Hello, this is Hamamoto from TIMEWELL.
I first wrote this article in early 2026. Back then, OpenAI's "Sora 2" was at the center of the conversation, and side hustles and social media monetization built on it were spreading fast. Yet Sora is no longer available to consumers. OpenAI announced its discontinuation in March 2026, and the app and web experience closed on April 26[^1]. The lead changed hands in barely six months.
That is the speed at which the AI video world moves. So I have rewritten this article to reflect the situation as of June 2026. The tool names change, but the fundamentals of approaching this as a side hustle — copyright risk and the reality of monetization — have barely shifted. If anything, people keep falling into the same trap every time the tools turn over. That is what I want to nail down here.
Why Sora 2 Disappeared: Then and Now
Let me make clear what happened to Sora 2 first. Working from outdated information means searching for how to use a tool that no longer exists.
Sora 2 was the video generation model OpenAI released on September 30, 2025, and at the time it stood out for the accuracy of its physics and the naturalness of its synchronized audio. Because it could produce polished video quickly, it spread explosively among creators. Then, on March 24, 2026, OpenAI announced it was shutting down the Sora app[^2]. The reason was blunt: cost. According to reporting, Sora's inference costs reached an estimated $15 million per day, while lifetime revenue was around $2.1 million[^3]. Its user base, which peaked near one million, had fallen below 500,000 by the time the shutdown was announced[^3].
The deciding factor, reportedly, was the collapse of a major Disney partnership in early 2026 that OpenAI had been counting on for funding[^3]. The economic case for a dedicated video product was gone. OpenAI pulled the large bank of chips that had been running Sora and redirected them to training its next-generation model. Video generation capabilities are expected to continue inside ChatGPT, and a successor model reportedly in development is called "Spud"[^3].
This is the biggest "then to now" shift.
| Dimension | Then (early 2026) | Now (June 2026) |
|---|---|---|
| Leading tool | OpenAI Sora 2 | Google Veo 3.1, Kuaishou Kling 3.0 |
| Sora's status | Available to consumers | App and web shut down (API ends in September) |
| Copyright stance | Opt-out, widely contested | OpenAI shifted to opt-in |
The developer Videos API is also slated to stop on September 24, 2026[^1]. In short, Sora is on its way out for both consumers and developers. Anyone who was excited six months ago about "making money with Sora 2" now has to switch tools. That is the reality of AI video.
The Leaders Now Are Veo and Kling — Here Is How to Choose
So who leads after Sora? The short answer is Google's Veo and Kuaishou's Kling.
Google's Veo was updated to the "Veo 3.1" family in October 2025 and gained native 4K output in January 2026[^4]. It supports vertical video, which suits short-form clips and Reels well. The family comes in three tiers: the quality-first "Veo 3.1," the faster "Veo 3.1 Fast," and the cheapest, high-volume "Veo 3.1 Lite"[^5]. Lite reportedly delivers the same speed as Fast at less than half the cost, which makes it a practical choice for a side hustle that needs volume[^5]. Google has also signaled that it will replace the Veo model inside the Gemini app with a conversational video generation and editing model called "Gemini Omni"[^4].
Kling, made by China's Kuaishou, has reached the point where "Kling 3.0" is used as an industry standard by agencies and marketing teams in 2026[^6]. Kling's strength is length. While many models produce only five to ten second clips, Kling can generate up to three minutes in a single pass, which puts it ahead for anyone who wants to create long footage at once[^6]. Its physics and lip-sync are also well regarded.
On sets that demand consistency, like filmmaking, Runway's "Gen-4.5" is the pick. You can lock a character's appearance using up to three reference images, keeping the same face whether the scene is a dark alley or a bright office[^7]. Gen-4.5 generates dialogue, ambient sound, and music as a single multimodal output[^7]. The catch is that clips run five to ten seconds, so a two-minute video means stitching together around twenty of them[^7].
Here is my take. For high-volume social content, Veo 3.1 Lite or Kling; for long-form narrative, Runway — that is the practical split as of June 2026. We are not in an era where learning one tool sets you up for life. Sora vanishing in six months is the clearest proof of that. Do not fall too hard for any single tool. Keep yourself light enough to switch at any time.
Looking to optimize community management?
We have prepared materials on BASE best practices and success stories.
Copyright Risk Does Not Go Away When the Tools Change
This is the heart of it. The tools turn over, but the thorniest problem is unsolved: copyright.
The first popular use of AI video was generating content based on famous anime — One Piece being a frequent example — and uploading it to YouTube or Instagram with ads enabled. Platforms have been removing this content at volume. The mechanism is clear: these are derivative works of existing IP uploaded without rights-holder permission. Ads get stripped, videos get deleted, and in the worst case the whole account is suspended. This use case always carries the risk of losing accumulated followers and revenue overnight.
Japanese rights holders did not watch quietly. On October 28, 2025, the Content Overseas Distribution Association (CODA) sent a written request to OpenAI, asking it to stop using member companies' works — including those of Bandai Namco, Square Enix, and Studio Ghibli — to train Sora 2[^8]. CODA stated it had "confirmed that a large amount of Sora 2's output closely resembles Japanese content or images," concluding that unauthorized training was the cause[^8]. Shueisha, the publisher of One Piece, warned it would take legal action against any unauthorized use[^9].
The government moved too. Minoru Kiuchi, Minister of State for IP and AI Strategy, called anime and manga "irreplaceable treasures" and a source of Japanese pride at a press conference[^9]. Together with Digital Minister Masaaki Taira, he warned that if OpenAI did not comply voluntarily, the government could invoke provisions of the AI Promotion Act, which came fully into effect in September 2025[^9]. Japan's copyright system requires prior permission, so the opt-out model common overseas — "tell us if you don't want to be used" — does not protect against liability for Japanese rights holders[^8]. Under this pressure, OpenAI shifted to treating copyrighted content as opt-in, using it only where rights holders grant permission[^10].
The legal framework is worth knowing. Japan's Agency for Cultural Affairs distinguishes between the learning stage and the generation and use stage. The learning stage is broadly permitted under Article 30-4 of the Copyright Act, but publishing or selling output that resembles an existing work at the generation stage can infringe just as a human-made work would[^11]. Reproducing a celebrity's face without consent raises portrait rights and defamation issues on top of copyright. On YouTube, simply putting the name of a copyrighted work in a video title has, in some cases, led to account suspension.
If you want to make AI video a side hustle, the first thing to decide is not "what to make" but "what not to make." Do not ride on anyone else's IP. That alone removes most of the risk of revenue vanishing overnight. Build on original characters and brands, and have a place to publish consistently with others. Our community platform BASE helps you lay that groundwork.
Revenue Models That Are Safe, and a System You Can Sustain
Talking only about risk leaves you paralyzed, so let me be specific about what actually earns.
The reachable scale is genuinely large. Some Instagram channels using AI video tools have reportedly posted monthly revenues above ¥30 million. This is not unique to AI — it rides the familiar structure in which short-form video that grabs attention in the first three seconds spreads well on algorithmic platforms. What AI changed is the production cost and the speed of iteration. Advertising video that once required a film crew and hundreds of thousands of yen can now be produced quickly and reworked many times over.
That said, the revenue models I consider safe have an order to them.
- Ad revenue from original content. High-volume content with zero third-party IP scales best.
- Brand deals and sponsorships. Channels with an established audience attract corporate partnerships.
- Corporate video production services. Income is more stable than platform ad revenue.
- Education and consulting. Teaching AI video workflows has itself become a market.
What these four share is that they do not borrow anyone's IP. Original characters, original brands, grown with AI used only to speed up production. It is both the legally safe path and the one that builds lasting channel equity. Reported outcomes include first-month contracts above ¥4 million and going independent at ¥1.2 million per month, but those are the upside cases. Grounded income comes from the unglamorous, safe, original route.
Let me stress one thing that often gets overlooked: publishing a video is not the end. Now that AI can turn a face or character into an avatar and mass-produce footage that looks like you are on camera, it is getting hard to tell whether the person on screen is real or AI. That is exactly why people with genuine connections to their audience hold the advantage. A channel that only broadcasts one-way quickly drowns in a sea of similar mass-produced content. Whether you can turn a video into fans who gather, interact, and move on to the next event or product — that cycle is what separates winners in the AI era.
This is where community design comes in. Our BASE is an AI-native community and event platform that lets you spin up event pages quickly and give attendees a place to keep interacting after the event. You turn the attention gathered through video into a fandom rather than a one-off spike. As a way to avoid resting your AI video income on a single leg, this kind of foundation pays off. And if you want to redesign how your business uses AI from the ground up, our AI consulting service WARP can advise on strategy.
Summary
To close, here is what stays important in a field where the lead changed in six months.
- Sora 2's consumer side shut down in April 2026. The leaders now are Google Veo 3.1 and Kling 3.0; choose by use case and cost.
- Copyright risk persists no matter the tool. Reproducing others' IP, like anime or celebrities, is high risk and leads straight to account suspension.
- Japan moved OpenAI toward opt-in through CODA's request and government pressure. The premise that prior permission is required is not going away.
- The safe path to earning is to build on original content and brands, using AI only to speed up production.
- Beyond publishing video, whether you can design lasting connections with fans is the long-term differentiator.
The tool names may well have changed again six months from now. Even so, compete with your own expression rather than borrowing someone else's, and build real relationships with the people who watch. Hold to those two, and whatever comes next, you won't be caught off guard. Rather than chasing the trendy tool, I believe firming up the foundation takes you further in the end.
Community and Event Management with AI | TIMEWELL BASE
You want to turn the attention gathered through AI video into lasting connections with fans, not a one-off spike. TIMEWELL BASE is the platform for that.
It covers everything from event page creation to member interaction and payments in one place. In practice, it supported Adventure World in running a Dream Day with 4,272 attendees, and TechGALA 2026 in centrally managing 110 side events.
We will think through how to convert the awareness you build with AI video and social media into a sustained community.
Related Articles
- Accelerating Growth in the AI Chip Market: 2026 Trends and TSMC/NVIDIA Strategies
- ElevenLabs 2026: Valued at $11B, Surpasses $330M ARR, Dominating the Voice AI Market
- From 2026 to 2027: The Real Prospect of AGI Arrival, Anthropic CEO Expresses Conviction
Footnotes
[^1]: OpenAI Help Center, "What to know about the Sora discontinuation." Consumer app and web ended April 26, 2026; Videos API ends September 24, 2026. https://help.openai.com/en/articles/20001152-what-to-know-about-the-sora-discontinuation
[^2]: CNN Business, "OpenAI is shutting down its Sora video app just months after launch," March 24, 2026. https://www.cnn.com/2026/03/24/tech/openai-sora-video-app-shutting-down
[^3]: TechCrunch, "Why OpenAI really shut down Sora." On the estimated $15M/day cost, user decline, the collapsed Disney deal, and the shift of compute to the "Spud" successor. https://techcrunch.com/2026/03/29/why-openai-really-shut-down-sora/
[^4]: Google DeepMind, "Veo 3.1" model page. On the October 2025 Veo 3.1 release, January 2026 4K support, and the move to Gemini Omni. https://deepmind.google/models/veo/
[^5]: Google, "Build with Veo 3.1 Lite, our most cost-effective video generation model." On the three-tier Veo 3.1 family and Lite's cost advantage. https://blog.google/innovation-and-ai/technology/ai/veo-3-1-lite/
[^6]: Atlas Cloud Blog, "Runway Gen-4 vs. Kling 3.0." On Kling 3.0's up-to-three-minute generation and its standardization in professional use. https://www.atlascloud.ai/blog/guides/runway-gen-4-vs-kling-3-0-which-image-to-video-ai-wins-for-professional-filmmaking
[^7]: Runway Research, "Introducing Runway Gen-4." On character consistency via up to three reference images and Gen-4.5's multimodal output. https://runwayml.com/research/introducing-runway-gen-4
[^8]: Variety, "Trade Group Representing Studio Ghibli, Other Japanese Companies Tells OpenAI to Stop Using Their Content." On CODA's October 28, 2025 request. https://variety.com/2025/digital/news/studio-ghibli-openai-sora2-japanese-trade-group-coda-letter-1236568751/
[^9]: GameSpot, "Studio Ghibli And Japanese Game Publishers Demand OpenAI Stop Using Their Content In Sora 2." On Shueisha's warning, Ministers Kiuchi and Taira, and the AI Promotion Act. https://www.gamespot.com/articles/studio-ghibli-and-japanese-game-publishers-demand-openai-stop-using-their-content-in-sora-2/1100-6535845/
[^10]: The Japan Times, "Japan's government asks OpenAI to seek permission amid Sora 2 copyright concerns." On OpenAI's shift to an opt-in model. https://www.japantimes.co.jp/business/2025/10/16/companies/japan-opt-in-model-sora2/
[^11]: Agency for Cultural Affairs (Japan), "AI and Copyright." On the two-stage framework separating the learning stage from the generation and use stage, and Article 30-4 of the Copyright Act. https://www.bunka.go.jp/seisaku/chosakuken/aiandcopyright.html