VibeVO – Advertisers and creators in one ecosystem

Blog

  • AI Voice: How to Create the Perfect Text-to-Speech in 2026

    AI Voice: How to Create the Perfect Text-to-Speech in 2026

    In 2026, speech synthesis technologies reached an incredible level, transforming the complex process of audio content creation into a matter of minutes. Now, it’s enough to insert text, choose a model and voice, and your file is ready. However, practice shows that the first result is often far from ideal. The problem is not in the quality of the neural network, but in the nuances of text preparation that most users ignore. Let’s figure out how to avoid mistakes, choose the right model, and scale AI production to create the perfect text voiceover.

    Text Preparation: The Secret to a Perfect AI Voice

    A neural network reads symbols, not meanings that a human constructs from context. To avoid errors, the text must be adapted.

    Dealing with Homographs and Stresses

    The Russian language is rich in words that are spelled the same but have different meanings depending on the stress (e.g., “за́мок” (castle) and “замо́к” (lock)). The model chooses a variant based on statistics, which often leads to errors. There are two solutions:

    • Rephrasing: Change the sentence to eliminate ambiguity. For example, “Замок заело” (The lock got stuck) can be rephrased as “Дверной замок заело” (The door lock got stuck).
    • Placing stresses: Use special characters. In Yandex, this is a “+” before the stressed vowel (зам+ок), in most other services – Unicode U+0301 immediately after the stressed vowel (замо́к). This method is more reliable but makes the text less readable.

    Numbers, Abbreviations, and Latin Script

    • Numbers: For correct pronunciation of numbers, especially with cases and units of measurement, it is better to write them out in words. For example, “к 15 марта” (by March 15) will become “к пятнадцатому марта” (by the fifteenth of March), and “1 250 000 ₽” will become “один миллион двести пятьдесят тысяч рублей” (one million two hundred fifty thousand rubles).
    • Abbreviations: Models handle abbreviations read letter by letter (НДС, МФЦ) or as words (ГОСТ, вуз) well. However, mixed constructions, like “ГОСТ Р 34.10-2012”, are often pronounced character by character. In such cases, it is better to write them out in words or move them off-screen.
    • Latin script: English words and abbreviations (API, OK) within Russian text may be read incorrectly. The solution is transliteration: “эй-пи-ай вернул двести о-кей” (API returned two hundred OK).
    • Letter “ё”: The absence of “ё” can change the meaning of a word (“все” (all) and “всё” (everything)). In literary texts and proper names, place “ё” manually.

    Choosing a Neural Network for Voiceover: The Effectiveness of AI vs. Real UGC

    The choice of model depends on the task. Modern neural networks offer a wide range of functionalities, from basic narrator voice to emotional rendering.

    OpenAI TTS: A Reliable Workhorse

    The tts-1 and tts-1-hd models offer a smooth, narrator-like reading. They are ideal for educational videos, audio versions of articles, and interface prompts. The request limit is 4,096 characters. These models are available in BotHub: 1,767.86 ₽ and 4,278.21 ₽ per million tokens, respectively.

    ElevenLabs: Leader in Expressiveness

    ElevenLabs v3 offers maximum expressiveness in Russian, with support for audio tags, pauses, and tempo changes. Multilingual v2 is more predictable for long texts. Flash v2.5 and Turbo v2.5 are fast, inexpensive, and process up to 40,000 characters at a time. However, the more expressive the model, the smaller the context (v3 has only 5,000 characters), which requires dividing the text into smaller fragments. All these models are also available in BotHub.

    Google Text-to-Speech: For Large-Scale Projects

    Google offers classic voices (Standard, WaveNet, Neural2, Chirp 3 HD) with per-character billing and Gemini TTS with per-token billing and a “Style instructions” feature for delivery control. Over 380 voices in 75+ languages are available. This service is suitable for those who already have projects in Google Cloud and need a large volume. The entry barrier is high: an account and a foreign card are required.

    Yandex SpeechKit: For the Russian Market

    Integrated into Yandex AI Studio since 2026. The cost starts from 1,342 ₽ per million characters. All calculations are made in rubles and include VAT, which is convenient for Russian legal entities. Ideal for voice robots, answering machines, and IVR in the Russian-speaking segment. The Brand Voice service is available for creating a unique company voice.

    Local Solutions: Confidentiality and Scalability

    For projects with high confidentiality requirements or a zero budget for volume, synthesis can be run locally. Examples include Kokoro 82M (without Russian language) and Fish Audio S2 Pro (80 languages, commercial use is paid). This approach is suitable for closed circuits, working with personal data, and large volumes if you are willing to manage the infrastructure.

    Testing and Optimization: Deepfake-Ethical and Efficient

    Voice selection is not just about timbre. Run your own paragraph through several voices to understand how they handle your content.

    AI-голос: как создать идеальную озвучку текста в 2026 году — illustration 2

    Voice Cloning: Ethical Aspects

    The Voice Clone feature in ElevenLabs allows you to create a copy of a voice from a short recording. Technically, it’s simple, but legally complex: cloning someone else’s voice without written consent is prohibited. It is important to adhere to deepfake ethics.

    Iteration Process: From First Pass to Final Render

    1. First Pass: Record a short text fragment (paragraph). Listen carefully to how numbers, names, abbreviations, and sentence boundaries are pronounced. Correct errors in the text, not in the settings.
    2. Second Pass: Record the full text. Evaluate the transitions between fragments and the overall pace. Models with a larger context reduce the number of transitions but may be less expressive.
    3. Final Render: Choose the desired format. WAV or FLAC for further editing without loss of quality, MP3 or AAC for podcasts.

    “Scaling was impossible — now it’s possible. Subscription-based neuro-production opens new horizons for content creators.”

    Comparative Analysis: Cost of 1 Minute of AI Video and Tariffication Nuances

    To evaluate the effectiveness of AI versus real UGC and understand how much one AI creative costs, it’s important to understand the tariffs.

    Test Run: 289 Characters

    Stress-test text: “On September 8, 2026, OOO “Yolochka” signed a contract for 1,250,000 ₽ with VAT. The lock in the warehouse jammed, and a 16th-century lock has nothing to do with it. Deadlines have already been missed, and this is an expensive lesson. According to GOST R 34.10-2012, the signature is correct, the API returned 200 OK, and the ElevenLabs v3 model read it without a hitch. Or not?”

    • OpenAI tts-1: 23 seconds of audio, 0.76851 ₽ charged (45,200 characters per hour, 120 ₽/hour).
    • OpenAI tts-1-hd: 23 seconds of audio, 1.85998 ₽ charged (45,200 characters per hour, 291 ₽/hour).
    • ElevenLabs v3: 34 seconds of audio, 7.68625 ₽ charged (30,600 characters per hour, 814 ₽/hour).

    Conclusions on Pricing

    • Duration: ElevenLabs v3 reads slower, which increases the audio duration and, consequently, the cost when billed by time.
    • Tokens vs. Characters: Token-based billing in Russian differs significantly from character-based billing. 289 characters can turn into 435 tokens, increasing the actual cost. Always check the actual charges in the interface.

    Frequently Asked Questions

    How to choose the best neural network for video generation?

    The choice depends on the specific task: for a narrator’s voice, OpenAI TTS is suitable; for emotional voiceovers, ElevenLabs; for large volumes within the Russian context, Yandex SpeechKit. Consider character limits, cost, and the ability to control intonations.

    Can AI voice be used in commercial projects?

    Yes, most services offer commercial licenses. However, when cloning a voice, always obtain written consent from the copyright holder to avoid legal issues.

    How much does mass video generation by a neural network cost?

    The cost depends on the chosen model, the volume of text, and the duration of the final audio. For example, for a 10-hour audiobook, the price can range from 1,200 ₽ (tts-1) to 8,100 ₽ (ElevenLabs v3). For AI UGC for a crypto project or an AI video factory for e-commerce, it is important to consider the scale and optimize the process.

    Where to order AI content production?

    Services like BotHub offer access to various speech synthesis models, including OpenAI and ElevenLabs, with payment in rubles and trial access. This is a convenient solution for getting started with AI avatars with subtitles, AI voiceovers, and AI clips with face swapping.

    Conclusion: The Future of AI Content is Here

    AI video generation and audio are not just a trend, but a new reality. With the development of technology, we gain access to tools that allow us to create mass video generation by neural networks and an AI conveyor of 500 videos per day. The key to success is understanding the nuances of text preparation and choosing the right tools. Start experimenting with BotHub today to evaluate the capabilities of neuro-production and scale your content!

    Prices as of September 8, 2026:

    Model or Service Characters per request Free Payment in Rubles (approximate)
    tts-1 in BotHub 4 096 trial access 1 767.86 ₽ per 1 million tokens (2.66 ₽ per 1000 characters)
    tts-1-hd in BotHub 4 096 trial access 4 278.21 ₽ per 1 million tokens (6.43 ₽ per 1000 characters)
    Eleven v3 in BotHub 5 000 trial access 26.60 ₽ per 1 000 characters (by measurement)
    Eleven Multilingual v2 in BotHub 10 000 trial access 17.53 ₽ per 1 million tokens
    Eleven Flash v2.5, Turbo v2.5 in BotHub 40 000 trial access 8.76 ₽ per 1 million tokens
    OpenAI directly 4 096 no $15 per 1 million characters, HD $30
    ElevenLabs directly depends on model 10,000 credits/month (without commercial license) from $6 per month
    Google Standard, WaveNet depends on model up to 4 million characters per month $4 per 1 million characters
    Google Chirp 3 HD depends on model up to 1 million characters per month $30 per 1 million characters
    Google Gemini 3.1 Flash TTS depends on model 32,000 tokens $1 per 1 million text tokens + $20 per 1 million audio tokens
    Yandex SpeechKit, API v1 not specified according to AI Studio terms 1 342 ₽ per 1 million characters including VAT
    Yandex SpeechKit, API v3 not specified according to AI Studio terms 0.1626 ₽ per request (long requests are counted proportionally)
  • Is CPM profitable? Should Perplexity be excluded from AI visibility tracking?

    Is CPM profitable? Should Perplexity be excluded from AI visibility tracking?

    In the world of AI search, leaders are constantly changing, and the question of which platforms to track becomes critical for SEO specialists. Recent discussions have revolved around the Perplexity platform: should we continue monitoring it or focus exclusively on giants like ChatGPT, Gemini, and Claude? We will examine how changes in market share affect visibility assessment and how much money each spent ruble will bring to understand if CPM is profitable.

    Is Perplexity really losing ground?

    Ross Hudgens, CEO of Siege Media, claims that Perplexity’s market share is shrinking. Including it in general metrics alongside ChatGPT, Gemini, Claude, and Google AI products can distort the real picture of visibility. Hudgens urges: “Everyone should remove Perplexity from their LLM trackers today.”

    However, as experience shows, hasty conclusions can be erroneous. In March 2002, at the Search Engine Strategies conference, the reduction of tracked search engines from 15 to 5 was discussed. Google was then excluded from the list. This turned out to be a serious oversight, as Google was rapidly gaining momentum. A similar situation could repeat itself today.

    Data confirms Perplexity’s decline

    • According to StatCounter, in June 2026, Perplexity held 7.91% of AI chatbot referral share, almost on par with Gemini (7.94%).
    • By August 2026, Perplexity’s share had fallen to 4.31%, while Gemini grew to 10.9%.

    “A small market share does not always mean small strategic importance. It’s like investing in stocks: you shouldn’t always look only at current capitalization, but also at growth potential.”

    The market is consolidating, but not around a single leader

    The latest Similarweb data for May 2026 shows the dominance of ChatGPT (53.9% of global web visits among seven major AI assistants). Gemini accounts for 27.9%, Claude – 9.2%, DeepSeek – 4.1%, Grok – 2.4%, and Perplexity and Copilot – 1.3% each.

    Significant changes occurred not in the disappearance of Perplexity, but in the emergence of a strong second player. ChatGPT’s share of AI chatbot web traffic decreased from 76.4% a year ago to 52.7% in May 2026. Gemini grew from 9% to 27.3%, and Claude – from 1.6% to 8.9%.

    Key players and their advantages

    • ChatGPT: over 900 million weekly active users (February), over 1 billion active users across all products (end of July). This is a huge direct consumer base.
  • Gemini: The Gemini app exceeded 1 billion monthly users (August). It has Google distribution through Search, Android, and the entire product ecosystem.
  • Claude: Has gained a strong position in the enterprise segment and among developers. In April, over 100,000 customers used Claude on Amazon Bedrock. Annual revenue exceeded $65 billion (August).

These are not three versions of the same business, but three different distribution advantages. The market looks like an emerging oligopoly, not a “winner takes all” situation.

What SEOs Should Track

Google AI Overviews and AI Mode should not be viewed as ordinary LLMs. They are much more important. In June, Google reported that AI Overviews cover more than 2.5 billion users per month, and AI Mode exceeded 1 billion monthly users. AI Mode queries have doubled every quarter since launch.

According to Similarweb, AI Overviews appeared in 43% of Google search queries in the US by May 2026, compared to 15% a year earlier. AI Mode visits grew from 126 million in June 2025 to 279 million in May 2026.

This means that Google AI is a separate layer in the world’s dominant search ecosystem. For B2B companies, Claude is also of immense importance, despite relatively low consumer traffic. PwC, TCS, and Cognizant are actively integrating Claude into their workflows, training tens of thousands of employees. Over 1,000 business customers spend more than $1 million annually on Claude.

Three-Tier AI Visibility Measurement System

  1. First Tier: Platforms with scale and strategic importance. Track ChatGPT and Gemini separately. Add Claude for B2B, corporate, and professional audiences.
  2. Second Tier: AI search embedded in existing ecosystems. Track Google AI Overviews and AI Mode separately from ChatGPT, Gemini, and Claude. The goal is to measure how AI changes the search journey. Microsoft Copilot can also be included in this tier if the company has a significant presence in Microsoft 365.
  3. Third Tier: Emerging or specialized platforms. This includes Perplexity, Grok, DeepSeek. Do not ignore them, but do not give them equal weight. Monitor them for unusual visibility, referral traffic, citations, and growth.

Don’t Let Averages Mask Real Data

An aggregated LLM visibility metric can create a false sense of accuracy. For example, if a brand has 40% citation rate in ChatGPT, 35% in Gemini, 30% in Claude, and 90% in Perplexity, a simple average would be 48.75%. This number says little if Perplexity accounts for only a small fraction of the traffic important to the business.

CPM выгодный? Стоит ли исключать Perplexity из отслеживания ИИ-видимости? — illustration 2

Instead of asking, “What is our average LLM visibility?” you should ask, “Where do the people important to this business actually encounter our brand?”

This requires combining three data sets:

  • Audience Reach: usage, visits, platform distribution.
  • Visibility: mentions, citations, referring URLs, queries that generate them.
  • Business Impact: linking AI referrals to engagement, leads, sales, and other conversions.

My Verdict on Perplexity

Ross Hudgens is right about the problem, but not entirely right about the solution. Don’t remove Perplexity from your LLM tracker, but reduce its weight. If Perplexity generates 1% of your AI traffic, giving it 25% or 33% weight in your visibility score is unjustified. However, if it accounts for more than 5% of referrals, as Hudgens himself suggests, the argument for it becomes stronger.

Perplexity’s trajectory is worth watching. Similarweb’s August 2026 analysis still describes Perplexity as an active player in the AI search market, noting its ad-free strategy and focus on subscriptions and enterprise deals. This shows why “small” and “insignificant” are not synonyms.

Frequently Asked Questions

What CPM should be considered profitable for AI platforms?

Profitable CPM in AI platforms depends on your target audience and ROI. If a platform generates high-quality leads or sales, even with a relatively high CPM, it can be profitable. The main thing is to consider not only the cost per impression but also the conversion rate.

Is it worth testing hypotheses cheaply on new AI platforms?

Yes, cheap hypothesis testing on new AI platforms is essential. This is how you can discover the next Google. You shouldn’t spend huge budgets on them, but monitoring and minimal testing will allow you not to miss a potential breakthrough.

What is cheaper: a banner or a blogger, when compared to AI platforms?

The question of what is cheaper, a banner or a blogger, is incorrect in the context of AI. AI platforms offer a different type of interaction. Price comparison should be done through efficiency metrics: how much does 1 million impressions via a banner cost versus the cost of generating 100 AI videos or the cost of publishing 500 videos via AI tools. Each channel has its own unit economics.

How to estimate the cost of uniqueizing one video using AI?

The cost of uniqueizing one video using AI depends on the complexity of the task and the tool used. It is important to evaluate not only direct costs but also the time that AI will save. The price per minute of AI video with an avatar also varies, but it often turns out to be more profitable than manual labor, especially for large volumes.

Conclusion

The AI search market is dynamic. Ignoring Perplexity or other niche platforms entirely means risking missing important signals. However, blindly giving them equal weight with giants distorts the picture. A mathematical approach to analysis, weighing platforms by their real business impact, and constant monitoring of new players are key to success in SEO and a profitable CPM strategy.

We count cash, not just the number of platforms in the tracker. Don’t overlook the outsiders who are gaining momentum. Perhaps among them lies the next market leader. Start analyzing your data today so you don’t miss new opportunities!

  • Molson Coors: How to Speed Up Marketing and Quadruple Engagement with Content Creators

    Molson Coors: How to Speed Up Marketing and Quadruple Engagement with Content Creators

    Major beverage producer Molson Coors is radically changing its approach to interacting with content creators, abandoning outdated approval processes characteristic of the television era. The company is now focusing on speed and flexibility, which has led to a fourfold increase in audience engagement. This decision is particularly relevant for companies seeking to optimize their marketing budgets and maximize returns on investment in content marketing.

    Molson Coors’ new approach: “freedom within boundaries” for creativity

    Since March, Molson Coors, in partnership with The Shake Squad consulting group from Movers+Shakers agency, has begun implementing an updated strategy. Justin Stauffer, Senior Director of Creative Effectiveness at Molson Coors, did not disclose commercial figures but noted that the new system has been scaled to 230 marketers working with over 100 brands in the US and Canada, including Miller High Life, Fever Tree, and Zoa energy drinks.

    “We’re moving much faster than we were before because our legal team has been involved in the process from the beginning, they better understand the ecosystem and consumer behavior in this space,” says Stauffer.

    The problem of outdated thinking and its solution

    Evan Horowitz, CEO of Movers+Shakers, emphasizes that Molson Coors’ main problem is typical for many large companies: senior marketers grew up in a “television” world. This mindset was passed down to junior specialists.

    “The world has changed dramatically,” Horowitz notes. “The ecosystem has become much more complex, and brands that still use a television strategy have a broadcast mentality. They talk to customers without understanding that there are hundreds of parallel dialogues. This requires a fundamentally different approach to brand building.”

    To address this problem, Molson Coors introduced a “freedom within boundaries” system with the legal department. It classifies potential solutions into three categories:

    • Fast track: for quick approvals.
    • Requires discussion: for issues needing additional analysis.
    • Categorical refusal: for unacceptable proposals.

    Less is more: abandoning “TV campaigns” on social media

    The second key element of Molson Coors’ strategy is reducing the importance of individual publications and abandoning the treatment of organic social content as a “high-budget TV campaign.”

    Influencers vs. creators: what’s the difference?

    Horowitz notes: “There is no better platform, no better creator; there is only the question: ‘what audience is this brand trying to reach?’”. Stauffer and Horowitz distinguish between influencers and content creators:

    • Influencers: community and reach-oriented.
    • Content creators: focused on craft and creativity.

    This helps in drafting briefs: creators can receive more open-ended assignments, which requires some adaptation from a large company like Molson Coors.

    Measuring effectiveness and trusting new approaches

    For large companies, one of the challenges is the “measurement gap” in creator performance. “Marketing budgets don’t change as fast as consumer attention. Many large companies underfund organic social content and creator work because they can’t prove ROI, while their competitors allocate funds on faith and see results,” comments Horowitz.

    Experimental approach to content

    Horowitz’s team helped Molson Coors rethink its approach to measurement and content itself. “It has to be experimental. It has to be a place where we can test, learn, observe signals, learn about our communities, and build our brands from fans for the feed,” says Stauffer.

    The Shake Squad developed a training system for the entire organization, tailored to each brand and their “consumer cohorts.” This allowed Molson Coors to understand how each brand is represented in these groups’ social feeds. “We could see what was in the feed every day for a core Miller Lite consumer,” Stauffer adds.

    This culture-first approach helped shape briefs and embrace the power of “lo-fi” content, which feels more authentic and less polished.

    “We really had to redefine content quality in this space differently than we were used to doing for other channels,” Stauffer emphasizes.

    Molson Coors: how to accelerate marketing and quadruple engagement with content creators — illustration 2

    Trust as the foundation of success

    The last, and perhaps most crucial, element of a successful new social strategy for a company worth over $7 billion is trust. Trust in The Shake Squad’s advice and belief that a company that has existed for hundreds of years (Molson and Coors merged in 2005) can radically change its marketing strategy, becoming more agile and responsive.

    Molson Coors’ marketing team worked closely with The Shake Squad for several months, applying their research, advice, and expertise to develop a new approach. Horowitz noted that marketing executives approached the collaboration with a humility not common to all large companies, which made the process smoother and more experimental.

    According to Stauffer, this humility led to a complete overhaul: fewer restrictions, less hesitation, more focus on culture-centric content.

    Perhaps the future of Molson Coors’ marketing strategy will be more like e.l.f. — a cosmetics brand that Movers+Shakers helped grow from $220 million to $1.5 billion and become the number one brand among Alpha, Gen Z, and millennial generations in seven years. This “social-first” brand does not have a traditional marketing agency and only occasionally uses TV ads to create buzz.

    “They completely flipped the script,” Horowitz concludes.

    Frequently Asked Questions

    Why did Molson Coors change its marketing strategy?

    Molson Coors sought to adapt to the rapidly changing digital landscape, where traditional “TV-first” content alignment methods were outdated. The goal was to accelerate processes, increase flexibility, and boost audience engagement through collaboration with content creators.

    What is the “freedom within a framework” system?

    This is a new content alignment system with Molson Coors’ legal department that categorizes decisions into three categories: fast track, requires discussion, and outright rejection. It significantly speeds up approval processes and increases interaction efficiency.

    What is the difference between an influencer and a content creator?

    According to Molson Coors and Movers+Shakers, influencers are focused on reach and community, while content creators are focused on producing high-quality and creative material. This approach helps to create more precise briefs for different types of collaborations.

    How does Molson Coors measure the effectiveness of the new strategy?

    The company is moving towards an experimental approach where content is seen as an opportunity for testing and learning. The emphasis is on observing signals from communities and building brands “from fans for the feed,” which allows for a better understanding of the audience and the creation of culture-centric content.

    What benefits does a company gain by ordering “turnkey” video production?

    Ordering full-cycle video production allows a business to get all necessary services in one place, which significantly saves time and hassle. This eliminates the need to coordinate multiple contractors, reduces risks, and ensures a unified vision for the project from idea to implementation, which is critical for obtaining high-quality and effective content, as Molson Coors’ experience has shown.

    Conclusion

    Molson Coors’ experience demonstrates that for successful marketing in today’s digital world, it is necessary to abandon outdated approaches and trust new strategies. A comprehensive approach to content production, which includes not only creation but also strategic planning and adaptation to platform specifics, is key to increasing engagement and effectiveness.

    If you strive for the same results and want to optimize your marketing efforts, consider ordering full-cycle video production, which will allow your company to move at the speed of creators and reach new heights.

  • AI videos for advertising: Grom TV – Russia’s first generative online TV

    AI videos for advertising: Grom TV – Russia’s first generative online TV

    Mass video generation by neural network is now available in the format of endless online television. GPTunneL launches Grom TV – an innovative platform where anyone can launch their own AI-generated video in just 20-25 seconds. This revolutionizes the concept of creating AI video for advertising and opens new horizons for AI content production.

    Imagine you are the director of your own channel, where your prompt turns into a video in real time. The effectiveness of AI versus real UGC in this format reaches new heights, demonstrating how quickly and extensively content can be generated.

    What is Grom TV and how does it work?

    Grom TV is generative online television where users enter text queries (prompts), and a neural network generates videos based on these queries. Your creations are broadcast live along with clips from other users. This is not just video generation from text; it’s a whole interactive experience.

    Your neuro-production by subscription in real time

    • Visit the website: gptunnel.ru/lab?section=tv
    • Register: The process is as simple as possible.
    • Write a prompt: Enter your query in the chat to the right of the screen.
    • Wait 20-25 seconds: This is the time from entering the prompt until the video appears on air. The generation itself takes about 7.8 seconds.
    • Watch: Your creation is broadcast live.

    For example, you can create a Gandalf and Scooby-Doo crossover, a film noir scene, or a KISS concert with “The Big Bang Theory” characters. The neural network confidently works with different styles, whether it’s Disney plasticity or black-and-white noir, and knows what Homer Simpson looks like without additional explanations.

    Behind the Scenes Technology: How does the neuro-pipeline work?

    Normal video generation is an asynchronous process. Grom TV, however, works live, which requires a completely different architecture. Every millisecond is crucial here so that the broadcast doesn’t “stutter.”

    Secrets of high speed and stability

    One Grom TV video lasts eight seconds with sound, including lip-sync if there is speech. Its generation takes about 7.8 seconds. This means a real-time coefficient of 0.97 — generation is ahead of the broadcast, albeit with a small margin. The entire process runs on one Nvidia B200 chip; sometimes a second one is connected to optimize speed.

    «Scaling was impossible — now it is. A single Nvidia B200 chip allows us to generate about 10,000 unique clips per day without a single repeat. This is the very AI video factory for e-commerce and beyond,» notes the founder of GPTunneL.

    Next, ffmpeg comes into play, which in 0.3 seconds manages to:

    • Adjust playback tempo to generation speed.
    • Apply a «film» style: grain, vignette, soft color.
    • Cut the video into two-second segments for smooth transitions.

    The use of the HLS protocol ensures compatibility with all devices: browsers, VLC, smart TVs. This guarantees that AI avatar with subtitles or AI clips with face replacement will be accessible to the widest possible audience.

    Infinite Content: When the Queue is Empty?

    Grom TV does not stand idle. If the queue of user prompts becomes empty, Grom GPT – GPTunneL’s own LLM model – steps in. It generates plots and feeds them into the broadcast, ensuring continuous transmission. This is like a neural network writing scripts and editing automatically.

    The ability to generate 10,000 unique clips per day without repeats demonstrates unprecedented efficiency. This is much more than a team of 500 people could create, making the AI conveyor of 500 videos per day a reality, not a dream.

    Safety and Moderation: Ethical Deepfake

    Live broadcast does not allow to «rewind», so the moderation system is built to filter unacceptable content before video generation. This ensures an ethical deepfake approach.

    Two Lines of Defense

    1. GromMod: The first filter from GPTunneL. Analyzes the entire prompt and rejects unacceptable content (e.g., violence) even before it is sent to the GPU. Rejection is instantaneous.
    2. Grom at the script stage: The second line of defense. Transforms the user’s request into a safe script for the video model, cleaning out sensitive details and softening formulations, but preserving the creative intent.

    This two-level system ensures that only safe content reaches the model. This is critically important for creating AI UGC for a crypto project or any other niche where reputation plays a key role.

    AI video for advertising: Grom TV – Russia's first generative online TV — illustration 2

    GPTunneL: A Platform for the Future

    Grom TV is a laboratory experiment of the Russian AI platform GPTunneL, which combines over a hundred AI tools for generating text, images, video, music, and voice. The project, which started in 2023, grew into a platform with over one and a half million users by 2026.

    This is a vivid example of what happens when a platform has its own models (video, LLM-moderator, LLM-scriptwriter) and its own hardware. AI video for advertising, mass video generation by neural network, creation of AI commercials – all this becomes accessible and scalable.

    Technically, it is possible to raise the generation quality to full-fledged streams with digital avatars, stable picture, and HD resolution. If a business is interested in a live generative video channel, such a solution can be implemented on a B200 server chip in real time.

    Conclusion: The Future of AI Production is Already Here

    Grom TV is not just an experiment; it’s a demonstration of the principle: generative video can be broadcast live on a single graphics card, and it will be a full-fledged, working channel. From AI UGC for a crypto project to a neural network for UGC advertising – the possibilities are endless. Where to order for crypto? Perhaps this is where the answer begins.

    We continue to refine the pipeline, improve speech quality, splice stability, and moderation. To see this for yourself, visit Grom TV, write your prompt, and become part of this technological revolution. It’s like an analog of Synthesia, but live and with the ability to generate 100 videos per day!

    Frequently Asked Questions

    How much does one AI creative cost in Grom TV?

    As part of the Grom TV experiment, video generation is available for free. The cost of 1 minute of AI video in commercial solutions will depend on the complexity of the request and the volume.

    Can Grom TV be used to create AI videos for advertising?

    Yes, Grom TV demonstrates the potential for mass video generation by neural networks, which is ideal for creating AI commercials. In the future, specialized B2B solutions may be developed.

    What is a real-time coefficient of 0.97?

    This means that the video generation time (7.8 seconds) is slightly faster than the clip’s duration (8 seconds), ensuring stable live streaming without delays.

    How does GPTunneL ensure content moderation?

    GPTunneL uses a two-level moderation system: GromMod filters out inappropriate prompts before generation, and Grom softens formulations at the script stage while preserving the creative intent.

    Where can I order mass video generation for crypto?

    Grom TV is currently a demonstration platform. For commercial inquiries regarding AI UGC for a crypto project and mass video generation, it is recommended to contact GPTunneL directly to discuss B2B solutions.

  • ChatGPT Ads: 6 Months Later — What is a Good CPM?

    ChatGPT Ads: 6 Months Later — What is a Good CPM?

    Six months after the launch of OpenAI’s advertising platform, ChatGPT Ads still raises questions among advertisers. The figures vary widely: cost per click (CPC) ranges from less than $3 to $13, and a profitable CPM is not always clear. Some campaigns bring quality leads at a competitive price, while others yield minimal results. The problem is that there are no clear criteria for evaluating effectiveness, and the platform’s reporting is limited.

    Reporting is available at the campaign, ad group, and ad levels. Conversion tracking is implemented through OpenAI Pixel and Conversions API, supporting actions such as purchases, leads, and registrations. OpenAI can also use modeled conversions. This is where the reporting functionality ends, which complicates price comparison and the assessment of real ROI.

    Reporting Limitations and Competition

    The lack of comprehensive competitive reporting is a critical issue. If CPC increases, current data does not allow understanding whether this is due to increased competition, changes in relevance, inventory volume, bids, or the types of dialogues in which ads appear.

    This hinders accurate cheap hypothesis testing and campaign optimization.

    Hostinger’s Experience: $70,000 on Tests

    One of the largest public tests was conducted by Hostinger, spending almost $70,000. Hussein Ograk, Head of PPC, initially noted that CPC was no higher than in Google Search, and campaigns generated purchases. However, as the budget increased, his assessment became more nuanced.

    “CPM exceeded $65, but the main problem was CTR. More specific cases worked better, while broad messages consistently failed. Traffic quality was unstable, and evaluating ROAS based solely on direct conversions proved difficult,” Ograk noted.

    This example demonstrates that what CPM to consider profitable is a matter of context and traffic quality.

    Other Tests: Attribution and Lead Quality

    CTC reported attributed revenue ranging from $19,000 to $38,000, depending on the attribution model, with ROAS from 3.3x to 6.8x. They used Triple Whale to assess downstream performance, as ChatGPT Ads reports were insufficient.

    Floyd Blakey, analyzing a B2B campaign with a budget of $7,000 CAD, recorded an average CPC of $9.29, CPM of $64.34, and CTR of 0.7%. Using visitor de-anonymization, her team identified 146 organizations from 336 paid clicks. Of these, only five matched the ideal customer profile.

    This highlights the problem of traffic quality and shows that cheaper than influencers does not always mean more effective.

    Geographical Differences and Recommended CPC

    The Synter test showed significant geographical differences. With an overall CPC of $9.89 (at $4,428.84), it ranged from $5.10 in the UK to $10.62 in the US. In Australia, CPC reached $17.59, in New Zealand — $22.89, although volumes there were significantly lower. Cost per click heavily depends on the region.

    OpenAI recommends a starting maximum CPC from to . However, this recommendation is often perceived as a benchmark, although it is not. This creates false expectations regarding a budget alternative to seeding and how much 1 million impressions cost via banner compared to AI advertising.

    ChatGPT Ads: 6 months later — what is a profitable CPM? — illustration 2

    ChatGPT Ads Audience

    Advertising in ChatGPT is only available to Free and Go plan users, who constitute the majority of the audience. More expensive subscriptions (Plus, Pro, Business, Enterprise, Edu) remain ad-free, as do accounts of minors. This means that the advertising audience does not coincide with the overall ChatGPT user base.

    There is insufficient public information about the demographics and purchasing power of this audience. General ChatGPT user statistics are also irrelevant, as they include those who do not see ads. Geography further narrows the comparison.

    • Pilot launch: USA, Canada, Australia, New Zealand (February).
    • Expansion: UK, Japan, South Korea, Brazil, Mexico (May-August).
    • Access to Ads Manager: 52 countries (August 31).

    How to Approach ChatGPT Ads Now

    When starting to test hypotheses cheaply on this platform, it is important to understand its current limitations. Set expectations before spending, based on what will make the channel profitable for your business. Someone else’s CPC or CTR will provide little value until we know so little about the competition, audience, and types of dialogues behind these results.

    Early results require more thorough analysis than on mature platforms. Due to a lack of diagnostic signals, it is difficult to distinguish a real performance problem from display peculiarities, relevance, or available dialogues during the test period. There is no universal number yet that determines the success of a campaign in ChatGPT Ads. “Good” is what the channel brings to your specific business, not abstract figures.

    Frequently Asked Questions

    What is a profitable CPM in ChatGPT Ads?

    A profitable CPM in ChatGPT Ads does not have a fixed value. It depends on the niche, traffic quality, campaign goals, and the ROI you receive. Hostinger’s experience showed a CPM above $65, but with CTR issues. It is important to evaluate not only CPM but also the quality of leads and the final ROAS.

    Is it worth investing in ChatGPT Ads for a cheap hypothesis test?

    ChatGPT Ads can be a tool for a cheap hypothesis test, but with caveats. Limited reporting requires additional analytics tools (e.g., Triple Whale, visitor de-anonymization) for a full evaluation. Be prepared for the price vs. result to vary greatly.

    What’s cheaper: banner ads or ChatGPT Ads?

    Comparing what’s cheaper: banner or blogger with ChatGPT Ads is not directly correct. CPC in ChatGPT Ads can be significantly higher than in traditional banner advertising, reaching $10-$20 in some regions. However, if the quality of traffic and ROAS justify these costs, then ChatGPT Ads can be more effective. It is important to calculate unit economics.

    What CPC should be considered normal for ChatGPT Ads?

    OpenAI recommends a starting maximum CPC of $3 to $5, but this is not a benchmark. The actual CPC can be significantly higher, up to $13-$22. A “normal” CPC is one that provides a positive ROAS and aligns with your unit economics. Do not rely solely on recommendations; always conduct a cheap hypothesis test and analyze your data.

    Conclusion

    ChatGPT Ads is a new platform with potential, but also with serious limitations, especially in terms of reporting and transparency. Advertisers accustomed to detailed statistics face the need for deep off-platform analytics to assess real effectiveness. To understand how much money each ruble spent will bring, do not rely on general figures, but focus on your metrics and goals.

    Conduct a cheap hypothesis test, analyze each stage of the funnel, and optimize campaigns based on your own data and expert opinions. This is the only way to find your profitable CPM and achieve success.

  • MrBeast and Google: A Multi-Year Partnership to Promote Gemini, Health, and Fitbit

    MrBeast and Google: A Multi-Year Partnership to Promote Gemini, Health, and Fitbit

    The largest YouTube blogger MrBeast, whose real name is Jimmy Donaldson, has signed a multi-year agreement with Google. This partnership aims to promote the company’s key products: the Gemini chatbot, the Google Health platform, and Fitbit wearables. For many businesses facing the challenge of choosing a contractor for integrated promotion, this approach demonstrates a “one-stop shop” model, where all services can be ordered in one place.

    According to an official Google statement, Donaldson, as the first creator to reach 500 million YouTube subscribers, constantly pushes the boundaries of what’s possible. Now he will use Gemini to realize his ambitious ideas, showing the audience how artificial intelligence can help bring the most incredible concepts to life. In addition, the blogger will demonstrate how Google Health supports fitness and wellness goals.

    Partnership Details: How MrBeast Will Promote Google

    The partnership kicks off on September 5 with a video in which MrBeast and his team will attempt to survive in the Earth’s most extreme conditions: the Arctic, jungles, and desert. In this challenge, they will rely on Gemini, which will help them identify dangers, cope with sudden weather changes, and stay one step ahead. This is an excellent example of how to use content outsourcing for social media, entrusting the production of large-scale projects to external specialists.

    Gemini as a Tool for Logistics and Survival

    • Danger Identification: The chatbot will provide information about potential threats in extreme conditions.
    • Weather Forecasting: Gemini will help the team adapt to changing weather conditions.
    • Project Logistics: Donaldson will also be the face of a new Gemini campaign, demonstrating how the app helps him organize and execute his large-scale stunts.

    “Jimmy Donaldson, better known as MrBeast, is always pushing the boundaries of what’s possible. Now he’s using Gemini to realize the boldest concepts, showing his fans how they can use AI to bring incredibly big ideas to life,” Google stated.

    Promoting Fitbit and Google Health

    In an upcoming video on his channel, MrBeast will use a Fitbit Air device, highlighting its functionality in everyday life and during challenging trials. This demonstrates integrated promotion through short videos, where each product is integrated into the natural context of the content.

    Why did Google choose MrBeast?

    The choice of MrBeast as a partner is no accident. He is the most popular content creator on YouTube with over 515 million subscribers and generates 4 to 6 billion views per month on his main channel. His participation in advertising campaigns for brands such as Disney, Lowe’s, and Jack Links, as well as his appearance in a Salesforce Super Bowl commercial, underscores his influence.

    Strategic importance of the partnership for Google

    While Google sometimes collaborates with YouTubers to promote products or events, this level of long-term commitment is unusual. This could be part of YouTube’s strategy to retain leading content creators who might otherwise move to other platforms, such as Netflix, which offer lucrative deals. For businesses, this is a lesson in the importance of long-term relationships with key partners and investing in them.

    MrBeast and Google: multi-year partnership to promote Gemini, Health and Fitbit — illustration 2

    Prospects for cooperation and ROI

    Google promises that this is just the beginning of their collaboration with MrBeast. Such a partnership provides high audience reach and demonstrates the potential of full-cycle short video production for promoting complex technological products. For those wondering what is cheaper – AI or real clipping, or whether uniqueization or new production is worth it, Google’s example shows that investments in high-quality, organically integrated content always pay off.

    Frequently Asked Questions

    How to choose a contractor for video content creation?

    When choosing a contractor for video production, especially for full-cycle short video production, pay attention to their portfolio, experience with similar projects, and their ability to offer comprehensive solutions from idea to posting. It is important that the contractor understands your business goals and can provide a replacement for an SMM team, taking on the entire process.

    What’s better: using AI for content creation or shooting from scratch?

    The choice between AI generation and shooting from scratch depends on your goals and budget. AI can be effective for quickly generating a large volume of content or repurposing, but for unique, high-quality, and emotionally rich stories, like MrBeast’s, shooting from scratch or deep uniqueization remain preferable. It’s important to find a balance between economy and quality so that the content truly captivates the audience.

    How to evaluate the effectiveness of mass posting?

    Evaluating the effectiveness of mass posting involves analyzing metrics such as reach, engagement, click-through rate, and conversions. It is important to track which types of content and platforms yield the best results to optimize the strategy. Content quality, even with mass distribution, plays a key role in achieving high performance.

    Conclusion

    The partnership between Google and MrBeast is a prime example of how large companies leverage the influence of popular personalities to promote their products. For businesses looking for effective solutions to order everything in one place — from video creation to its promotion, this model demonstrates the potential of a comprehensive approach. Remember that investing in quality content and strategic partnerships can significantly increase brand awareness and audience loyalty. If you feel that your current SMM team is not coping with the tasks, consider outsourcing content for social networks to get the maximum return on your marketing efforts.

  • Mass video posting: how an AI model created a pipeline for 127 n8n nodes for $4

    Mass video posting: how an AI model created a pipeline for 127 n8n nodes for $4

    Mass video posting and automatic publication of short videos is a dream for many content creators. Imagine being able to schedule auto-posting so that publication to 5 platforms happens without manual labor, while you focus on other tasks. This article is a detailed breakdown of how a mysterious AI model, later identified as Z.ai GLM 5.3 Flash, generated a complex mass posting service of 127 nodes in n8n, spending only $4 in the process.

    We will examine the architecture, the economics of the experiment, and the key lessons learned. Get ready to discover how artificial intelligence can radically change the approach to content creation and distribution, making it not only efficient but also incredibly economical.

    Introduction to the Experiment: AI Coder for Complex Pipelines

    From August 20 to 25, OpenRouter provided free access to the stealth/ox-alpha model, later de-anonymized as Z.ai GLM 5.3 Flash. This model, with a 1M token context and native multimodality (text, images, video), is positioned for “efficient coding and long-term agent tasks.”

    • Experiment Goal: Stress testing a new LLM as a coder for a complex distributed pipeline.
    • Task Scale: Five external APIs, binary streams, three dozen branching conditions, asynchronous polling – all in one workflow.
    • Result: In three days (August 21 to 23), the model “vibecoded” a workflow of 127 nodes, capable of interacting with APIs, generating media and publishing Reels.
    • Economics: 60.9 million tokens were spent, of which 85.3% were cache hits. The total cost at a blended price of $0.07/1M was about $4.

    “In the Western market, there’s a boom in autonomous bundles like ‘scraping-photo/video generation-auto-posting’. I’m interested in pipelines and fault tolerance.”

    Workflow Architecture: 5 Circuits and 127 Nodes

    Initially, the workflow had 156 nodes, but after refactoring, their number was reduced to 127. There are 113 working nodes, excluding triggers and stubs. Each circuit performs its specific function, ensuring short video distribution.

    Circuit 1: “Reconnaissance” (23 nodes)

    This circuit is responsible for finding viral content. It runs on a schedule (24 hours) and analyzes competitors’ Reels.

    • Scheme: Schedule (24h) → list of competitors → Reels via ScrapeCreators → virality math (views > avg × 2.5) → TRENDING tag.
    • For trending videos, caption (/v1/НЕЛЬЗЯgram/post) and transcript (/v2/instagram/media/transcript) are retrieved.

    • Circuit 1.5: “Semantic Filter” (17 nodes)

    Here, content relevance is evaluated using an LLM re-ranker.

    Circuit 2: “AI Dispatcher” (17 nodes)

    The model does not copy content, but “extracts the DNA of virality” and generates an original idea.

    Circuit 3: “Media Workshop” (37 nodes)

    This circuit is responsible for creating media files.

    Circuit 4: “Distribution” (19 nodes)

    The final stage is video cross-posting and content publication.

    Key Techniques for Effective Interaction with LLM

    The success of the experiment largely depended on the correct formulation of prompts. A unified panel for TikTok, YouTube, Instagram requires clear instructions.

    Engineering Solutions and Fault Tolerance

    The workflow contains many engineering solutions that ensure stability and fault tolerance.

    PCM to WAV on the fly

    The TTS model outputs raw PCM, which n8n and NELZYAGRAM do not understand. The model generated a Code node that adds a correct RIFF header using n8n’s built-in binary-helpers.

    Static Data: cycle accumulator without duplicates

    $getWorkflowStaticData('global') is used to collect trends, which avoids bloating the standard context and duplicating data.

    Mass video posting: how an AI model created a conveyor for 127 n8n nodes in — illustration 2

    Error Handling: 402 vs 5xx

    The system distinguishes between critical errors (402 — no funds) and temporary failures (5xx — service temporarily unavailable).

    Fail-open for reranker

    If Qwen3 Reranker does not respond after three retries, all documents are assigned a neutral score of 0.5, which is filtered out. The pipeline does not break but quietly skips the round.

    Model errors and my own

    Despite the impressive result, both model errors and shortcomings in my approach were identified during the process.

    Model flaws:

    My own blunders:

    Economics of the experiment: $4 for 60.9 million tokens

    The cost of 500 publications or even more, thanks to this experiment, turned out to be minimal.

    Price comparison with other models (average prices per 1M tokens):

    Model Price per 1M (in/out) Estimated for 60.9M
    Anthropic: Claude Opus 4.8 $5 / $25 about $426
    OpenAI: GPT-5.6 Terra $2 / $12 about $183
    Anthropic: Claude Sonnet 5 $2 / $10 about $171
    Qwen: Qwen3.8 27B $0.35 / $2.75 about $36
    DeepSeek: V3.1 Terminus $0.27 / $1 about $21
    stealth/ox-alpha + cache free about $4

    The paradox is that the free model built a factory that uses inexpensive services, saving significant funds on a monthly mass-posting package.

    Workflow growth areas and further development

    Further project development may include:

    Conclusion

    The experiment with stealth/ox-alpha showed that AI models are capable of creating complex and fault-tolerant pipelines for automatic publication of short videos. The key success factor is clear technical specifications and a structured approach to interacting with LLMs. This technology opens up huge opportunities for mass uploading with proxies and anti-detect, allowing to post to 10 accounts and more, significantly reducing labor costs.

    If you are looking for how to choose a mass posting service, pay attention to solutions that use similar AI approaches. Mass posting via API is becoming more accessible and effective than ever. Try applying these principles in your projects and see their power!

    Frequently Asked Questions

    What is stealth/ox-alpha?

    stealth/ox-alpha is the codename for an AI model that was later de-anonymized as Z.ai GLM 5.3 Flash. It has a 1M token context and native multimodality, designed for efficient encoding and agent tasks.

    How much did the workflow creation experiment cost?

    Thanks to a free trial period and a high cache hit rate (85%), the experiment cost about $4 for 60.9 million tokens.

    What platforms are supported for cross-posting?

    In this workflow, publishing to Instagram Reels via the Graph API was implemented. However, the architecture allows for expanding the list of platforms, including TikTok and YouTube, to create a unified dashboard.

    Is it possible to publish 100 videos a day using such a system?

    Yes, theoretically it is possible. The system is designed for automatic short video publishing and mass posting. Limitations will depend on the API throughput of the platforms used and computational resources.

    How to ensure fault tolerance in mass posting?

    Fault tolerance is ensured by protective logic: separation of 402 and 5xx errors, fail-open strategies for critical nodes, as well as asynchronous polling and retries. This guarantees that the schedule across 10+ accounts will be executed seamlessly, even during temporary failures.

  • AI Content Factory: How to Scale Content Production Without Losing Quality

    AI Content Factory: How to Scale Content Production Without Losing Quality

    Creating an effective AI-powered content factory is not just about automation, but about building an entire system where neural networks handle routine operations, freeing up time for strategic planning. We, as producers, know: a tired streamer cannot generate ideas. How do you package raw material into highlights without losing quality and without turning the process into chaos? This article will help you build an assembly line where AI will not be a replacement, but a powerful assistant.

    What is a content factory and why is it needed?

    A content factory is a systematized process of content production, based on the principle of “One idea – many formats”. AI here does not completely replace humans, but automates repetitive actions. Our goal is not just to increase volume, but to create mass cutting of 100+ clips and other formats, while maintaining quality and relevance.

    The role of AI in the content conveyor

    AI can be implemented at almost every stage of content production, from idea to publication. However, its role should always be auxiliary. For example, a neural network can suggest 100 blog topics, but the choice of the most relevant and valuable ones remains with a human.

    This is especially true for expert content, where factual accuracy is critically important.

    “A neural network can easily generate a hundred blog topics for a dental clinic, an English school, or an online store. The harder part is understanding which of these are actually worth pursuing.”

    Defining goals and initial data

    Before building a “conveyor,” it is necessary to clearly define what problem it should solve. The phrase “We need more content” is too vague. We need to understand what exactly is missing: time, people, or money.

    Choosing a content factory strategy

    • A) Accelerating publication for multiple platforms: If the task is to quickly manage Telegram, VKontakte, and other social networks with a small team, the factory should reduce the time from idea generation to a complete set of publications.
    • B) Increasing organic traffic: In this case, the content factory is built around semantics. AI helps analyze search queries, identify gaps, and generate ideas for new articles optimized for SEO.
    • C) Reducing production costs: If the team spends a lot of time gathering factual material, the factory will help repackage one expensive source into several formats, saving budget.

    • Searching for “Raw Material” for a Content Factory

    Quality raw material is the key to successful production. For each strategy, it is different:

    Developing a Route: From Raw Material to Finished Content

    After defining the goal and raw material, it is necessary to think through the content movement route. This will help to understand how long each stage takes and where automation can be implemented.

    Route Examples

    1. For multi-platform publication: 🛠️ Facts → Format Selection → Tasks → Drafts → Editing → Visuals → Publication.
    2. For SEO traffic: 🛠️ Semantics → Analysis of existing materials → Cluster Selection → Sources and Facts → Article → Publication → Indexing → Next Material.
    3. For cost reduction: 🛠️ Expensive source material → Processing → Maximum repackaging (e.g., cutting a podcast into video clips, cropping 9:16 from horizontal footage).

    Automating Repetitive Operations

    AI is most effective where there are repetitive actions that do not require a new solution each time. These can be:

    Establishing Rules for AI

    For AI to work effectively, it needs clear rules and boundaries. Without them, it can generate meaningless or irrelevant content.

    Content Factory with AI: How to Scale Content Production Without Losing Quality — illustration 2

    Examples of Rules

  • To save budget: Determine what material is considered standalone, when a derivative format adds value, and what cannot be cut without losing meaning.

    • Implementing AI Tools

    Only after defining goals, raw materials, routes, and rules can specific AI tools be selected. It is important to understand what limitation we are removing and what area we are automating.

    Recommended Tools

    Evaluating the Effectiveness of a Content Factory

    The content factory’s performance should be checked at three levels:

    1. Production speed: How much time passes from factual material to publication? Where do delays occur?
    2. Control cost: How much time is spent checking and correcting AI drafts?
    3. Content results: Read-throughs, saves, clicks, leads, organic traffic, video views. It’s important that 50 additional posts don’t go “into the drawer.”

    “Good automation should reduce the cost per useful unit, not just increase the number of files in a folder.”

    Frequently Asked Questions

    How long does it take to cut a two-hour stream?

    Using specialized AI tools, such as Creoscan, cutting a two-hour stream into main highlights can take from 15 to 30 minutes, including automatic subtitle addition and adaptation to different formats, for example, cutting for TikTok from Twitch.

    What moments are usually cut from streams?

    From streams, highlights, funny moments, important announcements, key answers to audience questions, gaming achievements, and the host’s emotional reactions are usually cut. Anything that can attract attention and go viral.

    Are subtitles needed for cuts?

    Yes, subtitles for cuts are extremely important. They improve content accessibility, allow viewing videos without sound, and increase audience engagement, especially on social media, where many users watch videos without sound.

    Can I order 20 clips from a stream?

    Of course! Many studios and freelancers offer services for cutting without loss of quality. With AI tools, this can be done quickly and efficiently, resulting in 20 clips from a stream, ready for publication.

    Who is the person who cuts streams?

    A person who cuts streams is a video editor or content manager specializing in creating short, engaging videos from long live streams. They select the most interesting moments, add graphics, music, and subtitles to transform raw footage into ready-made highlights for various platforms.

    Conclusion

    Building an AI content factory is not an instant solution, but a sequential process. Start small: choose one repeatable content stream, work it out manually, and then automate the sections that do not require a new solution. This will allow you to effectively scale production, create 20 clips from a stream or dozens of posts, and ultimately, monetize a stream through cuts. Ready to start your conveyor? Try the tools we’ve discussed and turn your ideas into ready-made content today!

  • Google’s Ad Business Will Stand: Court Rejects Asset Split

    Google’s Ad Business Will Stand: Court Rejects Asset Split

    A U.S. federal court has ruled that Google will not have to break up its advertising business, despite the company being declared a monopolist in April. Judge Leonie Brinkema rejected the Department of Justice’s demand to sell the AdX ad exchange and, potentially, the DFP ad server, both part of Google Ad Manager (GAM). This decision calls into question the ability of regulators to rein in giants like Google.

    Why did Google avoid a breakup of its advertising empire?

    Judge Brinkema found that a forced breakup of Google’s business could cause more harm to publishers than current anti-competitive practices. She pointed to potential negative consequences for small publishers who use DFP for free. Additionally, the acquisition of AdX by another party (e.g., Microsoft) could create new complexities.

    Google’s proposed measures and their impact

    Google had previously proposed a number of measures to enhance competition:

    • Providing competitors with real-time bid data for display advertising sold through AdX.
    • Abolishing “Unified Pricing Rules” and allowing publishers to set different minimum prices for individual bidders in GAM.
    • Refraining from using “first-look” and “last-look” privileges to adjust its bids. Google claims it stopped this practice several years ago.

    “What is a web publisher to do if they want to use a different ad server but still get demand from Google buyers?” asks Jay Friedman, co-founder of CartographAI and former CEO of Goodway Group.

    Market and expert reaction to the court’s decision

    While many Google critics are disappointed, some ad industry players are willing to see how the proposed measures will play out. For example, PubMatic, an AdX competitor, stated: “We expect the court’s adoption of behavioral remedies to create a level playing field for all market participants. The court’s focus on behavioral remedies will likely provide a faster path to addressing the current competitive harm from Google’s unlawful monopoly.”

    Behavioral remedies versus full divestiture: arguments from both sides

    The Department of Justice insisted that forced divestiture would be a “cleaner and less risky solution,” expressing distrust in Google’s ability to comply with prescribed behavioral measures. However, Google’s lawyers countered that “distrust is not a lever to circumvent established antitrust principles.”

    Previous rulings and the future of Big Tech regulation

    This is not the first time Google has been found to be a monopolist but avoided a breakup. In August 2024, Judge Amit Mehta found Google’s monopoly in online search but did not force the sale of Chrome or Android. Instead, Google was ordered to share search data with competitors.

    Google's Ad Business Will Stand: Court Rejects Asset Split — illustration 2

    Such decisions make ad industry insiders doubt the ability of regulators to deal with Big Tech. One former media buyer anonymously remarked that even “trust-buster” Theodore Roosevelt would be astonished by the lack of courage in the judiciary today.

    Frequently Asked Questions

    What is Google Ad Manager (GAM)?

    Google Ad Manager (GAM) is an integrated ad management platform that combines the AdX ad exchange and the DFP ad server. It allows publishers to sell ad space and advertisers to buy it.

    Why did the U.S. Department of Justice want to break up Google?

    The U.S. Department of Justice accused Google of monopolizing the ad tech market and insisted on breaking up the company to restore competition and prevent anti-competitive practices.

    What behavioral measures did Google agree to take?

    Google agreed to a number of measures, including providing bid data to competitors, abolishing “Unified Pricing Rules” for publishers, and refraining from “first-look” and “last-look” privileges in bidding.

    Conclusion: pennies for a million or millions for pennies?

    So, banner advertising in video remains a battleground where Google is currently holding its ground. For us, arbitrageurs, this means one thing: CPM cheaper than 50 rubles in Shorts, Reels, and TikTok is still possible. Don’t wait for these “behavioral measures” to level the playing field. While the court deliberates, competition, however skewed, offers loopholes. Continue to test hypotheses, seek cheap reach, and scale mass reach for pennies. And remember, real math is always on our side if we know how to calculate and act quickly while the “big guys” are sorting things out in court. Forget about bloggers and their million-dollar price tags – buying reach without creating a video through programmatic platforms is our path to profit. Learn how to set up your first campaign and start earning right now!

  • TikTok’s AI Revolution: How Neural Networks Are Changing Advertising and UGC Content

    TikTok’s AI Revolution: How Neural Networks Are Changing Advertising and UGC Content

    The era of mass video generation by neural networks is already here. If you’ve ever considered running ads on TikTok, but the Ads Manager interface seemed too complex and the entry barrier too high, then this article is for you. We will tell you how new AI videos for advertising and AI-powered tools like Agentic Hub, Symphony Creative Studio, and Content Suite simplify campaign creation, management, and optimization, allowing you to generate AI UGC for a crypto project or any other business without deep expert knowledge of the platform.

    TikTok AI: New Horizons for Advertisers

    TikTok is actively implementing AI tools to democratize the process of creating and managing ad campaigns. According to Melissa Lorie, a short-form video expert, if TikTok Ads Manager is a restaurant, then MCP (Marketing Campaign Platform) is the waiter, and AI tools (like Claude or ChatGPT) are the menu.

    Thanks to this integration, marketers can interact with platforms in a normal way, asking questions about campaign performance, requesting reports, and generating ideas.

    MCP: Two-Way Data Exchange with AI

    Before MCP, connecting AI tools to ad platforms required complex API integrations implemented by developers. MCP provides two-way communication between AI and the platform, allowing not only data extraction but also actions to be taken.

    Some connections are read-only (for reports), while others are read-write (for creating campaigns or adjusting settings directly from an AI tool).

    Pro Tip: AI recommendations should not be taken as gospel. If a new creative has only been running for a few days, TikTok’s AI might suggest pausing it because it’s performing worse than older creatives. However, this ignores the fact that the platform prioritizes ads that have been running longer.

    Marketers need to rely on their own judgment and challenge AI suggestions by asking questions like: “Is there another way to do this?” or “Can I do this better?” Often, AI revises its recommendations when challenged.

    AI Skills: Shortening the Path to Success

    Available skills include Viral Video Creator, Ad Group Optimizer, Account Diagnosis Co-pilots, and ByteDance’s C-Dance video creation tool. These skills can be found by searching for “TikTok AI skills” online.

    • Viral Video Creator: This skill, a favorite of Melissa Laurie, analyzes millions of videos processed daily by TikTok to identify the most effective ads. Marketers can use this data to model their own videos, from the catchy beginning to the call to action. Melissa estimates that the skill brings creative work to 70% completion; the remaining 30% requires human refinement, such as emphasizing specific product benefits. This allows the AI for UGC advertising to operate at peak efficiency.
    • Skill Flexibility: Since skills are files, they can be modified, copied, and improved. Marketers familiar with AI can add additional instructions on top of official TikTok skills, customize recurring tasks, and create their own automations.
    • Cross-Platform Compatibility: An unexpected benefit is that skills are not tied to the platform on which they were created. A marketer can use a TikTok copywriting skill to create ads for Meta or find that Meta’s Instagram copywriting skill works better for their TikTok campaigns. This interoperability represents a complete paradigm shift.

    Campaign Management Automation

    For campaign management automation, Melissa recommends starting with the following basic tasks:

    • Automating daily performance reports.
    • Querying AI about which creative has the strongest “hooks.”
    • Monitoring creative fatigue.

    When setting up a campaign, marketers can tell the AI their budget and goals (e.g., “I have $5,000, and I want to achieve X”) and request specific recommendations for campaign setup. This allows mass video generation by AI to become truly manageable.

    Symphony Creative Studio: AI-powered Production by Subscription

    Symphony Creative Studio is TikTok’s hub for creating AI-powered commercials. The updated Symphony Agent uses TikTok trends, signals from top-performing ads, business goals, and brand guidelines to create creatives at scale.

    The tool can take a campaign from brief to finished video, generating product briefs, insights, storyboards, scripts, avatars, and videos. Marketers then intervene to revise scenes, add voiceovers, and refine calls to action. This is true AI content production.

    Instant Video Generation from Text

    The interface is simple and resembles the query window of any other AI tool. Marketers add their prompts, brand guidelines, and other details, and Symphony generates videos almost instantly. This allows for generating video from text with unprecedented speed.

    Product Videos: Where AI Shines Brightest

    Product videos are an area where Symphony is currently delivering the strongest results. Melissa gives the example of a furniture company that uses this tool to create 100 videos a week, each showcasing a separate piece of furniture.

    Since these videos don’t require people on camera, the company can produce content at a scale that would be impossible with traditional video production. This is an AI pipeline of 500 videos per day for e-commerce.

    AI Avatars: Challenges and Prospects

    Animated AI characters in videos are not yet ready for most brands. The consensus among many leading companies is that the technology is not yet mature enough to create convincing human presenters. The trust issue is real: an AI avatar saying “all my friends are talking about this” raises obvious concerns. However, AI avatars with subtitles and AI voiceovers are already actively used for other purposes.

    Hybrid Approach: AI-Enhanced Real Videos

    Melissa sees potential in using AI to enhance real videos, rather than completely replacing them. Her agency, Oysterly Media, works with 3M on content for a mop. Creators film themselves demonstrating the mop, and then AI adds visual elements, such as animated “angry germs,” to dramatize the before-and-after effect.

    AI-революция TikTok: Как нейросети меняют рекламу и UGC-контент — illustration 2

    This hybrid approach—real creator plus AI enhancement—is one of two main ways that prominent brands are using AI video now. The other is product demonstration videos without people, especially in the beauty industry, where AI can render product materials, textures, and effects, such as how lip gloss looks when applied.

    Content Suite: Market Research with AI

    Content Suite is a library of UGC-style videos and ads running on the platform. The new AI search update significantly simplifies finding relevant content for research and ideas, solving one of the biggest challenges marketers face: coming up with new content ideas. This is a powerful tool for AI UGC for crypto projects and other niches.

    Smart Search and Competitive Analysis

    Marketers can search Content Suite by keywords, industry categories (e.g., airlines or beauty), time filters (e.g., last 30 days), campaign goal (sales, lead generation, or brand awareness), and geography. The tool functions similarly to Facebook Ads Library, but is easier to navigate, especially with AI search.

    Melissa recommends starting by searching for the company’s own brand, then expanding to competitor products and the broader industry category. The real value lies in seeing what creatives other brands in the same niche are launching and using that insight to create original content.

    Content Suite can also be used for direct competitive research, identifying what types of ads competitors are investing in and what creative approaches they are testing. For those who have not yet used TikTok advertising, Content Suite is the best starting point. Before spending money, browse the library to see what videos are being created in the relevant industry. This research expands creative thinking and provides concrete models to work with.

    Anatomy of a High-Performing TikTok Video

    As AI tools make ad creation more accessible, the quality bar on the platform is rising. Melissa argues that the way to stay above this rising bar is to understand what makes a video high-performing. The first frame matters most, not the first three seconds.

    Key elements of a captivating video:

    1. First Frame: The initial image of the video that a viewer sees before deciding to scroll past. This single frame determines whether someone stops.
    2. Billboard Text: A large text overlay on the screen. Melissa calls it “billboard text” because it works like a roadside billboard: the viewer has one moment to glance at it while scrolling, so it must instantly grab attention and convey the video’s value.
    3. Opening Words: If there’s a voiceover or someone speaking, the first few spoken words must immediately hook the viewer after the visuals have stopped the scroll.
    4. Connecting Tissue: Melissa’s term for the element that connects the beginning of the video to its middle section. Without this connecting tissue, viewers drop off between the “hook” and the main body.
    5. Middle Content: When sharing tips, Melissa recommends grouping them in sets of 3 or 5. Three is the ideal number to keep content concise.
    6. Abrupt End: The video should end with a sharp cut so that viewers don’t realize it’s about to end. A common mistake is slowing down and speaking more slowly towards the end, which signals the video’s conclusion and gives viewers a reason to scroll past. A clean cut tells the algorithm that viewers watched until the end, which helps with distribution.

    Authenticity vs. Polish

    Melissa’s agency preaches creating videos that feel like content from a family member or friend. Polished, staged videos make viewers scroll past because they are immediately perceived as ads. Naturalness is what drives performance.

    Tips for creating natural content:

    • Don’t stand still and talk directly into the camera. Instead, creators should be doing something while talking: picking up a water bottle, applying makeup, or walking through a door.
  • Even raw videos benefit from light editing. Melissa recommends setting a clear start and end point, adding subtitles (as many viewers scroll without sound), and using occasional jump cuts. These jump cuts signal to the brain that something interesting is happening, helping to retain viewers.

    • Pro Tip: Don’t give away the ending at the beginning of the video. Build anticipation by first hinting at the implications, then deliver the reveal in the middle or at the end. For example, instead of starting with “Instagram just launched feature X,” start with “Instagram just made a very big update. This will change how you do X, Y, and Z.” Then explain the implications before finally revealing the specific update. When the answer is given first, viewers have no reason to keep watching.

    Conclusion: The Future of AI Production

    Melissa Laurie, founder of Oysterly Media, believes that marketing agencies still have an important place in this new landscape. She compares it to flying: economy, business class, and first class all get passengers to the same destination, but some companies prefer the full range of agency services.

    What has changed is that the economy option – doing it yourself with AI – is now a viable path for businesses willing to invest time in learning. Want to learn more about how AI writes scripts and edits, and how to use mass generation APIs? Explore new TikTok tools, browse Content Suite, and start experimenting. The future of AI clips with face swapping and AI video factories for e-commerce is already here, and it’s accessible to anyone willing to master these technologies. Start scaling your content production today!

    Frequently Asked Questions

    What is Agentic Hub and how does it help with TikTok advertising?

    Agentic Hub is a TikTok platform that allows the integration of AI tools for creating, managing, and optimizing advertising campaigns. It simplifies marketers’ interaction with Ads Manager, enabling them to ask questions and receive reports in natural language, and automate routine tasks.

    Can TikTok AI tools be used for other platforms?

    Yes, one of the key advantages of TikTok’s AI skills is their cross-platform compatibility. For example, TikTok’s copywriting skill can be used to create ads for Meta, opening up new opportunities for marketers and changing the paradigm of working with different ecosystems.

    How effective is Symphony Creative Studio for video creation?

    Symphony Creative Studio allows you to generate videos from text, using TikTok trends, effective ad signals, and brand books. The best results are achieved when creating product videos without people in the frame. Some companies generate up to 100 videos per week, which would be impossible with traditional production. This is an excellent Synthesia alternative for mass production.

    Should you use AI avatars in advertising videos?

    Currently, AI avatars are not ready for most brands due to issues with persuasiveness and trust. Experts recommend using AI to enhance real videos, for example, by adding visual effects or animated elements, rather than completely replacing human presenters. However, AI avatars with subtitles and AI voiceovers can be useful in other contexts.

    How to get started with AI advertising on TikTok?

    Melissa Laurie recommends starting by exploring new TikTok tools and reviewing the Content Suite to understand what content other brands in your niche are creating. This will help form the basis for developing original creatives without initial advertising costs.