I, a PPT designer, shut up after AI made me unable to stop

In the 2024 Spring Festival Gala, accompanied by Ren Suxi's warm and lingering singing, people across the country watched the first AI video of the Spring Festival Gala. "Looking through the window at a fairy tale, under the glowing clouds, the evening breeze gently blows through her silver hair, he smiles and waits for her to walk home together slowly", in the music, a man and a woman dancing to the music go from youth to old age.

2024 China Central Television Spring Festival Gala

What few people know is that behind this touching program is a highly difficult commission with a deadline of less than a month and no room for error.

And AI has achieved this seemingly "impossible task" .

The rapid development of generative artificial intelligence technology has brought the cooperation between artificial intelligence and humans to an unprecedented depth and breadth. When we no longer just talk about AI in science fiction movies in an abstract way, AI has already quietly begun to change our work and life. In view of this, Science Popularization China launched a series of talks to talk with industry insiders about everything related to AIGC.

Are you curious about how the AI ​​duet dance in the Spring Festival Gala work "She Who Pillows on Light" was realized? How does AI empower designers? Awen, the creator behind this video and a PPT designer, shared his experience in a conversation with us.

He said: When he first used AI, he never imagined that the situation would come to this day - "I basically can't live without AI."

The following is a summary of the conversation with Awen.

Copyright images in the gallery. Reprinting and using them may lead to copyright disputes.

How AI became part of my job

Q: Could you please briefly introduce your work?

Awen: My main job is a PPT designer for press conferences. I have a design studio in Beijing that specializes in PPT. I am also an AI artist.

Q: When did you start paying attention to AIGC?

Awen: I started paying attention to AI in April 2022. Because I often surf Weibo, a tool called Disco Diffusion was popular on Weibo in April 2022, and many artists and good friends around me were using it.

Q: When you first started trying out AIGC, what did you envision as the maximum capabilities of AI painting?

Awen: The first impression was definitely very shocking. Who has ever seen such a tool that can generate images by typing a few words in 2022? So I was very excited, but the quality of AI-generated images was still very average at the time, not high-definition enough. After DALLE-2 came out, I had a completely different view of this tool, thinking "it might be used in our work", but I never thought it would become what it is now - my daily work is basically inseparable from AI. At least in the field of static images, AI is fully usable.

Q: What has AI helped you do?

Awen: AI currently plays the biggest role in a very important part of my work - finding design materials. Press conferences often have some ultra-wide screen designs, but there are very few screens over ten meters long in the gallery that require very high-definition large-size image materials. In the past, we spent a lot of manual time to synthesize large-size materials, but now we just need to tell AI what size material I want.

In fact, in my main workflow of making press conference PPTs, AI currently accounts for a small proportion, about 25%~30%. Recently, I started to try some AI transfer painting creations, where AI accounts for 80%~90% of the entire workflow.

Q: Will your creative ideas change with the addition of AI?

Awen: I seem to have become lazy. In the past, when a creative need came, I would think about it first, but now I may subconsciously type a few keywords to feed AI. It is equivalent to having an extra super assistant .

Q: Do the images obtained from AI need to be manually modified?

Awen: In the beginning, we still needed to import the AI ​​materials into PS to “fix them up”, but now we can basically get it done in one go, and we hardly make any changes.

Q: Have you ever tried AI tools for making PPT?

Awen: Actually, our professional PPT designers don’t really appreciate the effects generated by this kind of tool, nor do they use it. It is more like a work report template for office workers to deal with their leaders.

Q: When did you start doing AI rotoscoping?

Awen: At the beginning of the year, Teacher Hai Xin and I received a commission from the Spring Festival Gala program team. When Ren Suxi sang the song "She Who Pillows on the Light", the big screen was going to use a duet dance as a background video projected on the stage, wanting to show a couple dancing to the music, from youth to old age. Because the production cycle was very short, less than a month, if the traditional path was used, whether it was motion capture scanning or modeling of the two dancers, it would take a lot of time . At that time, the Spring Festival Gala program team thought of using AI to see if they could produce a "not bad" effect in a very short period of time. In the end, we did it.

Q: How is this achieved specifically?

Awen: We encountered many challenges during the implementation of the project.

For example, the problem of character stability. The program is designed to be a duet dance with three stages of costume change, including marriage, post-marriage, and old age. Using AI to achieve smooth costume change is an important requirement . The reason why duet dance is difficult is that AI will confuse the characteristics of the two characters, so problems such as gender swapping often occur. We tried many methods and finally used the ControlNet tile model to fix the characteristics of the characters and solved this problem.

Another example is the realization of porcelain material. With the support of SDXL and Civitai open source models and LoRa, we quickly decided to choose the dancing figure made of white porcelain. But we encountered many problems in the process. Just when we thought we had to train the porcelain LoRa of SD1.5, we found that using a "keyword" could solve the material problem. In addition to keywords, we also found a plug-in called IP-Adapter, which can use a reference image to guide AI to generate a specified material effect.

Another challenge was the stability test of costume changes. We first aligned the clips in PR, and used prompt travel (different keyframes describe different content) during generation to achieve a result that satisfied the program team.

Vision of AI

Q: What room for improvement is there in current AI-generated images?

Awen: I think AI-generated images have reached their limit.

Q: Has it reached the limit of your imagination?

Awen: Anyway, if you put two pictures in front of me at random, I may not be able to tell which one is AI (generated) and which one is created by a real person.

The more AI develops, the more confused I am. Even designers in our professional field are like this. For the general public, the quality of AI images is completely sufficient. AI painting has reached the next level. In fact, the most arrogant group about the development of AI is ours. At the beginning, most of us looked down on AI-generated images. We thought, "How can AI be comparable to what we design or draw ourselves?" But the more we try, the better the quality of AI generation is. When we try it, we simply can't stop. It really reduces your workload and makes you more efficient. Then I slowly shut up.

However, if we have to say it, we need to combine it with the capabilities of large text models such as ChatGPT so that the text-based graph model can better understand "human language".

Now I do more AI transfer painting, creating a style that is more like oil painting. For example, I can turn a street scene in Shanghai into a scene from a famous painting by Van Gogh.

Q: What jobs can AI replace, and what jobs cannot be replaced?

Awen: Repetitive labor will definitely be replaced. For example, if your previous job was to cut out pictures every day, and you did work that had nothing to do with creativity, then you will definitely be replaced. If it cannot be replaced, it must be some softer skills, such as creativity. I think this type of work is completely irreplaceable, and the more you work, the more you rely on your personal aesthetics. Your personal content aesthetics or design aesthetics will affect your final image work. At present, AI can only bring some random inspiration, but humans can output their own aesthetics very subjectively, which cannot be replaced.

I have observed an interesting phenomenon. Two years ago, some laymen challenged painters, game original artists, and designers, saying that they would soon be out of work. But two years later, you will find that most of the top ten in the OPENART community are game original artists and designers. AI painting finally climbed to the top of the pyramid, and the people standing at the top are still those professionals.

Q: What advice do you have for AIGC practitioners?

Awen: Don’t worry too much about being left behind. New technologies emerge every day in this world. Based on my observations over the past six months, it is an efficient way to wait for everyone to try them out and then test the best tool.

Q: Do you think AI can create new jobs?

Awen: There will definitely be some. But currently, most people who use AI are traditional designers who have changed jobs.

Q: Do you think the ceiling of AI in the future will be the same as that of humans? Or will AI have another development direction?

Awen: I think it will surpass humans, because AI's knowledge reserves far exceed those of every human being, and it may even be the sum of all human knowledge. The key is how AI uses knowledge. I think it may only be a matter of time before AI surpasses humans.

Q: Can you recommend some interesting ways to play AIGC that you have discovered?

Awen: The best AI translation plugin I have ever used is Immersive Translation, which can turn all foreign language webpage content into bilingual translations with one click. It is backed by a large language model, and the translation is very accurate. I also recommend all programming novices to try cursor, which allows you to write websites and applications without any coding knowledge! In addition, I would like to share with friends who want to deeply play with AIGC the most appropriate AI tool for beginners: comfyUI. After getting started, all open source technologies are your plugins.

Q: Is there anything else you would like to share?

Awen: I would like to say that China’s achievements in the field of AI are actually very impressive . The media around the world have exaggerated the model capabilities of large foreign companies and ignored those low-key but shining Chinese teams.

In fact, in the open source community, at least in the field of AI painting and AI video, 90% of the components are written by Chinese or Chinese teams : LCM, AnimateDiff, instantID, IPadapter, LivePortrait, etc., not to mention KeLing. In fact, foreign open source communities are very respectful of Chinese teams, but Chinese teams have always been very low-key and rarely go viral in China, so many people always think that China's AI technology is not good and cannot beat foreign countries. In fact, in my opinion, it is not the case at all!

Planning and production

Author丨Dongding Oolong Popular Science Creator

Interviewee: Simon Awen, Co-founder of AbleSlide, AI Artist

Review丨Yu Yang, Head of Tencent Xuanwu Lab

Planning丨Lin Lin

Editor: He Tong

Proofread by Xu Lailinlin