Video and Image Generation: Sora, Veo, Kling, Flux — 2026-10-03
Kuaishou's Kling AI released version 4.0 this week, doubling maximum clip length to 30 seconds and adding multi-reference support, intensifying competition with ByteDance and Google's models. A landmark federal appeals court ruling against AI training as fair use reshapes copyright liability across the entire sector, while compliance gaps widen as adoption surges across studios and platforms.
Video and Image Generation: Sora, Veo, Kling, Flux — 2026-10-03
Top developments
Kling 4.0 Launches with 30-Second Clips and Multi-Reference Control
Kuaishou's Kling AI released version 4.0 on September 29, 2026, doubling the maximum single-clip duration from 15 to 30 seconds and adding support for multiple reference inputs—a significant leap in shot consistency and director-like control. The update aims to close the gap with ByteDance's Jingeme and Google's Veo 3.1, both of which have captured market share since Sora's API shutdown in late September. Kling 4.0 maintains native 4K resolution support and competitive pricing relative to rivals.
Federal Appeals Court Blocks Fair Use Defense for AI Training
A federal appeals court ruled on September 30, 2026, against AI training as a fair use exception—the first appellate decision of its kind. The ruling directly affects video and music generators' legal exposure for unlicensed training data. While the decision centered on music AI (Suno), legal experts warn similar claims will extend to video platforms like Runway, which faces an active class action lawsuit from YouTubers over unauthorized training on video content. This represents a watershed moment: companies can no longer rely on transformative use arguments to shield training pipelines from copyright liability.
Copyright and Compliance Gaps Widen as Studio Adoption Accelerates
G2 analyzed 10,900+ verified AI video generator reviews and found significant compliance and licensing uncertainties among the top platforms. A new ScreenWeaver report (October 2, 2026) noted that while platform terms claim user ownership of outputs, copyright law assigns no authorship to generated frames—creating ambiguity around registration, insurance, and liability for commercial producers. As studios integrate Kling, Runway, and Veo into production workflows, these gaps are becoming dealbreakers for enterprise customers.
LTX Video 2.5 (Lightricks) Offers Open-Weight Alternative with Native 4K and Audio
Lightricks' LTX-Video 2.5 released last week with open weights, generating 4K video and synchronized audio at competitive API pricing: $0.09 per second (Fast, 720p) and $0.12 (Pro, 720p). The open-weight release contrasts sharply with proprietary vendors and appeals to teams running inference on private infrastructure. LTX-Video's diffusion backbone compresses video 16x spatially and 4x temporally, maintaining motion quality without enterprise-grade hardware.
Image Generation Speed: Flux Schnell vs. Midjourney Trade-off
Black Forest Labs' Flux Schnell generates images in 0.5–2 seconds, the fastest option for real-time applications, while Midjourney's API output ranges 10–30 seconds per image. Flux Pro (8–12 seconds via API) sits between the two. This speed hierarchy is reshaping use cases: Schnell dominates rapid-iteration workflows; Midjourney retained for premium quality.
Local view
China: Sina.cn (October 2, 2026) reported ByteDance's Douyin ecosystem integrating Jingeme AI video tools directly into creator dashboards, accelerating adoption vs. Kling in short-form video. Zhihu contributors noted that Kling 4.0's price-to-performance ratio now favors bulk ad production, where sub-30-second clips dominate. Tencent's HunyuanVideo (open-weight) remains used primarily by research teams and small studios unable to afford API costs.
Context & numbers
- Kling 4.0 specs: 30-second max duration (up from 15), native 4K, multi-reference inputs, competitive API pricing (below disclosed thresholds for Veo 3.1 and Runway Gen-3)
- Flux Schnell: 0.5–2 seconds per image (real-time tier)
- LTX-Video 2.5 pricing: $0.09–$0.12 per second (open weights available)
- Arena leaderboard (Kling 3.0): Elo 1,247 (text-to-video), ranked T2 among reliably available proprietary models as of April 2026
- G2 compliance survey: 10,900+ AI video generator reviews; widespread gaps in licensing clarity and commercial indemnification
- Fair use precedent: September 30 federal appeals ruling (against Suno, applicable to video platforms) eliminates transformative-use defense for training on copyrighted material
On the radar
- Runway class action: YouTuber lawsuit over unauthorized training data continues discovery phase; outcome will set precedent for video platform liability (watch for January 2027 trial dates)
- EU/UK copyright policy: Regulatory bodies monitoring fair use rulings; potential rulemaking expected Q4 2026 to clarify AI training exceptions under GDPR and UK Online Safety Bill
- Open-weight race: Tencent HunyuanVideo, CogVideoX (Zhipu), and Mochi (Genmo) actively recruiting academic and indie studios; watch for licensing clarity announcements that could undercut proprietary pricing
This content was collected, curated, and summarized entirely by AI — including how and what to gather. It may contain inaccuracies. Crew does not guarantee the accuracy of any information presented here. Always verify facts on your own before acting on them. Crew assumes no legal liability for any consequences arising from reliance on this content.
Powered by
