Shulou(Shulou.com)11/24 Report--

In the past, to do AI, you need to study algorithms and write a lot of code, and the learning threshold is very high.

Now that algorithms and code have been integrated into mainstream learning frameworks, AI engineers can directly adjust parameters and training models, or download and use models that others have trained. If the ready-made algorithm and framework are used as a Dan furnace, then the process of adjusting parameters and training models is called alchemy.

On October 15, Pixso and Stable Diffusion Alchemy Pavilion jointly held a special Pixso alchemy conference, inviting domestic well-known AIGC creators, SD (Stable Diffusion) model training experts, AI video creators and other celebrities to share the practical experience of SD model training and application and the cutting-edge trends of AI in various fields.

01. AI Design and Future Imagination

How will AI design be applied to various industries? How should individuals deal with themselves under the tide of AI technological change?

Wang Baotian, founder of AICC and a senior product design expert, brought examples of artificial intelligence application in the design field, including specific application cases and practical experience of AI design in jewelry design, e-commerce booth design, architectural design and game industry design, analyzed the opportunities and risks of AI design, and thought deeply about the development of AI technology in the future.

The speech also covered the development of multimodal content and the relationship between data, information, knowledge, insight, wisdom and influence. At a time when AI technology is about to be widely used, we begin to pay attention to the importance of key AI technologies, and encourage people who have not yet made in-depth application of AI technology to begin to reshape their personal ability structure, make rational use of their own resources, and deeply apply AI technology to practical work scenarios.

02. Commercial landing case of SDXL model

What is the advantage of SDXL model in commercial landing scene compared with 1.5 model? What are the commercial landing cases at present? Why will SDXL once again detonate the commercial landing trend of SD models?

Zhongzhong, an expert in design model training, shared the training process and experience of BDiconSDXL models. He suggested that beginners start training with single-style materials in the processing and screening of BDiconSDXL training sets, with a resolution of 1024 to 1024, clean and tidy background recommendations, more than 20 training sheets, and consider simple marking experience. The whole process is full of practical information.

In addition, he also focuses on comparing the advantages of SDXL model and SD1.5 model, and introduces and shows the application results of SDXL models, such as BDicon 3D icon, Microsoft Design Microsoft style, IDillustration Internet illustration style, ZQillustration Mid-Autumn style, LSillustration abstract line-to-surface interpolation style, 3Drendering e-commerce booth 3D rendering effect, jewelry jewelry model and so on.

Finally, Zhongzhong believes that the starry sea of the SDXL model has just begun and will land in a blowout in the future, encouraging everyone to participate in the wave of SDXL landing.

03. Global TOP5, how did wheat orange realism come into being?

Stable Diffusion modeler Merjic Mai Jing, perennially ranked among the top five authors of the world's most famous open source SD model sharing website Creators. The "wheat orange realism" model has more than 2 million of the total model operation in foreign paid online biogram platforms, and nearly 7 million of the total model operation in major domestic online bioglyph platforms.

This time, the wheat orange not only shares the famous representative works, but also shows the birth process of the wheat orange realistic model and the principle and practice of model fusion, telling the alchemist that there is no absolute truth in the process of model fusion, only experience. For example, in the lora model, the IN input area is responsible for plane composition, and the OUT output area is responsible for color composition. The higher the number of the IN input area, the greater the influence on the plane composition. In the OUT output area, the higher the number, the more localized the influence on the refinement process, and the greater the influence on the coloring process. The best way to integrate the model is to use the XYZ axis to try to influence the results according to these hierarchical characteristics. in addition, the fusion of the lora model should be cautious.

04. Pixso helps the integration of AIGC design and exploration.

From commercial to civil use, from AI to AGI and AIGC, with the continuous progress of science and technology and society, the design content needed by the society is constantly changing, and the quality needs of designers are also changing. How will the product design integrated collaboration platform Pixso, which integrates whiteboard, prototype, design, delivery, enterprise management and other capabilities, efficiently empower designers in the AI era?

Cheng Junnan, head of Pixso products, said that Pixso AI's existing textual diagrams, graphic diagrams, design specification generation, design element checklist and multilingual translation capabilities have been widely used in all aspects of the final landing of designers, and will speed up the construction of AI-related functions.

In the future, Pixso will also empower the full link of the product design in the way of GPT+SD, fully embracing AI. The advice for individuals in the AI era is to "hug, hug, and then hug, just like the electric era when embracing electricity, like the Internet era embracing the network, fully embracing AI in life and work."

05. The present situation and application of AI video

2023 is the first year of AI video, generating video has begun to have more and more applications. To what extent does AI generate video at this stage, and what technologies can be used? How long will it take for everyone to be a director?

The director of BORDER Studio, visiting lecturer of AI course of Tsinghua Academy of Fine Arts and digital artist shared the development of AIGC video field, the play of mainstream tools and the most cutting-edge technology development direction in the past two years.

At present, the AIGC video game is AI to generate pictures to make video, AI to generate multiple pictures to produce video, Wensheng video, "picture + text" to generate video and "video + text" to generate video. With the help of AI, you can easily replace the role and style in the video to achieve the effect of allowing text, pictures to move and video shot control.

After an in-depth analysis of the principle of AIGC video production, teasing sand thinks that the development trajectory of this year's AI video is similar to that of last year's AI painting, and it has now reached a stage of exponential growth. The development of AI painting and other AI multimodal technologies have improved the technical puzzles for AI video.

In the coming year, AI video generation is likely to surpass AI painting to become a new hot spot, resulting in the emergence of more AI video tools.

In the Pixso special session of the Alchemy Conference, in addition to the speakers sharing their experiences in their respective fields, AI enthusiasts from Ali, Tencent and many other large factories were also in the question and answer session, contributed valuable questions, and actively discussed and exchanged at the scene.

Pixso will also provide more opportunities for AIGC enthusiasts to exchange and learn, and look forward to more professional fields of Dajia working with Pixso to empower domestic AIGC enthusiasts.

Welcome to subscribe "Shulou Technology Information " to get latest news, interesting things and hot topics in the IT industry, and controls the hottest and latest Internet news, technology news and IT industry trends.