Kling AI made Kling 4.0 fully available on Oct 15, confirmed by a Kuaishou press release on Oct 16. It natively generates up to 30 seconds per pass; Omni Reference combines up to 15 assets (≤10 images, 5 videos, 7 subjects) with up to 10 keyframes, 8,000-token prompts and 21:9 output. Kling 4.0 is also wired into Kling MCP and CLI so AI agents can call it. 10-bit HDR at 4K/1080p and enterprise/developer API access are marked "coming soon". No quantitative benchmarks were published.
Key Takeaways
- ✓Native 3–30 s generation per pass; video extension up to 2 minutes is marked coming soon
- ✓Omni Reference: up to 15 assets per task — ≤10 images, 5 videos (≤30 s total), 7 subjects
- ✓Up to 10 keyframes anywhere on the timeline; prompts up to 8,000 tokens
- ✓Available via MCP and CLI (
npm i -g @klingai/cli-global); agents can call omni_ref_video and more - ✓10-bit HDR (4K/1080p) and enterprise/developer API are still coming soon; no quantitative benchmarks published
Key Decision Metrics at a Glance
Turn your technical choice into a development budget
Compare 40 dev plans & simulate token costs vs $20/mo subscriptions
Project Links & Resources
Direct AccessIn-Depth Technical Analysis
Background and Pain Points
AI video competition has shifted from "can it generate" to "can it reliably deliver controllable key shots." Kuaishou’s Oct 16 press release names jittery large motion, complex camera moves and inconsistent characters as the main pain points. Kling 4.0 entered early access on Sep 30 and became fully available on Oct 15.
Architecture Highlights
Per the vendor, three upgrades: steadier large-scale motion and tracking/orbit shots; native single-pass generation up to 30 seconds; and Omni Reference, combining up to 15 assets per task (≤10 images, 5 videos, 7 subjects, with an optional voice per subject) plus up to 10 keyframes. Prompts now reach 8,000 tokens, with 21:9 output and stereo audio.
Benchmarks
No quantitative benchmarks (e.g. VBench) or numeric comparisons with Veo or Sora were published; the release lists specifications only. 10-bit HDR at 4K and 2-minute extension remain "coming soon". All figures are vendor-reported.
Developer Hands-on Guide
The developer angle is agent access: Kling 4.0 is wired into Kling MCP and the official CLI. Install with npm i -g @klingai/cli-global, run kling login, then kling who_am_i to list models and parameters; submit with text_to_video or omni_ref_video and poll with query_tasks. The official skills repo ships a SKILL.md agents can use directly. Enterprise API access is not open yet, so check credit costs before batch runs.
Benchmark side-by-side against alternatives, or calculate monthly token cost vs subscription break-even.
Discussion & Comments
0Sign in to join the discussion
Connect with AI developers to exchange benchmark insights.