Google Veo Fuels Synthetic Flood as Creators Shift to Direct Sales
Ai4
Surging generative video is inflating social metrics while eroding advertising payouts, forcing creators to abandon synthetic shortcuts in favor of verified human trust and structured authority to drive real conversions.
## Background
Published on YouTube by the Ai4 channel on August 27, 2025, this presentation features AI YouTuber and podcaster Dylan Jorgensson (Dylan Curious). According to the video description, Jorgensson shares how artificial intelligence has transformed his content creation strategies, offering practical workflows, emerging tools, and upcoming social media trends for personal brands and businesses.
## TL;DR
Dylan Jorgensson explores the evolution of generative media from early text-to-image models to photorealistic video engines like Google Veo and interactive world models like Google Genie. He demonstrates practical workarounds for character consistency—such as leveraging non-human mascots, image-to-image keyframing, and frame-splicing viral loops—while showing how platforms like Showrunner point toward community-co-created entertainment. Shifting from production to discovery, he outlines how Generative Engine Optimization (GEO) relies on trusted entities like Wikipedia to secure visibility across LLM search results. Ultimately, he addresses the rise of synthetic bot traffic and low-grade 'AI slop,' comparing the current gold rush to the dot-com bubble and emphasizing that enduring commercial success belongs to high-trust, authentic human brands.
## Atmosphere & style
The presentation carries a pragmatic, candid, and grounded tone that balances technical enthusiasm with realistic skepticism. Delivering an experienced insider's reality check, Dylan pairs tactical demonstrations with honest admissions of failed experiments, lending the talk the feel of an intimate, clear-eyed strategic debrief. The emotional arc transitions from excitement over hyper-realistic visual effects into an unsettling look at bot-filled comment sections and creator loneliness, concluding with steady resolve about building resilient brands amidst the noise.
## Key takeaways
- [00:04](https://www.youtube.com/watch?v=uRL89WmkDsc&t=4s) The evolution of generative AI has rapidly advanced from rudimentary text-to-image models like early Midjourney into high-fidelity video tools like Google Veo.
- [02:20](https://www.youtube.com/watch?v=uRL89WmkDsc&t=140s) Google Veo (Veo 3) is approaching the threshold of crossing the uncanny valley, allowing creators to produce photorealistic footage and simulate dangerous, expensive, or impossible scenarios.
- [05:18](https://www.youtube.com/watch?v=uRL89WmkDsc&t=318s) To bypass current generative limitations with human facial consistency, successful creators rely on non-human mascots, animals, and helmeted characters like Stormtroopers.
- [05:18](https://www.youtube.com/watch?v=uRL89WmkDsc&t=318s) Image-to-image interpolation and first-to-last frame keyframing provide narrative continuity across scenes, enabling episodic storytelling in synthetic media.
- [10:20](https://www.youtube.com/watch?v=uRL89WmkDsc&t=620s) Marketers combine context-aware models like FLUX or Midjourney with video synthesis engines to accurately embed authentic brand logos into cinematic AI scenes.
- [12:02](https://www.youtube.com/watch?v=uRL89WmkDsc&t=722s) The 'liar's dividend' drives algorithmic virality by splicing real footage with unexpected AI-generated endings, though it raises serious risks regarding synthetic media manipulation.
- [15:16](https://www.youtube.com/watch?v=uRL89WmkDsc&t=916s) Platforms like Showrunner signal an entertainment shift toward community-curated, interactive generative worlds where audiences co-create stories within persistent IP universes.
- [18:29](https://www.youtube.com/watch?v=uRL89WmkDsc&t=1109s) Interactive world models like Google Genie are overcoming spatial amnesia with persistent memory architectures, transforming static video into real-time navigable digital spaces.
- [20:22](https://www.youtube.com/watch?v=uRL89WmkDsc&t=1222s) Ad-supported business models can survive on synthetic engagement, but conversion-driven businesses must signal genuine human authenticity to build buyer trust.
- [22:01](https://www.youtube.com/watch?v=uRL89WmkDsc&t=1321s) Generative Engine Optimization (GEO) is replacing traditional SEO as users query LLMs, making verified bedrock sources like Wikipedia essential for activating AI recommendation weights.
- [24:40](https://www.youtube.com/watch?v=uRL89WmkDsc&t=1480s) The influx of automated content will cause an 'AI slop' gold rush and metric inflation, mirroring the dot-com bubble where transient churn collapses while enduring, high-value brands thrive.
## Detailed walkthrough
### [00:04](https://www.youtube.com/watch?v=uRL89WmkDsc&t=4s) Introduction: Two Years of AI Evolution and Content Creation
Dylan Jorgensson, known online as Dylan Curious, introduces himself and his background as an AI YouTuber and podcaster who has been covering artificial intelligence news for roughly two and a half years. The journey began in the early days of exploring large language model capabilities before progressing to text-to-image generation tools. He reflects on how rapidly the landscape has evolved, recalling the early days of using Midjourney via Discord roughly two and a half years prior, when generated images were "pretty rough and pretty ugly," followed by subsequent breakthroughs like DALL-E 2. Over time, his channel's coverage attracted an audience of entrepreneurs, gig economy workers building personal brands, startup founders establishing LLCs, and brick-and-mortar business owners—such as restaurateurs trying to generate enough social media virality to bring in customers and turn a profit.
Jorgensson offers a grounded reality check on using AI for social growth, emphasizing that while shortcuts exist to improve your odds, the process remains unpredictable and challenging. Even with his experience, several Instagram accounts he launched in recent months did not perform as well as anticipated, yielding valuable lessons. Conversely, he observed peers who seemed to be on the wrong track achieve unexpected success. His talk aims to break down the most practical tools, cutting-edge workflows, and actionable strategies currently shaping AI-driven content creation.
- **Speaker Background**: Dylan Jorgensson creates content under the channel name *Dylan Curious*, covering AI news and workflows for over two years.
- **Audience & Use Cases**: Gig workers building personal brands, new LLCs and startups, and brick-and-mortar businesses (e.g., restaurants) leveraging viral content for customer acquisition.
```mermaid
flowchart LR
accTitle: Evolution of AI Content Creation Tools
subgraph S1["Early Wave (~2.5 Years Ago)"]
A["Early LLMs"] --> B["Midjourney v1/v2 on Discord"]
B --> C["DALL-E 2"]
end
subgraph S2["Current Wave"]
C --> D["High-Fidelity Video (Google Veo)"]
D --> E["Image-to-Image & Keyframing"]
end
subgraph S3["Emerging Formats"]
E --> F["Persistent Mascots & Niche Channels"]
end
```
---
[02:20](https://www.youtube.com/watch?v=uRL89WmkDsc&t=140s) Google Veo and Crossing the Uncanny Valley
The most critical technological leap for social media content creators is Google's flagship video generation model, Google Veo (Veo 3), launched about two months prior. The model's realism is coming remarkably close to crossing the uncanny valley—the threshold where synthetic media feels unnatural or fake—by producing video that easily passes the "sniff test." Jorgensson highlights viral videos that spread without proper watermarks and convinced viewers of their authenticity, such as a widely circulated clip of rabbits jumping on a trampoline. By prompting for realistic camera artifacts, like grainy night vision or natural surveillance angles, creators make footage look indistinguishable from real life.
```
"It is very close to crossing the uncanny valley... It does video that at least at first passes the sniff test."
```
Beyond realism, Veo empowers creators to visualize scenarios that would otherwise be impossible, prohibitively expensive, or socially taboo to film. Creators can generate explosions, car crashes, deep underwater footage, remote Amazonian jungle expeditions, or historical time-travel scenarios—such as a viral trend depicting Jesus taking a selfie on the cross. Because creators can publish anonymously, they are pushing creative and cultural boundaries in ways a traditional human creator tied to their real-world identity might avoid. Accessible through Google's Gemini Ultra subscription (priced at roughly $300 per month, with introductory rates around $100 for the first two months), high-end cinematic generation is now readily available rather than remaining years away in 2030.
- **Google Veo (Veo 3)**: Google's state-of-the-art video generation model capable of highly convincing physics and natural camera styles (e.g., night vision).
- **Extreme Scenarios**: Enables seamless generation of dangerous or unreachable scenes (car crashes, underwater, Amazon jungle, historical reenactments).
- **Gemini Ultra Tier**: Costs approximately $300/month (discounted to $100/month for the first two months), making production-grade video accessible to individual creators.
---
[05:18](https://www.youtube.com/watch?v=uRL89WmkDsc&t=318s) Virtual Influencers, Mascots, and Image-to-Image Consistency
As virtual content matures over the next three to six months, social media will see a shift akin to the historic transitions from Hollywood to YouTube and later TikTok: the emergence of entirely unexpected synthetic brands. Because AI video models still encounter difficulties maintaining exact human facial consistency across multiple prompts, successful creators bypass this limitation by using non-human characters and animal mascots. A prominent example is the viral trend of placing a persistent Bigfoot character into mundane, humorous situations—such as grocery shopping, recording podcasts, commenting on political news, or sharing outdoor camping tips. Brands can easily adopt stylized mascots (such as rabbits or logo-based creatures) that react to real-time trending topics on X (formerly Twitter).
```
"I feel like Veo 3 might be, in my opinion, it's like the next tool like ChatGPT where maybe you don't use it, maybe you do, but it would almost be irresponsible to not know that it exists."
```
A major breakthrough in resolving narrative continuity is the recent introduction of image-to-image and first-to-last frame interpolation. Creators can supply a starting photo and a matching ending photo—such as an individual standing behind a conference podium—and instruct the model to morph the action in between while anchoring the character's exact appearance to the bookending frames. This technique unlocks cohesive long-form storytelling. Jorgensson points to the viral Instagram channel *Stormtrooper Vlogs* (created by PJ Accetturo), which documents the mundane, *Star Trek: Lower Decks*-style grunt work of Imperial Stormtroopers working for Darth Vader. By giving recurring names and personalities (like "Greg" and "Sam") to consistent helmeted characters, the creator turned simple AI-assisted sketches into a loyal, episodic narrative brand.
- **Mascot Strategy**: Using animals or non-human figures (e.g., Bigfoot, rabbits, helmeted characters) circumvents the human facial consistency problem in generative video.
- **Image-to-Image Keyframing**: Setting initial and final frames lets creators preserve environmental and character consistency across sequential video cuts.
- **Narrative Storytelling Example**: *Stormtrooper Vlogs* demonstrates how episodic, comedic framing of niche characters builds an engaged social media following.
[09:20](https://www.youtube.com/watch?v=uRL89WmkDsc&t=560s) The Uncanny Valley and Consistent Character Generation
Dylan Curious discusses how content creators are creatively circumventing the challenges of character consistency in generative AI video. He cites an example of a creator running a mock-vlog series featuring a Star Wars Stormtrooper: the character talks directly to a selfie camera about routine deployment orders to the forest moon of Endor—a setup loaded with dramatic irony for Star Wars fans. Moments later, behind-the-scenes chaos ensues as Jedi sprint past and the trooper flees in terror, gradually snowballing into an entire miniature cinematic universe. The clever technical trick behind choosing a Stormtrooper is that helmets bypass facial inconsistency. Generating a custom human face from varied angles and lighting invariably introduces subtle distortions that plunge the character into the uncanny valley. For now, non-human entities—masked troopers, animals, or stylized cartoon avatars—allow creators to achieve narrative continuity without triggering audience revulsion, while true photorealistic human character consistency remains just over the horizon.
```mermaid
flowchart TD
accTitle: Generative Character Consistency Workarounds
subgraph Bottleneck["Current AI Limitation"]
A["Named Human Faces"] --> B["Angle & Lighting Shifts"]
B --> C["Uncanny Valley Glitches"]
end
subgraph Workaround["Practical Creative Solutions"]
D["Masked Figures: Stormtroopers"]
E["Animals & Creatures"]
F["Stylized / Cartoony Models"]
end
subgraph Output["Production Pipeline"]
D & E & F --> G["Veo / Video AI Generation"]
G --> H["Consistent Episodic Storytelling"]
end
```
[10:20](https://www.youtube.com/watch?v=uRL89WmkDsc&t=620s) Preserving Brand Identity with Context-Aware Image Generation
Transitioning to commercial applications, Dylan explains that tools like Google Veo are not natively equipped to preserve precise brand iconography, such as accurately rendering a Coca-Cola logo onto a water bottle. To solve this, creators leverage cutting-edge image models like FLUX Context alongside Midjourney or DALL-E 3. These tools understand context sufficiently well to stamp authentic corporate branding or channel iconography directly onto generated assets.
Once a brand mark is baked cleanly into a base image, that frame can be fed into Veo for motion synthesis. Dylan illustrates the absurd creative freedom this unlocks for brand marketing:
- An engineering firm can place its genuine logo onto the hull of an Empire Star Destroyer with the tagline "we engineered this thing."
- An outdoor brand can place its custom cookware logo on a frying pan used by an AI-generated Bigfoot teaching wilderness survival tips.
- Creators can easily stage high-energy branded spectacles, moving from a static initial product render into a full-blown cinematic action set-piece or vehicle explosion.
[12:02](https://www.youtube.com/watch?v=uRL89WmkDsc&t=722s) Frame-Splice Illusions and the Algorithm's "Liar's Dividend"
Dylan explores an emerging trend combining authentic footage with AI-generated endings to engineer viral moments. He recounts a viral clip featuring a real pigeon standing on a car hood, pecking futilely at the windshield trying to reach a French fry on the dashboard. A creator grabbed the clip, trimmed it right at the point of impact, exported that single frame into Google Veo, and prompted the model to have the bird violently shatter the windshield glass to snatch the fry. The sudden shift from familiar real-world tension into an impossible, shocking victory leaves viewers stunned.
```mermaid
flowchart LR
accTitle: Frame-Splicing and Viral Algorithmic Feedback
A["Real Viral Video Clip"] --> B["Export Exact Split-Frame"]
B --> C["Veo Prompt: Pigeon Shatters Glass"]
C --> D["Composite Seamless Output"]
D --> E["Audience Confusion & Scrutiny"]
E --> F["Replay Loops: The Liar's Dividend"]
F --> G["Algorithmic Amplification"]
```
This illusion exploits a phenomenon known as the "liar's dividend": when viewers cannot instantly decipher whether footage is genuine, they loop it repeatedly, scrutinizing visual anomalies such as awkward lighting, facial quirks, or unnatural glass physics. Because platform algorithms interpret repeated viewings as high engagement, borderline-deceptive hybrid videos receive aggressive organic promotion. While an effective mechanism for social growth, Dylan warns creators and consumers alike of its darker implications:
- Bad actors can take real, hour-long political speeches or press conferences, excise a three-to-four-second fragment, and generate a photorealistic synthetic edit depicting a leader saying something they never uttered.
- Because the surrounding footage is authentic, casual viewers can be easily misled, democratizing visual manipulation techniques that once demanded costly Hollywood CGI pipelines.
[15:16](https://www.youtube.com/watch?v=uRL89WmkDsc&t=916s) Generative Worlds: Showrunner and Community-Curated Media
Looking ahead, Dylan forecasts a structural shift in social media and entertainment from static uploads to dynamic, 24-hour interactive content environments. He highlights Showrunner, a platform incubated over two years with support from *South Park* creators Matt Stone and Trey Parker. Because *South Park*'s cutout animation aesthetic is intentionally crude and flat, AI generation flaws blend into the visual texture without breaking immersion. Users can inject custom avatars into the universe, prompt and script narrative beats, and see polished community contributions produced into full 30-minute episodes.
- Following initial traction on Discord, the startup secured $10 million in funding from Netflix to explore integrating the engine into its streaming ecosystem.
- Social media strategy will evolve away from chasing individual view counts on static videos toward managing living, AI-driven digital universes where audiences co-create stories, memes, and inside jokes around shared intellectual property.
- Rather than authoring all content manually, brands and creators will serve as curators of AI-orchestrated virtual playgrounds.
[18:05](https://www.youtube.com/watch?v=uRL89WmkDsc&t=1085s) Real-Time Interactive Engines: Looking Ahead to Genie
Closing his forward-looking analysis, Dylan turns to real-time interactive generation, highlighting Google's debut of Genie just four days prior. While acknowledging that Genie may not deliver the immediate visual spectacle of high-definition video engines like Veo, he emphasizes that its real-time, playable world generation marks a foundational architectural leap toward fully responsive, interactive digital environments.
[18:29](https://www.youtube.com/watch?v=uRL89WmkDsc&t=1109s) Interactive World Models: Google Genie and Persistent AI Environments
The horizon of generative media is moving rapidly past static video clips toward fully interactive world generation. Emerging foundational models like Google DeepMind's Genie and Veo signal a shift where content moves beyond a pre-rendered, passive viewing experience into real-time interactive spaces. Creators will no longer just prompt isolated scenarios—whether that is a humorous "Stormtrooper vlog" or "little rabbits jumping on trampolines"—but will instead output navigable environments. In these experiences, the audience takes the reins: zooming in, turning around, and freely exploring the digital space while the underlying model dynamically synthesizes whatever exists behind the camera's original vantage point.
```mermaid
flowchart LR
accTitle: Diffusion Generation vs Persistent Interactive Environments
subgraph Traditional["Standard Video Diffusion"]
S1["Seed Prompt"] --> D1["Layer-by-Layer Denoising"]
D1 --> V1["Static Video Output"]
end
subgraph Genie["Genie / World Models"]
U1["Interactive Camera Nav"] --> P1["Persistent Latent Memory"]
P1 --> G1["Real-Time Environment Synthesis"]
end
```
Historically, a core limitation of diffusion models—which construct images layer by layer out of random noise seeds—has been a complete lack of spatial permanence; looking away from an object meant it disappeared from the model's memory. Google engineers have begun solving this through persistent memory architectures, demoed mere days prior, allowing rendered worlds to remain stable when the camera pans away and returns. This breakthrough completely reinvents user engagement. Instead of passively swiping past a short video, viewers are invited to dwell inside the scene, inspect their surroundings, and prompt additions to the world. That exploratory curiosity captures the same compulsive dopamine hit of social media feeds, but transforms it into a persistent, multi-dimensional playground.
- **Spatial Autogeneration**: Genie dynamically infers and renders unseen surroundings on the fly as the user maneuvers the camera through the scene.
- **Solving Hallucinatory Amnesia**: Novel persistent memory architectures allow diffusion engines to remember previously rendered spaces rather than discarding them when the perspective shifts.
- **Interactive Retention**: Content engagement shifts from passive short-form video swiping to exploratory viewing where audiences navigate, inspect anomalies, and co-prompt environments.
---
[20:22](https://www.youtube.com/watch?v=uRL89WmkDsc&t=1222s) Creating for Non-Human Audiences vs. The Human Trust Funnel
A quiet, unsettling reality facing digital media is the reality of creating for non-human audiences. Beneath surface-level metrics, creators and marketers are increasingly haunted by the suspicion that their engagement metrics are hollowed out by automated traffic—wondering whether incoming views and comment section discourse stem from actual domestic users or automated LLM agents. Navigating this environment requires understanding the fundamental divide between vanity-driven growth and commercial survival.
If an account's business model relies strictly on top-of-funnel social media advertising revenue, algorithmic inflation and bot traffic might not present an immediate threat; AI-generated advertisements continue to run, brand budgets are expended, and the platform delivers an ad-revenue split regardless. However, for businesses that rely on direct customer conversion—where success hinges on a human being deciding to "open up their wallet and buy something"—relying on automated engagement is fatal. Winning real human conversion demands strict authenticity:
> "If it's trying to get people to follow a link and you want a real human connection, it's about a funnel, then you're going to want to play to that strength and be very careful about curating your social media strategy... Don't make it half human. Don't generate too much B-roll because people will start to distrust you."
- **The Creator Audience Divide**: Ad-revenue models can survive monetizing synthetic engagement and automated placement, but conversion-driven businesses require verified human intent.
- **The Synthetic B-Roll Trap**: Over-saturating personal content with synthetic stock footage or generic AI generation undermines trust, alienating high-value prospects.
- **The Human Premium**: Creators must actively signal in-person reality, reminding audiences through physical, unscripted moments that there is an authentic human operator behind the screen.
---
[22:01](https://www.youtube.com/watch?v=uRL89WmkDsc&t=1321s) Generative Engine Optimization (GEO) and Establishing LLM Authority
Traditional web search is undergoing an existential transition: "there's also no more Google really like everybody's going to LLM." Search habits are moving rapidly away from indexing query links and toward synthesized generative answers, replacing standard SEO with Generative Engine Optimization (GEO). As large language models ingest the live web in near-real time through iterative evaluation loops and reinforcement learning, algorithms look for authoritative bedrock sources to establish ground truth before generating answers.
```mermaid
flowchart TD
accTitle: Generative Engine Optimization Workflow
subgraph Ingestion["Authoritative Ingestion"]
W["Wikipedia Entries"]
P["Major Publications / NYT"]
end
subgraph Synthesis["LLM Verification Engine"]
L["Real-Time Web Scraping Loops"]
R["Reinforcement Learning Signals"]
end
subgraph Discovery["User Retrieval"]
Q["Conversational User Prompt"]
A["Direct Ground-Truth Entity Recommendation"]
end
W --> L
P --> L
L --> R
R --> A
Q --> A
```
Under this regime, structured digital authority repositories like Wikipedia carry unprecedented leverage. LLMs assign disproportionate probability weights to verified encyclopedic data and major legacy press outlets like *The New York Times*. Dylan illustrates this mechanic through a personal case study involving his fiancée, a resident magician performing at the Westgate in Las Vegas. Initially, despite having an active business website, her digital footprint yielded negligible referral traffic from AI engines:
> "We kind of really started thinking about Wikipedia and trying to, you know, contact people there and say, 'Hey, like here's some credible links. Do you mind adding her as a Wikipedia page?' And it was like night and day. As soon as the Wikipedia page showed up, the generative engine optimization said, 'Oh, this isn't just some random website about a magician. This is like a trusted source.'"
Following the publication of her verified entry, conversational queries like "who are the top female magicians in the world" or "the best magicians in Las Vegas" immediately elevated her name into zero-shot conversational recommendations. For brands and public figures alike, winning visibility across the modern web requires deliberately feeding high-trust, verified citation nodes to activate neural weights across the dominant frontier models.
- **From Blue Links to Probabilistic Answers**: LLMs bypass conventional page rankings, directly returning synthesized recommendations based on the highest-confidence sources available.
- **The Wikipedia Ingestion Moat**: Because Wikipedia requires neutral volunteer contributions and strict sourcing criteria, LLMs treat its entity graphs as verified ground truth.
- **Algorithmic Weight Activation**: Securing profiles on structured encyclopedias and legacy publications directly increases the model's activation weights, cementing the subject as a canonical answer in conversational prompts.
---
[24:40](https://www.youtube.com/watch?v=uRL89WmkDsc&t=1480s) The Loneliness Behind the Spotlight and Navigating the AI Slop Gold Rush
The immediate future of social media carries an acute psychological and structural paradox described as "the loneliness behind the spotlight." As generative tools drastically lower production friction, the social landscape will undergo a disorienting shift where "everybody feels famous and nobody is." Creators will log in to watch their view counters skyrocket into inflated thousands or millions, yet experience flattening ad payouts as synthetic distribution floods the platforms.
```mermaid
flowchart TD
accTitle: Transition from AI Slop to High-Value Brands
subgraph Flood["Early Generative Wave"]
V["Unfiltered Generative Video Models"]
S["Surge of Low-Effort AI Slop"]
end
subgraph Filtering["Platform Adaptation"]
ML["Machine Learning Quality Filters"]
H["Human Feedback & Verified Engagement Signals"]
end
subgraph Outcome["Market Consolidation"]
D["Pets.com Commoditized Casualties"]
A["Resilient Amazon-Caliber Brands"]
end
V --> S
S --> ML
ML --> H
H --> D
H --> A
```
Platforms like YouTube and Instagram are set to be inundated with an avalanche of automated, AI-generated video. Because social networks do not yet have mature machine learning filters capable of discerning nuanced craft from low-effort churn, feeds will temporarily descend into an era of low-grade slop. Over time, algorithmic engineering will adjust, prioritizing genuine engagement metrics, human verification signals, and community feedback loops to discount empty automation.
Despite this incoming wave of noise, this inflection point functions exactly like an early-stage speculative boom. Drawing an explicit comparison to the original dot-com bubble, Dylan emphasizes that while a tremendous amount of low-grade generative output will inevitably implode like Pets.com, persistent founders who maintain clear creative direction have an unprecedented window to construct resilient, generation-defining brands like Amazon. Surviving the transition requires leaning fully into visionary toolsets like Veo while fiercely anchoring production to an authentic creative core.
- **Metric Inflation vs. Financial Parity**: Skyrocketing aggregate views across platforms are met with diluted revenue splits and pervasive uncertainty over audience authenticity.
- **The Approaching Slop Trough**: Consumer feeds will encounter an unfiltered influx of synthetic generative media before platforms roll out robust machine learning quality guardrails.
- **The Dot-Com Playbook**: In an environment crowded by transient, disposable automated content, lasting commercial value belongs to creators who deploy advanced generative tooling in service of an uncompromising creative vision.
> [02:20](https://www.youtube.com/watch?v=uRL89WmkDsc&t=140s) It is very close to crossing the uncanny valley... It does video that at least at first passes the sniff test. — Dylan Jorgensson
> [05:18](https://www.youtube.com/watch?v=uRL89WmkDsc&t=318s) I feel like Veo 3 might be, in my opinion, it's like the next tool like ChatGPT where maybe you don't use it, maybe you do, but it would almost be irresponsible to not know that it exists. — Dylan Jorgensson
> [10:20](https://www.youtube.com/watch?v=uRL89WmkDsc&t=620s) we engineered this thing. — Dylan Jorgensson
> [20:22](https://www.youtube.com/watch?v=uRL89WmkDsc&t=1222s) open up their wallet and buy something — Dylan Jorgensson
> [20:22](https://www.youtube.com/watch?v=uRL89WmkDsc&t=1222s) If it's trying to get people to follow a link and you want a real human connection, it's about a funnel, then you're going to want to play to that strength and be very careful about curating your social media strategy... Don't make it half human. Don't generate too much B-roll because people will start to distrust you. — Dylan Jorgensson
> [22:01](https://www.youtube.com/watch?v=uRL89WmkDsc&t=1321s) there's also no more Google really like everybody's going to LLM. — Dylan Jorgensson
> [22:01](https://www.youtube.com/watch?v=uRL89WmkDsc&t=1321s) We kind of really started thinking about Wikipedia and trying to, you know, contact people there and say, 'Hey, like here's some credible links. Do you mind adding her as a Wikipedia page?' And it was like night and day. As soon as the Wikipedia page showed up, the generative engine optimization said, 'Oh, this isn't just some random website about a magician. This is like a trusted source.' — Dylan Jorgensson
> [24:40](https://www.youtube.com/watch?v=uRL89WmkDsc&t=1480s) the loneliness behind the spotlight — Dylan Jorgensson
> [24:40](https://www.youtube.com/watch?v=uRL89WmkDsc&t=1480s) everybody feels famous and nobody is. — Dylan Jorgensson
## Mentioned
Dylan Jorgensson, Dylan Curious, Ai4, YouTube, Instagram, X (formerly Twitter), TikTok, Discord, Midjourney, DALL-E 2, DALL-E 3, Google Veo (Veo 3), Gemini Ultra, ChatGPT, PJ Accetturo, Stormtrooper Vlogs, Star Wars, Star Trek: Lower Decks, FLUX Context, Coca-Cola, Showrunner, Matt Stone, Trey Parker, South Park, Netflix, Google DeepMind Genie, Wikipedia, The New York Times, Westgate Las Vegas, Pets.com, Amazon
## Video description
Dylan Curious, AI YouTuber, Dylan Curious (YouTube)
Abstracts
Join me as I take you on a personal journey through the transformative impact artificial intelligence has had on my approach to social media. I'll share how AI has already reshaped my strategies and content creation process, and explore exciting insights into how I foresee it revolutionizing my approach even further.
You'll walk away with practical tips, innovative tools, and key insights on upcoming trends you need to know, whether you're building a personal brand or managing your company's online presence. Don't miss out on this opportunity to stay ahead of the curve!
Google Veo助推合成内容泛滥，创作者转向直销
激增的生成式视频在推高社交指标的同时侵蚀着广告收益，迫使创作者放弃合成内容的捷径，转向经核验的人际信任与结构化权威来推动真实转化。
Dylan Jorgensson探讨了生成式媒体从早期文生图模型向Google Veo等逼真视频引擎及Google Genie等交互式世界模型的演进历程。他演示了解决角色一致性问题的实用技巧——如利用非人类吉祥物、图生图关键帧和画面拼接的病毒式循环——同时展示了Showrunner等平台如何指向社区共创娱乐的新方向。在从制作转向内容发现的过程中，他概述了生成式引擎优化（GEO）如何依托Wikipedia等可信实体来确保在LLM搜索结果中的曝光度。最终，他指出了合成机器人流量与低质“AI垃圾内容（AI slop）”的激增，将当前的淘金热比作互联网泡沫，并强调持久的商业成功属于高信任度、真实的人类品牌。
Google Veo（Veo 3）等前沿视频生成模型正逼近跨越“恐怖谷”的临界点，创作者能够以极低门槛生成高度逼真的画面，并模拟车祸、深海探险等危险或高成本场景。 (2:20)
针对当前 AI 生成人脸容易出现形变的不稳定性，创作者通过选用动物、非人吉祥物或头盔角色（如《暴风兵 Vlog》），并结合首尾帧插值技术，成功维系了连续剧集式的角色叙事。 (5:18)
创作者通过将真实视频与 Veo 生成的惊悚反转画面拼接，利用“说谎者的红利”（Liar's Dividend）诱导观众反复回放核验细节，进而促使平台推荐算法进行病毒式扩散。 (12:02)
以 Showrunner 与 Google Genie 为代表的互动世界模型，正通过引入持久记忆架构解决空间遗忘问题，推动生成式内容从静态单向视频迈向可实时探索的数字空间。 (18:29)
传统 SEO 正在向生成式引擎优化（GEO）转型，LLM 模型在生成式问答中高度依赖维基百科等权威信源作为事实依据，成为品牌获取对话式推荐权重的关键资产。 (22:01)
Reading summary
AI content creator Dylan Jorgensson argues that while generative video and world-building models are radically lowering production friction, sustainable digital growth requires navigating algorithmic distortions, bot-inflated engagement, and character consistency limits by prioritizing human trust and verifiable authority.
Creative Workarounds for Video Consistency and Branding (2:20)
Google Veo (Veo 3) and emerging generative video engines can now produce scenes passing visual "sniff tests," allowing creators to produce impossible or dangerous sequences, like car crashes and underwater expeditions, via tiers like Gemini Ultra ($300 monthly, or $100 intro rates). However, generating consistent human faces across changing angles and lighting remains an obstacle, often triggering uncanny valley distortions. To circumvent this, Jorgensson observes creators utilizing non-human mascots like Bigfoot, helmeted figures like the Stormtrooper Vlogs series, and first-to-last frame interpolation to preserve visual continuity.
For commercial applications, models like FLUX Context stamp authentic corporate logos onto base images before animating them in Veo. This lets brands place real iconography directly inside synthetic cinematic scenes, turning static branded product renders into high-energy action set-pieces.
Algorithmic Promotion and the Liar's Dividend (12:02)
Creators are increasingly splicing authentic video clips into generative endings to engineer viral moments. In one cited example, a real recording of a pigeon pecking at a windshield was cut at the exact point of impact and extended in Google Veo to show the bird shattering the glass. Jorgensson explains that such hybrid clips capitalize on the "liar's dividend": viewers unable to discern reality loop footage repeatedly to inspect visual anomalies, which platform recommendation algorithms interpret as high engagement.
While this mechanism drives rapid organic reach, Jorgensson cautions that the underlying technique carries severe risks. Bad actors can take brief fragments of authentic political speeches and generate photorealistic synthetic edits, effectively democratizing visual manipulation techniques that once required expensive Hollywood visual effects pipelines.
Moving from Passive Feeds to Interactive Worlds (15:16)
Entertainment and social media formats are shifting from static, linear video uploads toward dynamic, persistent environments. The platform Showrunner—backed by South Park creators Matt Stone and Trey Parker and funded with $10 million from Netflix—leverages cutout animation aesthetics where generative flaws blend naturally into the visual texture, allowing communities to script episodes and co-create digital universes.
Concurrently, real-time world models such as Google DeepMind's Genie are introducing persistent memory architectures. By addressing diffusion models' historical spatial amnesia—where off-camera elements vanished when the viewpoint shifted—these systems enable audiences to navigate and inspect synthesized 3D spaces in real time, transforming passive video swiping into exploratory interactive environments.
Generative Engine Optimization and the Human Conversion Funnel (20:22)
As conversational AI engines supplant traditional search, Jorgensson emphasizes Generative Engine Optimization (GEO) over standard SEO. Large language models synthesize answers by drawing on verified authoritative nodes; establishing a Wikipedia page for a resident Las Vegas magician, for example, immediately elevated her into zero-shot conversational recommendations for top performers.
Meanwhile, creators facing bot-heavy engagement must separate vanity metrics from commercial intent. While programmatic ad revenue models can tolerate automated traffic, businesses requiring paying customers cannot survive on hollow views. Jorgensson warns that oversaturating personal channels with synthetic B-roll alienates prospects, advising conversion-driven businesses to visibly signal genuine, unscripted human presence.
Navigating the Slop Wave and Platform Inflation (24:40)
The sudden influx of generative production creates what Jorgensson terms "the loneliness behind the spotlight," where creators see astronomical view counts coupled with diluted ad payouts and uncertain audience authenticity. Social platforms currently lack mature machine learning filters to separate high-effort craft from low-grade synthetic spam, creating a temporary wave of unfiltered "AI slop."
Comparing the current landscape to the early dot-com boom, Jorgensson contends that while transient, low-effort generative channels will inevitably collapse like Pets.com, operators who anchor advanced tools like Veo to a clear, disciplined creative vision have an opening to establish durable, generation-defining media properties.
Jorgensson concludes that generative AI will reward operators who deploy advanced video and world-modeling tools as strategic accelerators rather than substitutes for trust. The broader challenge for both platforms and publishers remains filtering out low-grade synthetic churn while sustaining genuine human connection and conversion.
[0:04] Dylan Jorgensson: pretty rough and pretty ugly
[2:20] Dylan Jorgensson: It is very close to crossing the uncanny valley... It does video that at least at first passes the sniff test.
[5:18] Dylan Jorgensson: I feel like Veo 3 might be, in my opinion, it's like the next tool like ChatGPT where maybe you don't use it, maybe you do, but it would almost be irresponsible to not know that it exists.
[10:20] Dylan Jorgensson: we engineered this thing.
[20:22] Dylan Jorgensson: open up their wallet and buy something
[20:22] Dylan Jorgensson: If it's trying to get people to follow a link and you want a real human connection, it's about a funnel, then you're going to want to play to that strength and be very careful about curating your social media strategy... Don't make it half human. Don't generate too much B-roll because people will start to distrust you.
[22:01] Dylan Jorgensson: there's also no more Google really like everybody's going to LLM.
[22:01] Dylan Jorgensson: We kind of really started thinking about Wikipedia and trying to, you know, contact people there and say, 'Hey, like here's some credible links. Do you mind adding her as a Wikipedia page?' And it was like night and day. As soon as the Wikipedia page showed up, the generative engine optimization said, 'Oh, this isn't just some random website about a magician. This is like a trusted source.'
[24:40] Dylan Jorgensson: the loneliness behind the spotlight
[24:40] Dylan Jorgensson: everybody feels famous and nobody is.
Original transcript
[0:04] Hey everybody. I guess we might as well
[0:05] just get started. Um, my name is Dylan
[0:08] Jorgensson, my full name, but I create
[0:11] videos for YouTube under the name Dylan
[0:13] Curious. Uh, I am here with another AI
[0:16] YouTuber. We've been doing podcasts.
[0:18] We've been covering artificial
[0:20] intelligent news for um about two years
[0:24] now. So it started as just sort of
[0:25] seeing what large language models were
[0:27] capable of and then uh slowly starting
[0:30] to be familiar with some of these AI
[0:32] generation tools. Kind of started with
[0:34] this tool called Midjourney where you
[0:36] could go into Discord and create um
[0:39] basically whatever you could imagine and
[0:41] like that looked pretty rough and pretty
[0:44] ugly only two and a half years ago. And
[0:46] then it started getting pretty good and
[0:48] a whole bunch of other tools including
[0:49] the second version of Dolly um from
[0:51] Gemini showed up and we were covering
[0:54] this for our audience on YouTube and
[0:56] found that there was you know a lot of
[0:58] people that wanted to use these tools to
[1:00] sort of change the way they run their
[1:02] business and uh especially entrepreneurs
[1:04] who wanted to go out there and create
[1:06] some sort of a brand on their own. For
[1:08] some people that's just a person they're
[1:09] in the gig economy and they want to
[1:11] create a brand for themselves. In other
[1:13] cases, it's about new LLC's, new
[1:15] startups. Sometimes it's about taking a
[1:17] restaurant um from, you know, the ground
[1:20] level to something that is somewhat v
[1:22] viral on social media enough that it can
[1:24] bring in enough customers to make your
[1:26] restaurant profitable. And I've sort of
[1:29] over the last two and a half years
[1:30] talked with enough people in the
[1:31] communities to know um that it's not
[1:35] easy, it's not predictable, and there
[1:38] are some shortcuts that can make your
[1:40] luck a little bit better. But um I
[1:42] personally thought I had kind of it
[1:44] figured out in a few ways and I've
[1:45] created some Instagram accounts in the
[1:47] last few months that haven't done as
[1:49] well as I thought and um I've got some
[1:51] learning lessons from that too. But I
[1:53] have other people who I thought were
[1:54] going down the wrong track that did
[1:55] good. So I'll just kind of go through
[1:57] all of that with you and I think
[1:58] there'll be some stories maybe that will
[2:00] help you. I'll tell you about some of
[2:01] the tools that are kind of on the
[2:03] forefront right now that people are
[2:05] using to create content and how they're
[2:07] using it and the combinations of the
[2:08] workflows. So, um, but also totally open
[2:11] if there's questions along the way or
[2:12] I'll have a little question kind of Q&A
[2:14] thing at the end. So, it'll be a half
[2:15] hour talk, probably be done like around
[2:17] 2:30. So, um, yeah. So, I'll start by
[2:20] talking about what I think is the most
[2:22] important tool for content creators
[2:24] right now that are um, thinking about
[2:27] what they want to do on social media and
[2:30] that is V3. So, Google has launched
[2:33] about two months ago the newest video
[2:36] generation model. Um it's important to
[2:38] understand just how good this model is.
[2:41] Uh it is it is very close to crossing
[2:45] the uncanny valley. So the uncanny
[2:46] valley is this idea that when you
[2:48] generate something it looks fake, right?
[2:50] But this does video that at least at
[2:52] first passes the snuff test. You will
[2:56] see a lot of videos in the last month or
[2:58] so that have gone viral and they were
[3:01] convincing people. They were not
[3:02] watermarked correctly, but people
[3:03] thought they were real videos. Like I
[3:05] don't know if anybody did anybody see
[3:06] the rabbits that were jumping on the
[3:07] trampoline. Yeah, that thing was going
[3:09] crazy. But that wasn't real. But the
[3:11] fact that VO can with with a creative
[3:14] prompt can do something like make it
[3:16] look like it's night vision, make it
[3:18] look like it's some kind of a camera
[3:20] that's capturing something naturally.
[3:22] These sort of prompts move it across the
[3:24] uncanny valley to a point where it's
[3:26] totally believable. And you know, I
[3:29] don't know what they could have maybe
[3:30] tra a trampoline company would have
[3:31] maybe been able to link the original to
[3:33] some kind of a product and and done a
[3:35] lot on that. So, um, what I've seen with
[3:39] V3, uh, started with just prompting. Um,
[3:43] you can say put, you know, some guy at a
[3:45] conference, make him look like he knows
[3:47] what he's doing and put him in front of
[3:49] some sort of generic background and it
[3:50] will do a pretty good job of that. But
[3:52] more importantly, it can let your mind
[3:54] do stuff that you would have never
[3:56] thought about any other creator doing.
[3:59] And that's where I feel like the magic
[4:01] is. So, um, V3 will generate videos of
[4:06] explosions. It will generate videos of
[4:09] car crashes. It will put you in parts of
[4:12] the world that are very hard to get to,
[4:13] the Amazon, into the jungle, underwater.
[4:16] It can take you back in time. there was
[4:19] this entire viral trend where people
[4:21] were capturing um like Jesus on the
[4:24] cross doing a selfie cam right before
[4:26] like the moment you know and that just
[4:29] went incredibly viral because they're so
[4:30] hard to film and you can also do it
[4:32] mostly anonymously so people are able to
[4:34] push the boundaries of what you know a
[4:36] human connected to a piece of content
[4:37] like this probably wouldn't do because
[4:39] of the social impact and you know this
[4:42] world hasn't existed for more than I
[4:44] would say two or three months I mean
[4:45] certainly in the past if you had enough
[4:47] money, you could do it. But this is like
[4:49] a tool that anyone in this room can sign
[4:51] up for right now. Everybody has a Google
[4:53] account already. The Gemini package, the
[4:55] Ultra package, which would allow you to
[4:57] really create enough content for a
[4:59] social media feed, is um $300 a month, I
[5:02] think, and for the first two months,
[5:04] it's only $100. Um, so here we are, like
[5:08] we're in this world now that I really
[5:10] didn't expect to happen until maybe
[5:12] 2030. It it happened way faster than I
[5:14] expected. Um, but it means that if you
[5:18] want a virtual influencer, even if it
[5:20] doesn't take off right away, you're
[5:22] looking at a three to six month curve
[5:24] before the uncanny valley is totally
[5:26] squashed. I'm convinced that the way
[5:28] originally Hollywood created people and
[5:31] then the YouTubers sort of came in and
[5:33] then sort of the Tik Tok generation. The
[5:35] next thing is there will be brands that
[5:37] nobody um expected to be full brands.
[5:40] So, for example, there is an entire
[5:43] trend of taking Bigfoot right now, like
[5:45] a Bigfoot character, and putting them in
[5:47] funny situations. Like, they're
[5:48] shopping, they're on podcasts, they're
[5:50] giving commentary on the president,
[5:52] they're uh they're giving camping tips.
[5:56] And the reason you'll notice it's
[5:57] Bigfoot and not a person is because V3
[6:00] is still a little bit, it's a little bit
[6:02] hard to make the exact same character
[6:04] over and over again. However, something
[6:07] like an animal, like if you just want a
[6:08] rabbit, you just want a Bigfoot, it can
[6:11] and and you have enough description
[6:13] there, you can usually recreate that.
[6:16] So, this is a good time to create
[6:18] something that if your brand has a logo,
[6:20] like, you know, put a real face. I don't
[6:22] know if your character, maybe you have a
[6:23] rabbit as a character for your uh my
[6:26] girlfriend's a magician, so like we talk
[6:27] about rabbits a lot. I like rabbits as a
[6:29] logo. You can just make some kind of a
[6:31] rabbit. It can be doing really fun
[6:33] stuff. it can be making comments on
[6:35] anything that's in, you know, that's
[6:37] trending on on X right now. And it's a
[6:40] just a fascinating moment. So, I feel
[6:41] like VO3 might be, in my opinion, it's
[6:45] like the next tool like chat GBT where
[6:47] maybe you don't use it, maybe you do,
[6:49] but it would almost be irresponsible to
[6:51] not know that it exists and be able to
[6:53] create at least a few videos with your
[6:56] imagination with one paragraph of text
[6:59] that that you can send to your friends
[7:01] and just see if they think it's
[7:03] interesting. Um, it helps you take your
[7:05] imagination and put it down. And
[7:07] internally inside of companies, this is
[7:09] also a great tool for, you know,
[7:11] expressing what you want, your creative
[7:13] vision or whatever it is to somebody
[7:15] else or kind of building community. Like
[7:17] if you have a Discord community, if
[7:20] you've got um some sort of a thread or
[7:22] even an email list, it's a great way to
[7:24] totally customize it. Um you can find
[7:26] your best customers and maybe not put
[7:29] them in a video, but you almost could.
[7:30] Um, but there's also another aspect to
[7:33] V3 that came out in the last couple
[7:35] weeks, which is there is an image to
[7:38] image process. And this wasn't there
[7:40] originally. So, if you've noticed in the
[7:42] last couple weeks, videos have gotten
[7:44] even cleaner. It's because you can take
[7:47] a photo like you could snap a photo of
[7:49] me like in front of this podium and then
[7:51] you could take another photo of me in
[7:54] front of the podium and put it as the
[7:55] beginning and the end and then tell a
[7:56] story that I transform into something
[7:59] and then go back to it. And because it
[8:01] knows what the final frame and the
[8:03] initial frame looks like, this is an
[8:05] incredibly overlooked tool for creating
[8:07] a long- form narrative. If you're making
[8:09] a long- form narrative, what you need to
[8:11] do is sort of connect the scenes
[8:13] together to have some sort of character
[8:14] consistency and there are other tools
[8:16] out there like you can make a model in
[8:18] 3D or you can actually take photos of a
[8:20] real person and keep using them over and
[8:22] over again as the initial frame and you
[8:24] end up with a much a much better
[8:25] product. So, um, we're just on the cusp.
[8:28] There's going to be tools that probably
[8:30] can put this whole storyline together
[8:31] and sort of automate it really quickly.
[8:33] But what that means is I think that the
[8:35] future of social media now is very niche
[8:40] and it's very um non-changing when you
[8:43] get there. So for example, there is a uh
[8:46] really funny Instagram channel that's uh
[8:48] called Stormtrooper Vlogs. The idea that
[8:51] this creator came up with uh PJ Tretto,
[8:54] I think is his name, um was what would
[8:57] happen like what's what's the life of a
[8:59] Stormtrooper like working for Darth
[9:01] Vader? Like what is it like? you know,
[9:03] basically like lower decks on the
[9:04] Starship Enterprise. They do the grunt
[9:06] work. They they they document their
[9:08] stories. But he was so smart because he
[9:11] created a couple of them and they went
[9:12] viral. But then he just decided to make
[9:14] them characters, right? He decided to
[9:16] name one of them Greg and one of them
[9:17] Sam, I think, or something. And they and
[9:20] it's really funny like like one in one
[9:22] video he's like talking to his selfie
[9:23] camera and he's saying it's a
[9:24] Stormtrooper and he's like going to
[9:26] Endor. Should be routine, not expecting
[9:28] much, but of course everybody in the
[9:29] Star Wars universe knows what happens on
[9:31] Endor. And then he you can see it from
[9:33] like behind the scenes and the Jedi run
[9:35] by and they get scared and run off and
[9:37] all this stuff. But it's slowly become
[9:39] this entire little universe. And that
[9:42] one's that's a little weird because the
[9:44] reason he chose the uh the Stormtrooper
[9:47] is the Stormtrooper is a very generic
[9:49] face and you can generate him from many
[9:51] different angles and every time you're
[9:53] going to think, "Oh, that's that's the
[9:54] Stormtrooper." Whereas the problem is if
[9:56] you create a character, you name him or
[9:58] her, and you want that exact face every
[10:00] time when you look at them from
[10:01] different angles, it's it's going to be
[10:02] a little different. So that's why we're
[10:04] talking about the uncanny valley. Like
[10:05] now is the moment where if you have an
[10:07] animal or something with a mask or
[10:09] something kind of more cartoony or
[10:11] creative, you can get away with it. And
[10:12] then we're almost there for like photo
[10:15] human realistic stuff. But um V3 is how
[10:18] a lot of that's being done. Now there
[10:20] are other tools that are also on the
[10:21] cutting edge. Um there's a tool called
[10:24] Flux Context which you should probably
[10:26] know about. Um this tool would be a
[10:29] little bit more as far as I understand
[10:31] it's mostly for images but of course you
[10:33] can take an image and then you can
[10:34] animate it. So um the power of it though
[10:37] is it's very it's very good at
[10:39] understanding the context of something
[10:41] that would have a logo on it, writing on
[10:44] it, branding on it. So V3 isn't your
[10:47] tool, your go-to tool, if you want to
[10:49] put a like Coca-Cola logo or something
[10:51] like on this water bottle. But if what
[10:53] you care about is your brand, you
[10:55] probably have a a logo for your business
[10:57] or you have something that you'd want
[10:58] for your YouTube channel, you can take
[11:00] that now and stick it on there. So, if
[11:03] you have a Stormtrooper vlog or you have
[11:06] Bigfoot giving camping tips, you can now
[11:09] get your logo onto the pan that um you
[11:13] know that the the Bigfoot is like
[11:15] teaching tips on or I don't know if
[11:17] you're on you know the I guess Star
[11:19] Destroyer or something you could you
[11:21] know put your logo like we engineered
[11:22] this thing or whatever if you're an
[11:24] engineering company. So, the sky's is
[11:25] the limits. All it kind of depends on
[11:27] whatever you're working on. But the cool
[11:29] thing is you you can take that logo, you
[11:31] can bake it into an initial photo and
[11:34] you would use uh something like Dolly
[11:36] E3, you could use Midjourney. Um someone
[11:39] else just told me another tool, but
[11:40] there's a few of the image generators
[11:42] out there that I'm missing, but you can
[11:44] check those out. And um then generate it
[11:47] with your logo using context flux and
[11:50] then take that to VO, animate it, and
[11:53] now we're talking about a full branding
[11:55] car explosion Bigfoot thing. And it's a
[11:59] very fascinating world. Um, now to build
[12:02] even one more layer onto this, and this
[12:04] is something that I've only started
[12:05] seeing, so I'm kind of taking a guess
[12:07] here that it's going to become much more
[12:08] popular, but there's already a series of
[12:11] real videos that have gone viral over
[12:13] the last few years. Um, one for example
[12:15] was a someone was sitting in their car
[12:18] and there was a bird like a pigeon that
[12:20] landed on the front hood of their car
[12:22] and it was trying to peck through the
[12:25] glass to get to a French fry that was on
[12:26] the dashboard. Like just play with me
[12:29] here. Like this this really happened and
[12:30] it was very v it was kind of you're like
[12:32] a poor pigeon like wants the French fry
[12:34] but can't get through the glass. So
[12:36] someone took that exact video and at the
[12:39] moment when the bird pecks at the glass,
[12:43] cut the frame. So they took it into an
[12:45] editing software and then they said like
[12:47] this is the end of the video, but then
[12:49] also exported that frame, brought it up
[12:52] to V3 and then said make it so that the
[12:55] the pigeon actually breaks the glass and
[12:57] gets the French fry. Incredible.
[12:59] incredible video because you're watching
[13:02] it, you've got this tension, and then
[13:04] you see this moment where it it just
[13:06] shocks you. Like you're like, "Oh my
[13:07] god, that pigeon broke through the
[13:08] glass, got the French fry, and like is
[13:11] actually the champion and won." And it
[13:14] throws you off so bad because you didn't
[13:16] expect that. And you know, watching it
[13:18] over and over again, a few after a few
[13:20] minutes, I was able to notice that,
[13:22] okay, this was not completely AI
[13:25] generated, but it was beginning as a
[13:27] real video. It ended AI generated and I
[13:30] they also probably I made the video more
[13:33] viral because there's something called
[13:34] the liars dividend which is this idea
[13:37] that when you're not quite sure if a
[13:39] video is
[13:41] real or not you just you keep playing
[13:43] it. You're like studying it like
[13:44] something seems off like the eyes are a
[13:46] little weird or the way the glass breaks
[13:48] is a little weird and that unfortunately
[13:51] tells the algorithm, hey man, people are
[13:53] like re-watching this. They're into it.
[13:55] So I would warn that I would warn for
[13:58] people who are on social media and I
[14:00] would say to people who are trying to
[14:01] grow on social media to sort of be aware
[14:04] of this lighter dividend. It could be a
[14:05] tool for growth. It could also be
[14:07] something that could kind of mess up
[14:08] your feed. So just be aware of it. But
[14:10] um yeah, this idea that real videos and
[14:13] moments of them are cut and then put
[14:16] into tools like V3 and then brought back
[14:18] into the video. And in the in the case
[14:20] of the bird, the second half of the
[14:21] video is not real. But in theory, I
[14:24] don't see any reason why you couldn't
[14:25] take a a long form video of some the
[14:28] president speaking or something, cut out
[14:30] one specific moment, say this is the
[14:33] beginning frame, this is the end frame,
[14:35] generate him saying XYZ that he didn't
[14:38] say. There's maybe a, you know, a night,
[14:40] maybe you're on some random YouTube
[14:41] channel, you watch a 60-minute press
[14:43] conference, three or four seconds of it
[14:45] are fake, but it's really hard to tell
[14:48] and it can convince you of something,
[14:51] you know? I mean that stuff is very
[14:52] possible. Hopefully people don't use it
[14:54] that way, but just to know that these
[14:55] kind of tools are in everybody's hands
[14:57] now, not just the kind of effort that it
[14:59] used to take to do uh CGI. So yeah, so
[15:02] to start the conversation, that's the
[15:04] main thing. V3 makes these amazing
[15:06] videos. Um there's many ways to use it.
[15:09] There's many things that I'm starting to
[15:11] see on social media and in social media.
[15:14] And then I'll take it kind of to the
[15:16] next level. So now I'm going to jump
[15:17] ahead a little bit. So these not exactly
[15:19] practical today, but this is where I
[15:21] would be like aligning my self if I was
[15:24] working at a social media company and
[15:25] not just doing my own thing. And that is
[15:27] the tools that are about to create
[15:29] generative content on a more 24-hour
[15:32] process. And the type of content will
[15:34] become more interactive. So I'll start
[15:36] with uh showrunner. I'm not sure if
[15:38] you've ever heard of this product. It
[15:40] went a little bit viral in the last
[15:41] week, but it's actually been in the
[15:42] works for about two years. But the
[15:44] creators of South Park, uh, Matt Stone
[15:46] and Trey Parker, about two years ago saw
[15:48] this AI revolution coming. They invested
[15:51] into a group of people who created the
[15:54] South Park characters because like the
[15:56] style of animation is very chunky. It's
[15:59] very like uh, you know, it's like
[16:00] cardboard or however they make those
[16:02] kind of things. So, it plays really well
[16:04] to AI because the mistakes aren't as
[16:06] noticeable. It's not like human level
[16:08] realism. And they built a tool where
[16:11] anyone can go in there, they can say
[16:13] they can add themselves into the
[16:15] essentially South Park universe and then
[16:17] they can create they can stitch together
[16:19] storylines. They can submit it and then
[16:21] if the showrunner people, the people in
[16:23] charge of this company that are under
[16:25] Mess and Trey Parker decide it's good
[16:27] enough, they actually publish it as a
[16:29] full 30 minute video. Um, this has been
[16:32] something I've been keeping an eye on
[16:33] for a while. I've been in the Discord
[16:34] and seen how creative people are. And uh
[16:38] Netflix just recently gave them $10
[16:39] million and is looking to build them
[16:41] into the platform. And it tells me that
[16:44] what you're going to see is shows like
[16:46] South Park, but there will be dozens and
[16:48] dozens and dozens of them. They'll all
[16:50] be in the same sort of cinematic
[16:52] universe of the South Park universe. And
[16:54] when you or I go to one of their uh
[16:58] Showrunner episodes, we're not going to
[17:00] actually generate the whole thing from
[17:01] scratch. We're going to essentially be
[17:02] zooming into an entire little world
[17:05] that's going to be playing out and then
[17:08] clipping from that and uploading it to
[17:10] social media will be those viral
[17:12] moments. And there will be no way for
[17:14] other people to see them probably unless
[17:15] I don't know, maybe they're recorded or
[17:17] whatever. But this idea of showrunner
[17:19] will slowly become what I think social
[17:22] media feels like to us right now. Like
[17:25] we think of it as it's my job to upload
[17:27] this this video and get views. What it's
[17:30] about to become is I'm in charge of a
[17:33] tiny little universe and I have my
[17:36] community and this is the type of way we
[17:38] talk. This is the type of jokes we make.
[17:40] This is the type of thing we're
[17:41] interested in. If it's um a product that
[17:43] you're trying to like build the company
[17:45] around and you let your audience start
[17:48] to make little things around it. Um and
[17:50] they just they just go. It's not
[17:52] audience driven. It's like audience
[17:54] curated but it's AI driven. And um these
[17:57] models are just starting to come on. So
[17:59] kind of be aware that showrunner will
[18:01] not be the last one. It just seems to be
[18:03] one of the pioneering systems. Um and
[18:05] then to go to the other side, there's
[18:08] also this interactivity element that I
[18:10] think would be really important. So um
[18:12] fortunately I don't have any slides to
[18:13] show you this, but if you're on your
[18:15] phones or you want to write down Google
[18:17] only 4 days ago has released Genie 3.
[18:20] This I think will be not as gamechanging
[18:23] as VO, but I think this is a very
[18:25] important step towards what will be
[18:27] available from them in a higher like
[18:29] more polished form in the next four or
[18:31] five months. And Genie 3 is like VO. So
[18:35] you can create anything you want. You
[18:37] can imagine a Stormtrooper vlog. You can
[18:38] imagine those little rabbits jumping on
[18:40] trampolines. Um but it will be
[18:43] interactive. And what happens is you
[18:45] generate that image and it it's given to
[18:48] somebody else to view, but they have
[18:51] control over zooming in, walking around,
[18:54] turning around. And Genie will on its
[18:57] own start to generate what's behind it.
[18:59] So if you generate an image right now,
[19:01] there's no turning around to see what's
[19:03] behind it. But Genie is working on this
[19:05] and it's already it's they've got demos.
[19:06] Google had demos uh just four days ago
[19:08] that came out and it's persistent. So uh
[19:11] don't know how much you guys all know
[19:12] about this but one of the problems with
[19:14] these diffusion models meaning AI is
[19:16] starting from a seed and it's kind of
[19:18] imagining layer by layer all the way up
[19:20] to some final product is that when you
[19:22] turn what was there would normally
[19:24] disappear but the engineers at Google
[19:26] have figured out ways to have persistent
[19:28] memory. Um it's not perfect but it will
[19:31] get probably pretty close to perfect in
[19:32] the near term. I would say a year at
[19:34] least or so. And um yeah, that also
[19:38] means that when you create a piece of
[19:40] content, you don't want them to just
[19:41] watch it and swipe past. You want them
[19:43] to watch it and keep moving around in
[19:46] that environment. Like if you saw the
[19:49] rabbit jumping on the trampoline, you
[19:50] might look around like, "Is anyone
[19:51] watching these rabbits? Is somebody like
[19:54] is something else going to happen?" And
[19:56] part of that is users adding more
[19:58] content and prompting it. And part of
[19:59] that is people walking around and
[20:00] exploring it. And it just feels to me
[20:03] like it's going to have that same
[20:04] excitement, that dopamine kick of
[20:06] swiping, but it will be in a different
[20:08] persistent format. So, I don't know. I
[20:11] would just once again, I'm kind of going
[20:12] a little bit into the future here, and
[20:13] if you want to stay more practical with
[20:15] the tools, we can, but both of these
[20:16] tools are out now, so you can see them
[20:18] enough to uh kind of get your head
[20:20] around them. Um, now I'll jump to
[20:22] another topic, which is something I've
[20:24] been dealing with recently, is uh
[20:26] creating for non-human audiences.
[20:29] there. Right now, I know it seems like
[20:31] we create content for other humans, but
[20:33] we're all kind of deep down subtly, you
[20:36] know, worried some of those views aren't
[20:38] real. Is that bots? It a lot of
[20:40] comments. You're kind of wondering like,
[20:42] is that from the United States? Is that
[20:44] from some kind of LLM? And that is
[20:47] probably going to get more and more of a
[20:50] problem. And it's a problem if all you
[20:52] care about is talking to humans, which
[20:54] you probably do if you're talking about
[20:56] a human who needs to open up their
[20:57] wallet and buy something for you. It
[20:59] probably isn't if what you want is just
[21:02] growth and the way that you make money
[21:03] is from advertising on social media. So,
[21:06] I don't know where everybody sits on
[21:07] this. It's a little bit of a different
[21:08] strategy depending on who you are, but
[21:10] if what you want to be is a social media
[21:12] content creator, it's probably not that
[21:14] bad. It's means AI generated ads are
[21:16] sitting in there. Somebody's paying for
[21:18] those. You're getting a share of it. And
[21:20] if you have, you know, AI generated
[21:22] content, it doesn't matter. But if it's
[21:23] trying to get people to follow a link
[21:24] and you want a real human connection,
[21:26] it's about a funnel, then you're going
[21:28] to want to play to that strength and be
[21:30] very careful about curating your social
[21:32] media strategy. So you so it so it is
[21:34] human. Like don't make it half human.
[21:36] Don't generate too much B-roll because
[21:38] people will start to distrust you. have
[21:39] to sort of stay strong on that that
[21:41] thing that I'm a real person sharing
[21:43] real thoughts and um you know maybe even
[21:46] market that way and like mention that
[21:47] once in a while to your audience so they
[21:49] know that it is a real you do real life
[21:50] things like this um but there's also no
[21:54] more Google really like everybody's
[21:56] going to LLM I actually had some stats
[21:58] before about how many people have been
[22:00] driven to people's websites now through
[22:02] generative um engine optimization which
[22:05] is like the new buzz term for SEO but
[22:07] what that means is that LLMs like
[22:10] imagine chatgbt is taking and digesting
[22:14] the web and it's doing it closer and
[22:16] closer to real time. We saw with GPT5
[22:17] there's a new looping feature and what
[22:20] they're doing is trying to find out what
[22:23] the truth is. Get reinforcement learning
[22:26] for that signal and then give you a
[22:29] prompt result. So if you're not on
[22:32] Wikipedia, you probably should really
[22:34] put some effort into that. It sounds
[22:36] weird because Wikipedia normally
[22:37] wouldn't be what you would think would
[22:39] drive a search result through Google,
[22:41] but it's a very trusted source when um
[22:45] it just it's just got so much c that's
[22:46] what makes it hard. Like technically
[22:48] you're not supposed to buy your way into
[22:49] Wikipedia. It has to be somebody
[22:50] volunteering their time. It's a very,
[22:52] you know, slow process to get in there.
[22:54] But when you're in Wikipedia, your
[22:56] company, your brand, you as a person, if
[22:58] you're a celebrity of any sort, um that
[23:00] means that the LLM will start answering
[23:04] people much more. So, for example, my
[23:07] fiance is a magician here in Las Vegas.
[23:09] She's um over at the Westgate and she
[23:11] performs magic and we started noticing
[23:14] some of the traffic from her site was
[23:15] coming through LLM, but not that much.
[23:17] So, we kind of really started thinking
[23:19] about Wikipedia and trying to, you know,
[23:22] contact people there and say, "Hey, like
[23:23] here's some credible links. Do you mind
[23:25] adding her as a uh Wikipedia page?" And
[23:28] it was like night and day. As soon as
[23:29] the Wikipedia page showed up, um the
[23:32] generative engine optimization said,
[23:34] "Oh, this isn't just some random
[23:35] websites about a magician. This is like
[23:37] a trusted source." So when you ask who
[23:40] are like the top female magicians in the
[23:41] world or the best magicians in Las
[23:42] Vegas, like she started showing up and
[23:44] these were very similar queries to the
[23:46] ones I was running just seven or eight
[23:48] months ago before the Wikipedia thing.
[23:49] So at least you know it's an end of one
[23:52] so I don't know how much to trust it but
[23:53] to me it makes sense. It makes sense
[23:54] that credible sources. So, and that
[23:56] includes like New York Times or if you
[23:58] can get in any of the major uh
[24:01] publications like Google definitely has
[24:03] a priority for them in their searching
[24:05] and LLMs have a priority in in the trust
[24:08] that they're putting in when and that
[24:10] means when they digest that information
[24:12] they're assigning somewhat of a stronger
[24:14] probability to this being real meaning
[24:16] the weights are you know the neural
[24:17] network is activated a little bit
[24:19] stronger and you just you know end up in
[24:21] more prompts so um yeah when it comes to
[24:25] social media I think you have to start
[24:26] thinking about generative engine
[24:27] optimization on top of it. I know social
[24:29] media doesn't normally seem like it
[24:30] would include Wikipedia, but but
[24:33] credible sources, getting them to talk
[24:35] about your um social media is going to
[24:36] be kind of important. Um and then yeah,
[24:40] and then for the last few minutes, I
[24:41] guess I'll just kind of talk about uh
[24:44] what I would call like the loneliness
[24:46] behind the spotlight. I think I think
[24:49] what's also something you have to kind
[24:51] of be aware of is that um as AI replaces
[24:54] social media there's going to be a shift
[24:56] in tone where everybody feels famous and
[24:59] nobody is kind of like I think everyone
[25:01] will be like wow like I got all these
[25:03] views and but then the ad revenue will
[25:06] kind of level out because everybody's
[25:07] getting more views and you won't be
[25:09] really sure who's um real and who's not
[25:12] and there's going to be a bunch of slop
[25:14] right now like we're about to go through
[25:16] this era
[25:17] I'm sure YouTube and Instagram are going
[25:19] to get just bombarded with VO3 content,
[25:21] the same thing I'm recommending that you
[25:23] all try. And they don't have a great
[25:26] system in place yet for filtering that
[25:28] out, knowing what's good, what's bad.
[25:30] Um, they're working on it and there will
[25:33] be many things that they can do in the
[25:35] machine learning side to see what
[25:36] quality looks like uh in terms of
[25:39] engagement, in terms of actual humans,
[25:40] in terms of like taking feedback and and
[25:42] assigning it higher probabilities. But
[25:45] for a while, I think it'll be very
[25:46] sloppy. But that also means it's like
[25:48] the gold rush. Like this is the time to
[25:50] create a brand to use tools like VO to
[25:54] um try to break out of the slop. And
[25:57] just like during the revolution, there
[25:59] was some Amazon's in there and there
[26:00] were some pets.com. And you just have to
[26:03] stay strong to whatever the vision is
[26:05] that you have for it. So yeah, that's
[26:07] all I kind of have and then I have a few
[26:08] more minutes. Is there any questions you
[26:10] guys wanted to throw at
