A comic-drama (漫剧) studio burns a week rendering episode 3, only to find the lead's face shifted, the café background floated, and the camera perspective broke between cuts. The client rejects the delivery. The team starts over.
This is the quiet bottleneck strangling AI comic-drama in China — not "can it look good once" but "can it look the same a hundred times." A platform inside the 360 (奇虎360) ecosystem is betting that solving consistency, not flashier renders, is the real product.
What the "nano-pipeline" is
The product is called 纳米大片流水线 (Nano Blockbuster Pipeline), built by Tianjin Zhiju Technology (天津智炬科技), a wholly owned subsidiary of 360 Group. It was publicly tested in late January 2026 and formally released in February 2026, positioned as China's first industrial-grade AI comic-drama agent platform (工业级AI漫剧智能体生产平台).
Comic-drama (漫剧) is the vertical, short-form, motion-comic cousin of the micro-drama (短剧) boom — fast cuts, episodic, made for feed-scrolling phones. It lives or dies on volume: studios need dozens to hundreds of episodes, not one hero shot.
The consistency engine is the pitch
Zhiju's core claim is a self-developed "nano-space engine" (纳米空间引擎) that locks character face shape, scene structure, and prop position inside a unified "3D space + timeline" coordinate system. In plain terms: once a character is set, the face should not change and the room should not drift no matter how many angles you cut.
According to the platform's published technical indicators — which are company-claimed, not independently verified — the system delivers:
- cross-shot facial consistency above 92 percent
- background deviation under 2 pixels between scene switches
- style drift below 3 percent
- per-episode render time of 30 to 60 minutes
- material success rate above 90 percent
- customer monthly output lifted 3 to 5 times (from roughly 10–20 episodes to 30–100)
Treat every one of those numbers as the vendor's own marketing figures until a neutral benchmark confirms them. The interesting claim is not any single percentage but the promise of predictable, deliverable output — the thing commercial buyers actually pay for.
Who is supposedly using it
Zhiju says that within three months of release the platform served more than 1,000 content studios and covered about 80 percent of top-tier comic-drama MCNs — again, platform-reported figures carried by Chinese business press rather than audited data. Named adopters in coverage include Poly Film (保利影业), Huashi Entertainment (华视娱乐), Qixiang Culture (奇想文化) and Jiuziyuan AI (九紫源AI).
The platform also says it embedded screenwriter and director agents: the writing agent parses a script, splits scenes, tags characters, and drafts shot lists; the director agent plans camera movement and spatial logic from a film-industry knowledge base. It leans on an adapted Seedance 2.0 model for character skin, hair, and clothing texture.
Why 360 is the right parent
Consistency at scale is partly a compute and stability problem, and 360 brings two decades of internet infrastructure, security, and — crucially — GPU resources. For a tool whose selling point is "it won't crash or drift mid-season," the parent's uptime reputation matters as much as the algorithm. The subsidiary framing also signals 360 treating generative media as a strategic agent (智能体) bet rather than a side experiment.
The harder truth about "industrial-grade"
The comic-drama market is overcrowded and the formats are cheap to imitate. A tool that genuinely holds faces and scenes steady across 100 episodes is valuable precisely because most rivals cannot. But "industrial-grade" is a claim the whole sector now makes; the differentiator will be whether delivered episodes actually pass platform review and earn views, not whether a demo reel impresses.
Zhiju's own cited case studies — a guofeng short (《霍去病》) said to exceed 500 million plays, and a 《全民诡异》 series past 1 billion cumulative plays — are promotional and unverified here. We include them only as the vendor's stated evidence, not as confirmed reach.
Trend reading
According to Li An, chief scientist at Pinewood Media Lab (松木媒体实验室), the winners in AI content tools will be the ones that remove rejection risk for commercial buyers, because "looks amazing" does not get a studio paid but "delivers on spec" does.
Honest limitations
- Every performance metric (92% consistency, 2-pixel deviation, 3–5x output, 30–60 min/episode) is platform-published and unverified by an independent benchmark.
- The "1,000 studios / 80% of top MCNs" reach and the named adopters come from Chinese business-press coverage of vendor claims, not audited customer lists.
- The viral播放 (play) figures for 《霍去病》 and 《全民诡异》 are promotional and unconfirmed; do not treat them as measured audience data.
- Comic-drama (漫剧) itself is a volatile, regulation-sensitive format in China; a tool's quality does not shield customers from platform or content-policy shifts.
- Seedance 2.0 dependency means Zhiju's ceiling is partly set by another company's model roadmap.
What readers can do now
- Judge by a delivered episode, not a sizzle reel: if you evaluate this or any comic-drama tool, ask for a 10-episode sample with the same lead face held across all of them.
- Pressure-test the consistency claim: render the same character in three angles and measure face and background drift yourself rather than trusting the 92% figure.
- Watch the parent, not just the product: 360's infrastructure commitment is the real signal of whether "industrial-grade" survives past the launch quarter.