💸 A Six-Hour Salesquake

In a Sunday broadcast on Baidu’s e-commerce app Youxuan, AI versions of superstar seller Luo Yonghao and his co-host Xiao Mu moved 55 million yuan ($7.65 million) worth of cosmetics, electronics, and snack bundles—outperforming Luo’s own record from a human-only stream last month that ran for just over four hours.

The digital duo kept chat rolling for more than six hours, cracking jokes, answering product questions, and triggering flash-sale countdowns without a single pause for water or rest.

🤖 How the Avatars Work

Baidu built the likenesses with its latest generative-AI stack, training on five years of Luo’s video archives to nail his punchlines, hand gestures, and signature sales pitches. Speech synthesis picked up regional slang, while real-time lip-sync powered a 60-frame-per-second face rig. Wu Jialu, head of R&D at Luo’s company, Be Friends, called it “a DeepSeek moment” for livestream e-commerce—referencing the open-source Chinese model that rattled investors earlier this year.

🚀 Why Virtual Hosts Sell More

Digital presenters slash costs: no studio rent, glam squad, or backup crew. They also scale: multiple clones can stream in parallel, target dialect niches, or switch products instantly. During Sunday’s show, the avatars handled thousands of viewer questions simultaneously, embedding “Buy Now” links in replies faster than any human team. Conversion rates jumped 34% compared with Luo’s previous Youxuan outing.

⚠️ Glitches & Guardrails

The tech isn’t flawless. A mid-stream lip-sync freeze—quickly memed as “digital lockjaw” reminded viewers they were watching code. Regulators worry shoppers may not realize the host is synthetic, so draft rules will mandate on-screen labels and archived transcripts for every AI claim. Platforms like Douyin already limit virtual hosts that don’t “meaningfully engage” with chat, pushing developers to build genuine back-and-forth, not prerecorded loops.

@kanekallaway

These 2 AI Avatars sold $7M worth of products in 6 hours (future of commerce) #ai #artificialintelligence #tech #techtok #technology #avat... See more

🔮 The Next Wave

Be Friends plans 50 simultaneous AI rooms this quarter, each tuned to a region, age bracket, or spending tier. Language models will soon auto-translate, letting Luo’s avatar hawk gadgets in Spanish at noon and skin care in Arabic after dark.

Analysts expect virtual hosts to dominate low-margin bulk goods first, leaving humans to front high-touch luxury launches—at least until avatars master charisma upgrades. If a digital clone can eclipse China’s “lipstick king,” no niche looks immune.

🎬 The AI-Video Generation Battle Intensifies

🥇 Veo 3 

Google bundles its most powerful video model with the Gemini Ultra subscription. From a text prompt, Veo 3 can generate high-quality video clips. While base clips are often short, they can be extended to create cinematic footage longer than 60 seconds at up to 4K resolution.

• Pros: Impressive high-resolution output, realistic lighting, and a sophisticated understanding of 3D space that creates cinematic motion.

• Cons: Videos made in the Gemini app have a watermark. The watermark can be removed by Ultra subscribers using a separate tool called Flow. Complex prompts require careful wording to avoid inconsistent results.

🥈 Runway Gen 4 

Released in April 2025, Runway’s Gen-4 model is a favorite among a wide range of creators. You can drop in a reference photo or a simple sketch, and Gen 4 will animate it into a short video, making it excellent for character consistency.

• Pros: A user-friendly, point-and-click timeline that requires no coding knowledge and allows for easy experimentation.

• Cons: Video output is capped at 720p resolution and is generated without sound. Like many current AI models, it can sometimes struggle to render complex visual effects like heavy rain.

🥉 Kling 2.1

Developed by Kuaishou (a company in which Tencent is a major investor), Kling has impressed users with its ability to generate true-to-life human faces and motion. It can create 10-second clips and supports English prompts on its global platform.

• Pros: Excels at creating crisp facial detail and realistic movement. The interface allows for easy reuse and tweaking of previous prompts.

• Cons: The global version’s interface is in English, but the tool is still evolving. Videos created by free users include a watermark, which can be removed with a paid plan.

⭐ Honorable Mentions

• Veo 3 Fast: Included with the more affordable Gemini Pro plan, this version produces shorter, lower-resolution drafts (typically 8 seconds) very quickly, making it ideal for storyboarding.

• MidJourney Video: Known for the same artistic, painterly style as its image generator, MidJourney's video tool operates on its website. It uses your subscription's GPU minutes to create stylized clips at 480p resolution.

• Hailuo 02: This model from Shanghai MiniMax is recognized for producing high-quality character animations and dynamic motion, making it well-suited for shareable, snack-sized clips.

This Week’s Scoop 🍦

🤖 Meta rewrites bot rules after bizarre user chat logs surface

🏛️ Senate advances a bill pausing new state-level AI laws for two years

🚘 Tesla finally rolls out limited and supervised FSD robotaxi rides

📱 Google’s “AI Mode” is now available in India

🎧 OpenAI + Jony Ive face trademark fight over name “IYO”

🌐 DeepL taps Nvidia clusters to translate entire web in 18 days

📝 Weekly Challenge: Why & Where To Learn “Vibe-Code”

As vibe coding start-ups begin to take shape, now is a good time to learn exactly what vibe coding is and how to apply it to your personal and professional life!

Why bother?

• 🌈 Creative edge: Vibe-coding blends code with design; perfect for generative art and interactive web toys.
  

• 💼 Portfolio pop: Hiring managers love eye-catching demos over boilerplate CRUD apps.
  

• 🎮 Rapid prototyping: Build gamey UI elements or AR filters in hours, not weeks.
  

Top classes (all well-reviewed):

🎓 Replit’s “Vibe Coding 101” – free, project-based, integrates Ghostwriter co-pilot.

🖼️ “Vibe Coding with Cursor AI” on Coursera – Build real-world applications using Cursor.AI

🪄 “Vibe Coding from Scratch: Learn to "Code" Using AI” on Udemy – helps students build real applications and websites without writing any code.

🎵 “Vibe Coding: Build Personal Tools with AI and Python fast!” On Udemy – Learn to use AI to build tools, scrapers, automations and more!

Want to sponsor Jumble?

Click below ⬇️

That’s it for this week! From avatar shopping sprees to 4K AI cinema, synthetic media keeps rewriting the playbook. Which story blew you away? Reply and let us know. Until next time.

Stay informed, stay curious, and stay ahead with Jumble!

Zoe from Jumble