AI Creative Tools News: Sora Exits, Agents and House Models Move In

Here is the short version. OpenAI has shut down Sora, the video generator that defined this category a year ago. Adobe has turned Firefly into an agent that runs work across its apps. Google has folded its video tools into a model called Gemini Omni. And Canva, which spent years renting other companies’ models, now ships one of its own. Anyone following ai creative tools news today watches a race change shape, from who builds the best single generator to who owns the whole desk.

This roundup covers the moves that matter, and it checks each one against the maker’s own announcement or help pages before printing it. It starts with the exit, then the agent, then the video and design races. It ends with the thread under all of it: who owns the model, and who holds the rights to what the model makes.

The biggest story is a goodbye: OpenAI ends Sora

Sora 2 launched at the end of September 2025 and owned the conversation for months. It made short videos from a written prompt, with sound, inside an app that felt like a social feed. On March 24, 2026, OpenAI announced it was shutting the product down. The company’s help center page on the Sora discontinuation now lists the two dates that closed it: the Sora web and app experiences ended on April 26, 2026, and the Sora API ended on September 24, 2026.

OpenAI gives no detailed reason on that page, and this magazine will not invent one. Reporting at the time pointed to the heavy cost of running video generation and to a company shifting its people and computing power toward business products. What the help page does cover is practical. People can still export the videos they made through a sunset page, and OpenAI says it will permanently delete the data tied to Sora once the period of any final export window passes. If your old clips live there, that export is the one deadline in this story that still belongs to you.

The shutdown also ended the most famous licensing deal in AI video. In December 2025, Disney agreed to license more than 200 characters from Disney, Marvel, Pixar and Star Wars for Sora videos, alongside a planned 1 billion dollar equity investment in OpenAI. We covered that deal in our ai news today december 2025 roundup. Reuters reported in March 2026 that the investment never closed and no money changed hands, so the deal died with the product it was built on, months before the first licensed videos were due to reach Disney+.

Adobe answers with an agent that directs the apps

Adobe chose the opposite direction. On April 15, 2026, the company unveiled Firefly AI Assistant in its newsroom, calling it a creative agent. Instead of opening Photoshop, Premiere or Express one at a time, a creator describes the outcome in plain words, and the assistant plans the steps and runs them across the apps. Adobe says the assistant asks questions back, surfaces the decisions that matter, and lets the creator step in, guide or adjust at any point.

The same announcement widened the studio around the agent. Firefly’s roster passed 30 models, including Google’s Veo 3.1, Runway’s Gen-4.5 and Kling 3.0, sitting next to Adobe’s own models in one place. The Firefly Video Editor gained speech cleanup borrowed from Premiere, detailed color controls, and a door into the Adobe Stock library. Two image tools, Precision Flow and AI Markup, target the complaint every designer knows: an AI edit that changes the whole picture when you asked it to change one corner.

Read the strategy plainly. Adobe cannot out-demo a video lab, so it sells the desk the work already sits on. It also repeats one promise about its own Firefly models: Adobe trains them for business use and markets them as commercially safe. That promise matters more to a brand team signing a contract than any single demo does, and it leads straight into the rights thread below.

Google folds video into Gemini Omni

Google’s video story moved in stages. The ai creative tools news today october 2025 headline was Veo 3.1 arriving in Flow on October 15, with sound added to its Ingredients, Frames and Extend modes. Google said at the time that creators had already made more than 275 million videos inside Flow in its first five months, a number that showed how fast the tool had spread.

In August 2026, Google introduced Gemini Omni 1.1 Flash and changed the shape of the tool. Omni takes text, images, audio and video as input and returns video, and it edits inside a conversation: extend a scene, fill in the motion between a first frame and a last frame, upscale the result to a crisp 4K. Google’s own August 2026 announcements roundup lists Omni 1.1 Flash inside Flow, Google AI Studio, the Gemini app and its enterprise agent platform. Veo 3.1 still runs inside Flow for dedicated video work, but the Gemini app now speaks Omni.

The practical change is where the work happens. A year ago, AI video meant a specialist tool and one careful prompt. Now the model sits inside the chat and the editor that people already open, and the skill shifts from writing a perfect prompt to directing a series of edits, the way a director works through takes with an editor.

Runway takes its models to the enterprise floor

Runway spent 2026 doing quieter work. Its Gen-4.5 model generates video from text and images, with close attention to scene direction and camera movement. Aleph 2.0 works the other side of the desk: it edits footage that already exists, changing one element or restyling a whole scene and carrying that edit across the shots that follow, up to 30 seconds at 1080p. On October 14, 2026, Baseten announced that companies can deploy both models in dedicated, single-tenant environments, with the model weights licensed directly from Runway.

That is a different kind of news from a launch demo. A studio or a brand can fine-tune the models on its own material, so the output reflects its own products and guidelines. Each deployment runs isolated, on reserved capacity, away from other customers’ work. Baseten says enterprise creative teams at companies including Amazon and Microsoft already run Runway in production. We repeat that as Baseten’s claim, not as our own measurement. The direction is clear either way: video generation now sells as infrastructure, not only as an app.

Canva stops renting and starts building

The loudest ai creative design tools news landed in the last two weeks. On October 22, Canva launched Craft, the first image model it built and trained from scratch, inside its Dream Lab tool for every user and through its public API for developers. Craft plans a picture before it renders it, separating the layout from the painting, which Canva says produces cleaner text and more precise placement. It writes text in 55 languages and renders at up to 2K resolution.

Canva frames Craft’s debut as a ranking result. The model opened fourth on DesignArena, a public leaderboard where people vote between AI designs, and Canva says that position makes its research lab the second most advanced for image generation, behind only OpenAI. Treat that as Canva’s claim, measured on a voting site, not as a settled title. The harder fact underneath is the build itself. Canva Research, an in-house lab of 160 people, trained Craft on millions of human made templates from Canva’s own content library, then tuned it with feedback from the company’s professional designers. Those designers also judged more than 25,000 head to head image pairs during the tuning, and Canva says it sided with them whenever automated scores and human judgment disagreed. On DesignArena’s category boards, Craft took first place for people and portraits and for landscapes and nature.

The same build points at editing, the job designers actually spend their days on. Because Craft plans a layout before it renders, its images arrive as layered pieces that Canva’s Magic Layers can turn into fully editable design elements, and Canva says this structure lays the technical foundation for more advanced editing tools in future releases. For readers tracking the wider field, our genai news coverage follows this same shift, labs becoming platforms and platforms becoming labs, across the industry.

The race, in one view

Here is ai creative tools news today as a lineup. Five companies, five different answers to the same question: what does a creative tool become when the model can plan, edit and build on its own?

Two details in that table tell their own story. OpenAI’s empty chair shows that even the best known name in AI will close a creative product that no longer fits its plans. And no row belongs to a simple prompt box anymore. Every surviving flagship either directs other tools, edits existing work, hosts rival models, or runs inside a platform with paying customers attached.

For working teams, the buying question changed with it. A marketing lead comparing tools this winter no longer asks which demo looks best. She asks which system her team can direct, brand and defend, the same shift we track in our guide to generative ai for marketing.

Three shifts behind the headlines

Step back, and the year’s generative ai creative tools news resolves into three moves. First, generators became agents. The new products do not stop at one picture or one clip; they plan a sequence, switch tools, check the result and report back, which is exactly how Adobe describes its assistant. The creator’s job moves up a level, from operating the tool to directing it.

Second, rented models became home built models. Canva’s Craft announcement explains the economics in plain terms: with more than a quarter of a billion people creating some 400 designs every second, renting a frontier model for every task adds up fast. So Canva distilled Craft down to a fraction of its original inference steps, built a custom engine that runs its planning stage an order of magnitude faster, and already promises Craft-fast and Craft-ultrafast variants at a fraction of the cost. A model trained for one job, on the company’s own material, cuts that cost and holds the style in house. Third, the open feed became a licensed stage. Sora’s rise ran on attention, and its fall ran through a rights deal that never closed. The tools that remain sell control over rights as a feature, not as fine print.

The rights thread under every story

Copyright and training data explain more of this year than any benchmark does. Adobe markets its own Firefly models as commercially safe and builds its studio pitch on that safety. Canva trained Craft on its own template library, work its designers made for its own product. Runway’s enterprise offer leans on isolation: your material, your private deployment, your fine-tune.

The Disney collapse marks the other side of the same thread. A licensing deal can carry a billion dollar headline and still vanish before it closes, because the product underneath it closed first. Creators who build a business on licensed characters, rented models or a single app now carry that risk in the open. Our explainer on ai governance business context strategic visibility sets out how companies keep track of exactly these exposures before they sign.

None of this slows the tools down. It decides which tools a careful team can sign. The contest readers watch in public is about quality. The contest that decides contracts happens in license terms, training disclosures and indemnity clauses, and it runs all year without a launch event.

What to watch next

Adobe MAX, the company’s big creative conference, runs November 10 to 12 in Miami Beach, and Adobe rarely saves small news for its biggest stage. Expect the Firefly agent and the video editor to headline it. Watch, too, whether OpenAI’s exit stays an exit: the company still offers image creation inside ChatGPT and points Sora users there, which leaves video as the obvious missing piece. And watch Canva’s lab. It already promises Craft-fast and Craft-ultrafast variants and calls this first version of Craft just the start.

Conclusion

The state of ai creative tools news, in one breath: the most famous video app in the world closed, the biggest design suite put an agent in charge, Google moved video into its chat model, Runway sold its models as infrastructure, and Canva became a model maker. The tools did not get quieter this year. They got employers, landlords and lawyers. The winners from here are the ones a creator can direct, trust and keep.