Let me tell you about a book I can’t stop thinking about — except the book, this time, is a software release note. On August 25, 2026, a video-generation model called Wan3.0 went live. It takes text, images, PDFs, or presentation slides and generates video up to 30 seconds long at 1080p, priced at 0.05 US dollars per second of footage. The release was announced by the company itself on August 25 and tracked by an AI-application weekly report carried on Tencent News.
Curiously, the same season marks the end of a more famous story. OpenAI’s Sora — the product that made text-to-video a household phrase — stopped serving its web and app versions on April 26, 2026, with its API scheduled to close on September 24. The old edition is going out of print, and the new one is already on the shelves.
Read the Marginalia of a Release Note
There is a marginal note in this release that says it all: 0.05 dollars a second. That is not a technical spec; it is a price list. A minute of generated footage costs three dollars. That is the moment a technology stops being a demo and becomes a service — the moment someone writes a price on it.
The second marginal note is the input list: text, images, PDFs, presentations. Not just a prompt box. The product is explicitly built to turn the documents you already own — your slides, your memos, your reports — into moving pictures. That is an editorial decision as much as a technical one.
Let me tell you why that matters. The first wave of AI video tools asked you to describe a scene from scratch. This one invites you to hand over your existing material. That is the difference between a toy and a workflow. A tool you can feed your deck into is a tool that inserts itself into how work is actually done.
The First Edition Tells You How the Author Thought
The first edition of the AI-video era was Sora. Its consumer service is gone now, and its API is winding down. Whether the shutdown was technical, commercial, or both, the record is unambiguous: a flagship product that defined the category in 2024 is no longer serving customers in 2026.
That is a sobering footnote for anyone who assumed the pioneer would stay the leader. In software, the first mover writes the grammar; the second wave writes the business model. The marginalia of this season suggest the second wave is pricing aggressively, integrating with office documents, and shipping at 1080p for a fixed per-second rate.
To be honest, I was skeptical about per-second pricing when I first saw it — it felt like a metered utility, not a creative tool. I have corrected that view. Metering is what makes the thing institutional. Creative teams can budget it; finance teams can approve it; the tool becomes purchasable, repeatable, boring — and boring is what industrial adoption looks like.
The Chapter Turn
Curiously, the two events land in the same week of August 2026: a Chinese model launching a 30-second, 1080p, document-friendly pipeline, and an American flagship leaving the consumer stage. I want to be careful not to overread a single product cycle. Sora’s API closure is a specific business decision; it is not proof of any industry-wide collapse. And one strong release does not crown a new dynasty.
But the direction of the marginalia is clear. The questions asked in 2024 were: can machines generate video, and will anyone care? The questions in 2026 are: how much per second, and will it read my slides? The first question was about wonder. The second is about workflow. That is what a maturing medium looks like — it stops being a marvel and starts being a cost line.
Let me tell you about the edition I’d keep. It is the one that understands the difference between generating a clip and running a production. The model that takes your PowerPoint and turns it into a film is not just faster; it is re-describing what software for video even is. The first edition told us what was possible. The new one tells us what is payable. Both are worth reading; only one is still on the shelves.
The Price Point Is the Story
Let me tell you why the number 0.05 dollars per second matters more than the model architecture. Every generation of creative tools goes through the same transition: first there is the demo, then there is the price list. Wan3.0 from Alibaba is the first video model to land with a public, industrialized price tag — 30 seconds of 1080p video for roughly $1.50, generated from text, images, PDFs, or a slide deck. The marginalia in that pricing sheet is the real news.
At that price, the arithmetic changes for everyone downstream. A commercial that used to cost tens of thousands of dollars per finished second can now be drafted by an in-house team for pocket change, then polished by a professional only where it matters. The release note does not say “the creatives are replaced”; it says “the first draft just got free.” Curiously, that is exactly how earlier tool generations behaved — the typewriter did not replace the writer, it replaced the clerk who copied the manuscript.
The second signal is the input list. Text, image, PDF, PPT — the model is being positioned as a document-to-video engine, not just a text-to-video toy. That is a deliberate industrial choice: the work product of most organizations is already a PDF or a deck, and the barrier to turning those into video has just dropped by orders of magnitude. First editions of creative tools that target the existing workflow tend to win the market; the ones that demand a new workflow tend to win the conference.
There is a quiet risk, and it deserves a paragraph. Cheap generation does not automatically mean cheap quality control. A pipeline that produces 100 unusable drafts per minute still costs money to curate, and the curation cost — not the generation cost — is what will separate profitable users from the rest. The marginalia of the next release will tell us who has figured that out.
What Sora’s Shutdown Actually Signals
Now the other side of the ledger: OpenAI’s Sora has shut down its web and app products, with the API closing in September. Let me tell you what that means, because the popular reading gets it backwards. Sora’s shutdown is not a verdict on video generation — it is a verdict on demo economics. A product that generated buzz but could not convert buzz into a paying pipeline at scale was a first edition that no publisher would keep in print.
Curiously, the two events arriving in the same month make the chapter break visible. One model lands with a public price and an input list aimed at existing documents; another model closes because it stayed a demo too long. The contrast is the industry’s real report card: the winners in this generation will be judged by pricing discipline and workflow fit, not by the prettiest showreel.
The chapter turn for creators is practical. If the tool you use publishes a price and a support schedule, you can build a pipeline around it; if it remains a demo, you cannot. The first editions that stay on the shelves are the ones with a price on them. The new chapter of AI video is not being written by the model with the most impressive footage — it is being written by the model with the most boring, reliable price list.
There is a second edition of this story waiting to be written, and it belongs to the independent creators. Curiously, the tools that industrialized generation usually start by empowering the smallest operators first — the ones who could never afford a production pipeline now hold one in a browser tab. The marginalia of the coming year will be written by the studios that are already using the price list as a production plan. First editions of the tool are out; the second editions of the businesses built on it are just beginning.
There is a footnote worth adding to the chapter, and it concerns the creative talent layer. Curiously, the tools that industrialize a craft usually trigger a temporary dip in perceived craft — the first waves of cheap output are formulaic. Then the curve bends back: the people who understand narrative, pacing, and taste become more valuable, not less, because they are now the constraint on a flood of generation. The marginalia of the next release will confirm it: the price list lowered the entry barrier, and the curation layer became the bottleneck. That is a good trade for the industry, and a better one for the people who kept practicing the human skills.
The chapter closes with a practical note for creators. The tools have industrialized generation; the discipline that remains is editorial. The people who will profit from this generation are the ones who treat the model as a first draft machine and reserve their own judgment for the final cut. Curiously, that is the same role the editor has always played — the difference is the volume. The marginalia of the next year will be written in the workflows of the studios that figured out the division of labor early.
The closing marginalia for this chapter belongs to the creators who will test the price point first. A tool that costs $1.50 per 30-second video changes the economics of pitching, of iteration, of client feedback loops. Curiously, the businesses that will feel it first are not the big studios but the freelance operators who live on iteration speed. The price list is the real release note; the model is just the instrument. First editions of the era are being priced now, and the ones that survive will be the ones with the most honest invoices.
The chapter’s final marginalia: the era of the demo is over, and the era of the invoice has begun. Tools that publish price lists will be integrated into pipelines; tools that stay demos will be remembered fondly and dropped. The creators who adapt will be the ones who treat the model as the first draft and their own judgment as the final editor. Curiously, that is the oldest division of labor in media — the machine just got faster. First editions of the new era are already on the shelves.
And the final marginal note for this chapter: the price list is the release note that matters. First editions of the tool era are priced now, and the ones that survive will be the ones with the most honest invoices. The model is the instrument; the invoice is the book.