Kling 4.0 Is Here, AI Short Dramas Hit 430,000 in a Year: What's Behind Chinese AI Video's Leap Forward

On September 28, Kuaishou's Kling AI posted a message on its official account: Kling 4.0 officially launches in October, and the lightweight Flash version has already opened beta testing for annual subscribers. The announcement was short, but the comments section was buzzing. Over the past year, the Chinese AI video sector has moved from "can it generate?" to "can it deliver?" — and the parameters Kling updated this time land exactly on the spots that actually give people headaches.

First, the numbers. The Flash version outputs 720p at 3 to 20 seconds per clip; the full version stretches a single clip to 30 seconds at up to 4K, with 10-bit HDR to follow. Aspect ratios from 21:9 to 9:16 are all supported. But these aren't the crucial part. What people making product videos fear most isn't insufficient image quality — it's not being able to change anything: when the client says to swap the angle on the hero image, you have to redo the whole thing, faces and color grading included. What Kling 4.0 is pushing this time is precisely turning "redo" into "edit."

The first is local editing. In an already generated video, you can change just one element — expression, action, camera movement, style, or background — while everything else stays put, with up to 5 video inputs. This follows the same idea as the local editing that Qwen-Image 2.1 does on the image-generation side, just moved over to video. The second is keyframes: image-to-video accepts up to 10 keyframe images, so you can lock in one for the hero shot, one for the usage scenario, and one for the packaging close-up, and let the model handle the motion in between. In practitioners' words, keyframes are the storyboard — and the storyboard can be confirmed by the client before you spend any credits, so rework naturally drops. The third is Omni Reference: up to 15 inputs at a time, mixing images, video, and voice, with @image1 and @video1 used in the prompt to specify what each one is for.

Pricing is worth a mention too. On an annual plan, the entry tier is $21 a month for 180 credits, roughly $1.9 per video; the advanced tier is $90, about $1.1 per video. All tiers are watermark-free, so the final cut can be delivered directly. This pricing is aimed at the overseas market: Google's Veo 3.1 standard tier costs $0.4 per second, while Alibaba's Wan 3.0 runs $0.05 to $0.2 per second. Chinese models are trading price for market share — that's already an open secret.

Behind Kling 4.0 lies an entire explosive curve of AI short dramas and anime-style dramas. Xinhua reported a set of figures: in the first 8 months of 2026, 430,000 micro-dramas went live, 13 times last year's full-year total, with AI series accounting for more than 90% of them. In DataEye's report, AI anime-style dramas rose from 7% of the top-100 list in the same period of 2025 to 38%. One intuitive comparison is cost: a live-action micro-drama costs 50,000 to 100,000 yuan per episode with a 2-to-4-week cycle; with AI involved, per-episode cost drops below 5,000 yuan and the cycle shrinks to 3 to 7 days.

The most interesting phenomenon in this wave is the popularity of "simulated live-action dramas." "Zhanxiantai Live-Action AI Edition," launched in January 2026, passed 110 million views on Douyin; "Feng Shui Master" topped 40 million on Hongguo's popularity index and is already into its sixth season. On September 30, four of Hongguo's top five trending shows were simulated live-action dramas, with only one actually shot with real actors. The line that circulated most in the comments: real actors are ugly in a thousand different ways, while AI is beautiful in exactly the same way.

That line is both praise and criticism. Just as AI dramas dominate the charts, a visceral aversion triggered by the "AI face" also trended. Some netizens compared screenshots and found that the male leads in different shows were almost the same face: same face shape, same hairstyle, even the same center-parted bangs. The technical explanation is that video generation has to keep faces consistent frame by frame, and symmetric, regular, flawless faces have the highest tolerance during shot transitions, so models prioritize outputting a "safe" standard face. Add to that the many teams that directly reuse hit templates and call public digital face libraries in bulk, and homogenization accelerates even faster. At the end of June, People's Daily published "Tired of AI Faces? It's Time for the 'Human Touch' to Come Back," whose core argument is: technology can be used to cut budgets and cycles, but assembly-line capacity cannot replace artistry and humanity. The China Netcasting Services Association is also leading the drafting of "Quality Evaluation Standards for AI Micro-Drama Content," listing character recognizability as an independent scoring item.

So this round of "cornering" by Chinese AI video isn't fundamentally one company's technological miracle — it's the formation of an entire industry chain: on the model side, Kling, Jimeng, Hailuo, and Vidu each hold their own ground; on the workflow side, tools like Doubao Skill and CapCut AI motion effects bring the barrier down to phone level; downstream, platforms like Douyin and Hongguo support AI content with traffic pools. For an AI anime-style drama, a skilled team can already compress the journey from script to rough cut from several days to a few hours. What once required an entire film crew can now be approached with a phone and free tools.

But the lower the barrier, the sharper the old question becomes: when anyone can produce a drama a day, why should the audience watch yours? Homogenization is a side effect of the technology-dividend phase, and also the signal of the next elimination round. Only teams that have already started working on character recognizability, differentiated storytelling, and treating AI as a tool rather than a stand-in have a chance to break out of the "all the same" trap. Kling 4.0's local editing and keyframes give exactly the people who want to do fine-grained work a set of handier tools. The rest still comes down to the people.