SPICYFRAME · PRACTICAL GUIDES
AI video generator news for 2026: a dated, neutral roundup of new AI video models, shutdowns, studio disputes and laws, as of October 2026, with sources.
5 seconds · 480p · $3.99. Pay per video, no subscription required.
Updated
A public SpicyFrame example illustrating the guide. Results depend on the photo and prompt.
AI video generator news in 2026 has been dominated by four things: faster model releases with native audio, a major shutdown, copyright disputes with film studios, and new laws on synthetic media. Between January and August, Google updated Veo 3.1, Kuaishou announced Kling 3.0, ByteDance released Seedance 2.0, OpenAI said it was ending Sora, and the EU’s transparency rules for AI-generated content began to apply. This roundup is dated and neutral: every entry names a source, and we only include items we could verify as of October 2026. It is not a ranking, and SpicyFrame is not affiliated with any company named here. Where sources disagree on a date, we say so rather than choosing one.
On January 13, 2026, Google announced an update to Veo 3.1 centred on its “Ingredients to Video” feature, which builds a clip from reference images. According to Google’s announcement and press coverage, the update added native vertical (9:16) output, upscaling to 1080p and 4K, and more consistent results from the reference images. The update was announced for the Gemini app, YouTube, Flow, Google Vids, the Gemini API and Vertex AI, and Google said videos generated with its tools carry an embedded SynthID watermark. Why it matters: vertical output and reference-image control are now treated as core features rather than extras, the same direction image-to-video tools have been taking across the year.
In early February, Kuaishou announced the Kling 3.0 family: Video 3.0, Video 3.0 Omni, Image 3.0 and Image 3.0 Omni. The wire release, dated February 6, 2026, describes clips of up to 15 seconds, native audio in several languages and dialects, and support for text-to-video, image-to-video, reference-to-video and in-video editing in one architecture. At launch, early access was limited to Ultra subscribers, with a wider release promised. Sources differ on the exact announcement day (February 4 to 6), so we use the wire date. Around the same time, ByteDance released Seedance 2.0, a multimodal audio-video model; trade coverage places its launch in the second week of February 2026.
Seedance 2.0 quickly became the year’s main copyright story. Within days of launch, the Motion Picture Association publicly called on ByteDance to stop what it described as copyright-infringing activity, and cease-and-desist letters from Disney and Paramount Skydance were reported over a weekend in mid-February. CNBC, as cited by heise, reported that ByteDance said it would stop Seedance 2.0 from generating videos with copyrighted characters or prominent individuals, without detailing how. Dates of individual letters vary between reports. Why it matters: providers’ safety checks and content rules are tightening, which is why you should expect prompts about franchises, celebrities and real people to be refused on mainstream tools.
On March 24, 2026, OpenAI announced in a post on X that it would discontinue Sora in its consumer app and API. Engadget reported OpenAI’s explanation as growing compute demand and a shift of the Sora research team to world simulation research for robotics, and noted that OpenAI had not yet said when the app and API would become unavailable. Secondary outlets later reported an app closure on April 26 and an API end date in September, but we could not confirm those dates from an official OpenAI page, so check OpenAI’s help centre for the current status. The lesson for creators is practical: download and keep your finished work, and avoid building a workflow on a single tool.
Two policy milestones affect AI video. In the United States, the TAKE IT DOWN Act, signed May 19, 2025, made publishing non-consensual intimate imagery, including AI-generated imagery, a federal offence, and required covered platforms to run a 48-hour removal process, with the platform obligations applying from May 2026. In the European Union, the AI Act’s Article 50 transparency obligations, including disclosure of deepfakes, began to apply on August 2, 2026; commentators report that the EU’s Digital Omnibus on AI did not delay them, apart from a transition to December 2, 2026 for machine-readable marking by systems already on the market. This is a summary of law-firm and research commentary, not legal advice.
Trade blogs also list further 2026 releases, including newer open-weight Wan versions, PixVerse V6 and MiniMax models. We did not find primary sources we could verify for them in time for this roundup, so we leave them out rather than repeat unconfirmed dates. The same caution applies to any “Veo 4” specifications online: as of October 2026 we found no official announcement of that name. When you see a release date in a blog post, look for the company’s own announcement, a wire service or a major outlet before you rely on it.
Announcements follow a pattern worth knowing. A model is often unveiled at a conference or in a social post, reaches a small group of subscribers or API partners first, and only later becomes broadly available, so “announced” and “available to you” can be months apart. Headline numbers such as clip length, resolution or the count of reference inputs describe the maximum of the top tier, not what every plan receives. Leaderboards, including LMArena, are useful snapshots but move as new models are added, which is why we quote a ranking only with the phrase “at the time of writing”. Finally, check whether a feature is in the app, in the API or in both, because these often ship on different dates. A useful habit is to note the date you read a claim and revisit it a month later.
Looking across the year, three trends stand out. Native audio is now expected, with ambient sound or dialogue generated together with the picture. Reference control has improved, so a still photo can anchor a clip’s look more reliably. And providers are investing more in content rules and provenance, from SynthID watermarks to tighter prompt checks. For adults making suggestive but non-explicit clips of consenting people, that means rules matter as much as model quality: understand what your chosen tool refuses before you pay for it.
SpicyFrame is an independent 18+ image-to-video studio that uses Seedance 2.5, ByteDance’s model, which was 2nd on LMArena image-to-video at the time of writing. The service turns one to 12 of your own photos (JPG, PNG or WEBP, up to 10 MB each) into a vertical 9:16 MP4 with ambient audio, in 5, 10 or 30 seconds, at Medium 480p, High 720p or Ultra High 1080p. Prices are shown before payment, from $3.99 for 5 seconds at 480p, and no subscription is required. It is built for glamour, lingerie, bikini, boudoir, dance and kiss scenes between consenting adults. It does not produce explicit content, generation is moderated, and a prompt can be refused; this section makes no quality comparison with other models.
| Date | Event | Source |
|---|---|---|
| Jan 13 | Google updates Veo 3.1: Ingredients to Video, vertical 9:16, 1080p and 4K upscaling | Google blog |
| Feb 6 | Kuaishou announces Kling 3.0 family; up to 15 s, native audio; Ultra early access | Bernama wire release |
| Mid-Feb | ByteDance’s Seedance 2.0 draws studio complaints; ByteDance says it will add safeguards | heise, TheWrap |
| Mar 24 | OpenAI announces it will discontinue Sora in its app and API | Engadget |
| May 2026 | US TAKE IT DOWN Act platform removal duties apply | Skadden |
| Aug 2 | EU AI Act Article 50 transparency duties begin to apply | Law-firm commentary |
Veo 3.1 gained vertical and reference-image upgrades in January, Kling 3.0 and Seedance 2.0 launched in February, OpenAI announced Sora’s end in March, and new synthetic-media rules took effect in the US and EU.
Verified releases in this roundup include the Veo 3.1 update, Kling 3.0 and Seedance 2.0. Others are reported by trade blogs but not confirmed here.
OpenAI announced on March 24, 2026 that it would discontinue Sora in its app and API. Check OpenAI’s help centre for the current status, because reported shutdown dates are not confirmed by an official page we could read.
The MPA, Disney and Paramount Skydance said the model generated content using their characters. ByteDance responded that it would tighten safeguards.
Yes. Laws on non-consensual intimate imagery and deepfake disclosure apply to creators and platforms. Only use images of adults who agreed, and check local law.
It is ByteDance’s model, used by SpicyFrame, and was 2nd on LMArena image-to-video at the time of writing. Rankings change quickly.
No. It makes glamorous and NSFW-style clips of consenting adults, never explicit content, and moderation can refuse a prompt.
External pages, opened in a new tab. Rankings reflect the time of writing.