Topic
Platforms that generate and edit video from prompts, orchestrating multiple underlying models behind a creator workflow.
Kuaishou's Kling AI team announced Kling 4.0 on September 28 with 30-second clips, up to 10 keyframes and up to 15 Omni Reference inputs. A dev.to explainer describes the input as a shot list plus a reference set, though the full model is not public yet.
Reality
OpenAI is in talks to give a filmmaker-focused startup access to its shelved Sora video model, The Information reported Oct. 12. No fee or term has been reported, so licensing is one reading of the talks and an equity-for-access swap is another.
Reality
HeyGen's Avatar V keeps one person recognisable across scenes and outfits from a 15-second recording. Its strongest evidence is a benchmark HeyGen ran itself, so buyers still have to prove it on their own footage and with their own viewers.
Reality
Kling AI has hired CICC, Goldman Sachs and UBS for a Hong Kong IPO of at least $1 billion, after a July round valued it at $15 billion or more. A float of about 6% would set the first public price for an AI video business valued privately at 30 to 36 times its revenue run rate.
Reality
Runway debuted Praxis-1, an open-weight world action model it says adapts to new robot bodies with light fine-tuning, alongside an ad agent called Runway Ads. Because the weights are open, outside robotics teams can measure what that fine-tuning costs on their own machines.
Reality
Higgsfield AI spent $400,000 of Hell Grind's $500,000 budget on compute and $100,000 on the 15 people who made the 95-minute film. Its reported $5.4 billion valuation turns on how much of that compute spend it keeps from the model makers behind its platform.
Reality
State broadcaster CCTV says the average minute of AI short drama in China fell from about 5,000 yuan to a few hundred in the first half of 2026. Local governments pay part of the compute and rent behind it, so buyers cannot yet tell the lasting saving from the subsidy.
Reality
Google charges $0.05 a second, or $3 a minute, for Veo 3.1 Lite video and made Vids generation free on the day OpenAI's Sora API shut down. Teams losing Sora can budget against that rate. Google has not published the free tier's allowance, and Decrypt's test of it produced only 720p.
Reality
Every net dollar goes to AI infrastructure, three days after quarterly profit fell 75%. For anyone buying cloud AI, that changes which number tells you whether capacity arrives.
Reality
Generating the interface as video instead of rendering it from code is a serious research bet. Runway's own preview still lists legible text and long-session coherence as open problems, which is roughly where ordinary interfaces begin.
Perspective Coverage
5 publishersReality
Just 1.3% of the 221,900 AI dramas uploaded to Douyin in the first half of 2026 cleared the 50 million-view break-even line, according to an industry report. Viewing is concentrating in fewer hits while cities subsidise the compute behind every title.
Reality
YouTube will let creators order their vertical clips as episodic series. The format ran up 6.5 billion views on the platform in six months, and the industry behind it is projected to make $14 billion in 2026.
Reality
An internal tool at a video-generation service gave editors a choice between roughly thirty confirmation stops on an attended run and runAll, which skips the lot. About thirty-five PRs later, the attended path stops four times.
Reality
Dantiva's video generation is three Google model calls. The schedule went to accounts, a reserve-capture-refund token ledger, support tooling, and making one wallet work behind three different sign-ins.
Reality
The Verge sat through all 150 minutes of Odysseus: The Fall, a film that cannot keep a cyclops at one height. The errors it documents cluster at the handoffs between generated shots.
Reality
The agent returns layers, keyframes and timing that a designer can revise inside Adobe. Running it means paying for Higgsfield credits, a ChatGPT plan and After Effects at the same time. The runtime figures are Higgsfield's own.
Reality
The speed figure comes from a single dev.to write-up and describes generation, not delivery. The design decision underneath it is that frames and stereo audio leave in one pass, so the smallest thing you can retry is the whole clip.
Reality
NVIDIA Research rebuilt the serving stack for someone else's open-weight video model and got a five-second clip out in 1.653 seconds, but the default profile is lossy and the dense comparison run is the honest baseline.
Reality
Google has put video continuation in front of paying Gemini subscribers with a 10-second context window and a 40-second ceiling, which is enough for a short spot but only if you understand what the model can still see by the fourth segment.
Perspective Coverage
3 publishersReality
fal's H3 Max clears real-time playback by two thirds at the fast end of its range, which turns continuous video into a metered compute bill. The 12.5-cent promotional clip price is what the arithmetic rests on.
Reality
Leadership · August 27, 2026 · One report1 publisher
Invest · August 17, 2026 · Confirmed3 publishers