and installed, eliminating the technical threshold of building your own workflow, and also breaking down the hardware barriers of self-deployment of models.
Just open the web page, you can click and use, the interface is simple and clear, whether it is an AI novice or a celebrity, any photographer has the opportunity to get started, and is no longer helpless in the face of all kinds of new scenes.
If you are an AI user, you will also have the opportunity to use your high-level prompt words skills to mix and match the control conditions in the application at will, and explore more diverse celebrity-level gameplay.
If you have been poked, you might as well try it yourself.
— END —
QbitAI · Headline number signed
Follow us and be the first to know about cutting-edge technology trends