Around China's National Day holiday in 2026, Kuaishou's Kling AI (Kling 4.0) and Shengshu's Vidu Q4 Preview rolled out in quick succession: 4K output and longer clips, while entry prices pushed from roughly "yuan per second" toward "fen per second." AI video is shifting from lottery-style demos to costed production tooling—the race is now "good and cheap."
What happened
– Sept 28: Kling 4.0 entered closed beta—up to ~30s native clips, 4K, plus a Kling 4.0 Flash variant pitched for speed and value (full API pricing not yet public). – Oct 8: Vidu Q4 Preview launched with up to 15 reference images, 3 reference audio tracks, 4K out, and a launch promo from about RMB 0.09/sec. – Earlier context: MiniMax H3 helped set a "floor price" wave from July; ByteDance Seedance continues to push flagship capability (and sometimes unit price).
Reported Kling Q3 API ballparks: ~RMB 0.9/s at 720P, ~1.2/s at 1080P, ~3/s at 4K—Flash is expected to cut the entry rung further. Vidu SaaS has limited-time deals into fen/sec territory; MaaS image-to-video sits near ~0.6/s (720P) and ~0.75/s (1080P), close to or below prior Q3 Pro (~0.63/s).
Seedance leans the other way on flagship: Seedance 2.0 ~RMB 46 per million tokens (720P, no reference gen) vs Seedance 2.5 ~RMB 70 for the same class—about +50% vs the prior generation.
Why price wars only now
A year ago, "gacha" waste dominated: many generations per usable clip, so sticker-price cuts barely lowered true cost. Per a Bank of China Securities note, Seedance 2.0 one-shot success rates reach as high as ~90%, versus a prior industry average near ~20%. Once quality crosses the usable line, unit price becomes a real lever—echoing the LLM "DeepSeek moment": near-frontier ability at far lower cost, not cheapness alone.
Buyers are showing up. Reports cite Yiwu accessory makers running AI-styled videos overseas with million-plus views and ~45% higher inquiry conversion; some teams say higher-value models can bring a short AI film in around RMB 8,000, versus hundreds of thousands for live-action shorts.
Three paths: Seedance / Kling / Vidu
China's top tier is diverging—but all are moving from "asset generation" to "deliverable cuts". Browse this site's AI video tools for product entries; for pricing and substitute dynamics see also Seedance 2.5 vs substitutes.
| Product | Focus | This round | |———|——-|————| | ByteDance Seedance | Long-form narrative, person realism, controllable edits | ~30s clips, timestamp-level local edit; C-end via Doubao / Jimeng | | Kuaishou Kling | Cinematic look, complex camera & action | 4.0 for fight scenes / tracking / orbits; 4K & wide formats | | Shengshu Vidu | Short drama / comic drama "director feel" + value | Q4 emotion granularity, voice-reference consistency; from RMB 0.09/s |
In reported couple-argument animation tests, Seedance 2.0 nailing micro-actions; Vidu Q4 leaning Japanese comedy-anime look. Vidu Q4 pushes silent tension (gaze, pauses) and up to 3 audio refs for stable character timbre; Kling 4.0 stresses lip-sync. On camera/VFX, Kling owns high-dynamic combat; Vidu strengthens tracking, orbits, FPV, and effects that light/smoke the surrounding world.
Overseas: Google Veo 3.1 Fast (with audio) ~$0.15/s, full ~$0.4/s; Veo 3.1 Lite (720P) ~$0.05/s—price-competitive with China's mid/high tier, still behind on capability. On OpenRouter-style usage boards, lite SKUs (Veo 3.1 Lite, Seedance 2.0 Fast) are becoming volume workhorses.
What it means for creators
Lower cost enables multi-cut A/B for ads and solo auteurs shipping team-level work. Case: AI short *The Girl Who Got Laid Off*—personal creator, Douyin past 200M views in July after a June launch; two seasons estimated over 500M across platforms. Content buyers now score emotion, narrative, and finish quality—models are entering real production workflows, not just demos.
FAQ
Q: What is the AI-video "DeepSeek moment"? A: Near-top usable quality plus a sharp price drop that makes scaled production viable—not a race to the cheapest broken output.
Q: How much is Kling 4.0? A: Full API pricing isn't public yet; use Q3 (~RMB 0.9–3/s) as a reference; Flash is positioned as faster and cheaper.
Q: Will short-drama teams always use the priciest flagship? A: Practitioners still reserve stronger models for top scripts and need affordable standardized throughput elsewhere—hence fen-level SKUs.
Source: https://36kr.com/p/4025642512896130
相关文章
