AI Video Explosion! 100,000 Clips a Day Flooding Douyin, Kuaishou, and Xiaohongshu

[baoshi.rao]

Under warm lighting, a vintage suitcase slowly opens to reveal gray-white sneakers, with light and shadow gliding across the shoe surface. The camera zooms in, showcasing the clear texture of suede. The scene shifts as the shoes rotate, transitioning from dim to bright lighting, creating a slow-motion contrast of colors at the heel—one side bright, the other elegant.

This 20-second product showcase video is rich in angles, meticulous in color, and varied in shots, yet it wasn't filmed by a camera but generated by AI from just a few photos.

The importance of short videos in e-commerce marketing is undeniable, and AI is set to drastically enhance production efficiency. "AIGC allows us to produce 100,000 short videos a day," said Mao Xuchao, co-founder of Shidai Yongxian, at the Yibang Summit.

This is the new productivity brought by large models.

From a technological development perspective, AI video generation has gone through three stages: image stitching, GAN/VAE generation, and autoregressive/diffusion models. It is now applied in film trailers, advertisements, virtual scenes/characters/effects, and the restoration of old films/rare footage.

With the significant improvement in short video industrialization, advertising agencies, MCNs, film studios, and gaming companies are all undergoing transformations.

01

Automated Short Video Generation at 1/10 the Cost

"Domestic demand for short videos is strong, primarily driven by traffic support from e-commerce platforms," analyzed Wu Bin, CEO of Jirui Technology.

Official support means massive traffic and high ROI. Brands only need to produce and publish videos to achieve traffic growth. "Because platforms offer traffic support, every SKU of a brand now needs to be videoized," Wu Bin pointed out.

However, in 2022, short video production capacity couldn't keep up with the explosive demand across platforms.

Compared to manual production, large models bring industrialized video production. For example, Shidai Yongxian's Super Wheat Video can transform all product images into videos. "We can generate countless videos from the same product page. Since each AI-generated video varies in angles and content, we can produce unlimited basic videos for brands to gain traffic in public domains," Mao Xuchao explained.

Large models reduce short video production costs to 1/5–1/10 of the original. "Previously, producing 10,000 videos a year cost over 1 million yuan; now, it might only be 200,000 yuan. The demand is limitless—every merchant in this industry needs it, but it was unattainable before," Wu Bin noted.

This new productivity is also changing brands' content marketing strategies and the survival models of ad agencies and MCNs.

Mao Xuchao observed that brands' content marketing used to follow an inverted pyramid:

• For 1%–5% of hit products: High-budget, high-quality content produced by ad agencies or 4A creative firms.
• For 10%–15% of core products: Low-budget, high-quality content from studios or production companies.
• For 70%–80% of long-tail products: Low-budget, low-quality content from e-commerce operators or internal teams.

With AIGC, brands can now create tailored AI content for different product tiers:

• For hit products: AI-generated scripts + high-precision 3D models to produce high-quality 3D videos.
• For core products: AI-powered editing, virtual influencer reviews, or 3D product videos.
• For long-tail products: AI-driven 2D product videos, Taobao detail page videos, AI voiceovers, or virtual model try-ons.

Domestically, live-streaming clip automation is also trending. For instance, Jirui's iCut automatically identifies key moments during live streams, generating short videos in real time to attract and retain customers.

In 2023, Jirui expects quadruple growth. Its iCut demo, launched in April, was well-received despite initial manual adjustments. "The traffic boost was significant, and clients started bulk purchases by July–August," Wu Bin recalled.

Shidai Yongxian also anticipates tripling or quadrupling revenue, expanding from online marketing to offline stores by replacing posters with screens displaying short videos.

02

The Rise of Text-to-Video Models

Recently, someone merged Oppenheimer and Barbie into a trailer using ChatGPT for scripts, Midjourney for images, and Runway Gen-2 for video. The result—a blend of pink glamour and industrial grit—was innovative and visually striking.

Since Runway's GEN-2 launch in April 2023, enabling text, image, or clip-to-video generation, creative possibilities have exploded.

A paragraph can become a short video.

A single image can become a short video.

However, text-to-video technology is still evolving. Challenges include high-resolution generation, long-text coherence, and infinite-duration consistency.

AI video generation is already used in film. Runway contributed to Everything Everywhere All at Once, where a five-person team completed post-production, earning praise for scenes like "hotdog hands."

The Wandering Earth director Guo Fan noted, "If we film The Wandering Earth 3, at least half the on-set crew could be replaced. AI is both a challenge and an opportunity—a chance to overtake Hollywood."

03

Marketing Tech Revolution: Who Benefits First?

Film demands higher duration, coherence, and realism, while marketing prioritizes cost and traffic.

Shidai Yongxian founder William Li (alias "Kongjie"), former head of Tmall Luxury Pavilion, focuses solely on marketing material production. Their FancyGPT model—trained on 60B LLaMA parameters—automates video creation and manages multi-platform content. "We solve asset challenges for entire brands, not just one platform," said CRO Moyo.

Beyond e-commerce, they target 4A agencies and offline marketing. "E-commerce is a 5B yuan market; ads and offline are 15B each. That’s 35B yuan to explore," Moyo added.

Current video models face hurdles like high computational costs—each second contains ~30 frames, requiring massive resources for consistency. Complex temporal data also complicates modeling.

Jack Welch's "10x Rule" states that when a new technology outperforms by 10x or cuts costs to 1/10, it disrupts old systems. AI video generation has achieved both in marketing, reshaping agencies and MCNs—and the transformation continues.