Meitu MiracleVision 3.0: Unleashing the Power of AI for Visual Creation

　　These days, every company has to ride on AIGC’s coattails.
　　Since the advent of ChatGPT, various large AI visual model companies have sprung up like mushrooms after a rain. Data shows that more than 30 large AI vision models have been released in China, and more than 100 entrepreneurial teams are on the way. But if you look closely, you will find that most large models only describe a perfect concept. Their products are rushed to market without much polishing, and the real user experience is very poor. Especially in the enterprise production process, the combination of many large models and workflows is still in the preliminary exploration stage. For a long time, the big AI vision model has been in vain. From enterprises to users, everyone is waiting for a real storm of baptism.
　　Recently, the author discovered that Meitu held a 15th anniversary birthday party. At the birthday party, founder and CEO Wu Xinhong officially announced a major event. Meitu’s self-developed AI visual model MiracleVision has released version 3.0. The latest model will be used in all images and product designs of Meitu to help e-commerce and advertising. , games, animation, film and television five major industries. The core capabilities of MiracleVision version 3.0 can be divided into two parts: “whimsical ideas” and “intelligent creation”. In addition to the basic operations of inputting text to generate corresponding images and inputting graffiti into masterpieces in seconds, the “intelligent brain supplement” of large models and Precision control capabilities have been further enhanced, lowering the threshold for drawing and editing, and achieving equal rights in creation.
　　For Meitu’s various new actions, the secondary capital market has also made its own answer. Meitu’s stock price has continued to trend upward in recent months. It is not difficult to find that in this long-distance race to explore AIGC, Meitu has pulled users to the front.
100 days of evolution, AI skills have reached the next level

　　The last time Meitu was updated was 100 days ago. In June this year, Meitu released MiracleVision, a large-scale visual model for the first time, which has brought surprises to the entire industry. It can reverse technological evolution from visual creation scenarios such as painting, design, photography, etc. It is especially good at Asian portrait photography, national style and national trends, etc. Generate direction. But at that time, Wu Xinhong said bluntly: “AI brings many opportunities and challenges, but this opportunity is not necessarily our opportunity, and the challenges are real. I am still walking on thin ice.” This emphasis on technology
　　and The sense of crisis allowed Meitu’s large model to be iteratively upgraded again in a short period of time. Wu Xinhong said that three months ago, MiracleVision’s painting skills were still at the initial stage, but now version 3.0 can depict real and delicate picture details.
　　The author personally experienced the AI ​​visual creation tool WHEE with MiracleVision 3.0 as the core. After entering “a little girl” in the dialog box, the platform will provide intelligent association and complete the keywords as “children’s drawings, little girl, innocent smile, green grass, high resolution, cartoon style, cute, sunny, “Watercolor Effect”, after setting the basic parameters and waiting for a few seconds, 4 images of little girls with different contents were generated. Specifically, these portraits pay attention to details, the pictures are exquisite and conform to the aesthetic structure, the characters are lifelike, and they even have reasonable and changeable light and shadow effects, creating a full sense of atmosphere.
　　Ordinary users may not understand the professional terms related to paintings and photos, such as light and shadow effects, texture, style, picture quality, etc. Meitu’s “whimsical ideas” can provide creative inspiration for users without art background. If you have more professional needs, MiracleVision’s “Precise Word Prompt Control” function can also easily achieve close-up, long-distance, forward light, backlight and other generating effects.
　　”Intelligent creation” is another major event that Meitu has accomplished. MiracleVision’s “Deepening Creation” function can further enrich the details of the work and improve its expressiveness; the “AI Screen Expansion” function makes the work larger in size and richer in details; the “Partial Modification” function allows you to modify parts of the picture where you want. Precise modification and adjustment; the “resolution enhancement” function can generate high-definition large images, making detail expression, color display, and object recognition more accurate and vivid.
　　It can be seen that this update of the Meitu large model attempts to solve the problem of image accuracy and controllability from the root cause, taking AI visual generation to a higher level.
Performance improvement tools more suitable for work scenarios

　　To a certain extent, using large AI visual models into the workflow is more difficult than simply doing a good job in visual presentation. Although the underlying technologies of life scenarios and productivity scenarios are common, the actual application scenarios are different and require a lot of refined adaptation work.
　　The greater significance of the launch of Meitu Miracle Vision 3.0 this time is that it can be used in the five major industries of e-commerce, advertising, games, animation, and film and television to help improve workflow efficiency. Taking the e-commerce industry as an example, making materials has always been a headache. Many shoe and apparel companies spend hundreds of thousands or even millions of dollars every year just to hire a photography team and model shooting. Moreover, real-life model shooting is a one-time deal, and it will be very troublesome to make adjustments later.
　　Nowadays, in the Meitu Design Studio, merchants can realize the entire process from graffiti generation to line drafts, line draft coloring, product pictures, model try-on pictures, to e-commerce material output, greatly improving production efficiency and reducing costs. Using the “AI Shoes and Clothing” tool, you can first complete a preliminary product image. After selecting the line drawing reference image, material, and text description, 8 images can be generated at one time, and the generated results can also be fine-tuned twice. Secondly, on the “AI product map”, you can automatically cut out the newly prepared product map, select the appropriate scene, and optimize the details. Finally, see the display effect in the “AI Poster”.
　　Meitu has also developed an “AI model fitting” tool. After uploading a mannequin or a real model picture, you can selectively select which part of the product content to keep or remove, and then select the required AI model, gender, and race from the model library. , skin color can be customized. After a moment, the AI ​​model fitting picture will be generated. If you are not satisfied with the effect, you can change the model and re-edit it.
　　It is not difficult to find that the emergence of Meitu visual large models has formed an AI native workflow. After brainstorming, AI can quickly generate various creative visual solutions. After determining the final idea, AI can quickly refine the idea and complete the final work. Therefore, working hours that would have lasted more than a week were compressed to 3 days or even less.
　　In Wu Xinhong’s view, the essence of the large AI visual model is an endless visual creative library, and the application layer is a content extractor. “According to the user’s needs, the required content is extracted from this huge creative library, allowing users to used in specific scenarios.” It is Meitu’s goal to bring the accuracy and controllability achieved by version 3.0 of the visual model to vertical work scenarios and reduce costs and increase efficiency for the industry.
A future highly integrated with AIGC

　　Today, Meitu and AIGC have been deeply bound. Two months earlier, Meitu announced its 2023 interim results. In the first half of the year, it achieved total revenue of 1.261 billion yuan, a year-on-year increase of 29.8%; net profit attributable to the parent company was 151 million yuan, a year-on-year increase of 320.4%. Meitu’s monthly active users reached 247 million, a year-on-year increase of 2.5%. In terms of specific businesses, revenue from imaging and design products reached 602 million yuan, a year-on-year increase of 62.2%; revenue from beauty industry solutions reached 286 million yuan, a year-on-year increase of 31.1%.
　　It is not difficult to find that, driven by AIGC technology, Meitu continues to innovate in the fields of AI pictures, AI videos, AI design, etc., driving a significant increase in revenue from imaging and design products (mainly through VIP subscription fees), thus helping Meitu achieve Revenue growth and overall profitability.
　　In fact, there is a broader market space for AIGC application scenarios in the imaging field. At the birthday party, Wu Xinhong shared his opinion that the popularization of large-scale visual model applications will go through three stages: the exploration period before 2024, the rapid development period from 2024 to 2025, and the maturity period from 2026 to 2030. The application of visual large models on the production side is gradually becoming mature. In the future, it will help tens of millions of design scenarios and lead the upgrade of aesthetics and social and economic growth.