Video and Image Generation: Sora, Veo, Kling, Flux — 2026-10-02
Kuaishou's Kling 4.0 launched this week, doubling max clip length to 30 seconds and introducing multi-reference frame control. Meanwhile, a federal appeals court ruled against AI fair use in training for the first time, creating legal headwinds for all video generators. Gemini Omni Flash and Wan 3.0 lead latest arena rankings for image-to-video quality.
Video and Image Generation: Sora, Veo, Kling, Flux — 2026-10-02
Top developments
Kling 4.0 Doubles Maximum Clip Length to 30 Seconds
Kuaishou's Kling released version 4.0 on September 29, 2026, marking a major competitive push against ByteDance's Seedance. The update doubles the maximum single-clip generation from 15 to 30 seconds and introduces multi-reference frame support, allowing directors to guide motion with keyframe anchors. The model now accepts over a dozen reference input types for tighter creative control.
This represents Kuaishou's fastest iteration cycle in response to ByteDance's market dominance, with industry observers noting the shift toward 30-second clips as the new commercial standard for ad and social content.
Federal Court Rules Against AI Fair Use in First Appeal—Impact on Training Data
On September 30, 2026, a U.S. federal appeals court issued the first precedent-setting ruling against AI training as fair use, significantly narrowing the legal defense available to generative AI companies. The decision increases legal exposure for video and music generators that relied on copyrighted training data without licenses.
Suno (music generation) faces two new lawsuits from Sony and UMG filed within days of its v6 release, as the company confirmed training on user-generated interactions. Independent artists also won the right to sue without identifying specific infringing outputs. The ruling creates immediate precedent for video generators: Runway faces a pending class action from YouTubers alleging misuse of platform content.
Gemini Omni Flash Tops Image-to-Video Rankings; Kling 3.0 Holds Proprietary Crown
As of October 2, 2026, Artificial Analysis ranks Gemini Omni Flash at Elo 1,369 for image-to-video generation, followed by Bach 1.0 Pro (1,362) and Wan 3.0 (1,362). Among closed-source commercial models, Kling 3.0 remains at Elo 1,247 with native 4K output and multi-shot consistency, holding second place in the proprietary segment.
Veo 3.1 and MiniMax H3 occupy adjacent positions, with pricing favoring Kling at $0.25–$0.50 per 5-second clip versus Sora's legacy rates of $1.50–$2.50 and Runway's $0.75–$1.75 per 5-second clip.
LTX-2.5 Scales to 4K with Native Audio; Open-Weight Video Models Gain Traction
Lightricks' LTX-2.5 generates video and audio synchronously up to 4K resolution at $0.09 per second (Fast, 720p) and $0.12 (Pro, 720p). Critically, weights are open-source, enabling on-premises deployment—a competitive advantage as copyright litigation escalates.
Open-weight alternatives including HunyuanVideo (Tencent), CogVideoX, and Mochi-1 now use Diffusion Transformer (DiT) architectures matching proprietary leaders. HunyuanVideo combines a 3D causal VAE with DiT, achieving 16× spatial and 4× temporal compression while maintaining smooth motion without enterprise-grade hardware.
Local view
Chinese media and industry: Sina and Sohu reported Kling 4.0 as a critical response to ByteDance's market leadership, with Kuaishou positioning the 30-second clip as the "new production standard" for TikTok and Douyin content. Zhihu contributors noted the competitive timeline: Kling 3.0 launched mid-2026, Kling 4.0 nine months later, compared to Seedance's faster iteration. The model's support for 10+ reference types mirrors director-feedback workflows common in Asian short-form studios.
Copyright Alliance September Summary: The Copyright Alliance flagged September 2026 as marked by "significant congressional movements" on copyright-related legislation ahead of mid-term recess, with two pending House and Senate bills addressing generative AI training oversight.
Context & numbers
Pricing leaders (per 5-second clip or per-second equivalents):
- Kling 3.0: $0.25–$0.50 per clip
- Runway Gen-3: $0.75–$1.75 per clip
- Sora (legacy): $1.50–$2.50 per clip
- LTX-2.5: $0.09–$0.12 per second
Arena rankings (Elo, as of Oct 2):
- Gemini Omni Flash: 1,369 (image-to-video)
- Bach 1.0 Pro: 1,362
- Wan 3.0: 1,362 (open-weight contender)
- MiniMax H3: 1,357
- Kling 3.0: 1,247 (proprietary leader for 4K native output)
Clip length standards: 30 seconds is now baseline for commercial video generators (Kling 4.0, Seedance, Veo 3.1); most platforms generate in 1–5 minutes per request.
On the radar
- Copyright litigation calendar: Runway class action discovery phase expected Q4 2026; Suno dual-track Sony/UMG lawsuits likely to set precedent for video generators by mid-2027.
- Open-source acceleration: HunyuanVideo and Mochi-1 inference optimization ongoing on Hugging Face; watch for 8K or 60fps variants targeting enterprise workflows in November–December.
- Sora API sunset completed: OpenAI shuttered Sora API September 24, 2026. Migration tracker reports show Veo 3.1 and Kling 3.0 absorbing legacy Sora API users; adoption curves suggest 60–70% migration to Kling by Q4.
Data freshness note: All developments reported after 2026-09-25. Arena rankings and leaderboard snapshots valid as of 2026-10-02.
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
