Expired50% confidenceFactExact time
Current mainstream video generation models such as CDance 2.5, MiniMax, Wanxiang 3, and Kling all have maximum generation duration limits (e.g., 15 seconds or 30 seconds), so long-form finished videos require multiple unit generations to be spliced together.
1
Sources
50%
Confidence
Short-term (~14 days)
Relevance
9/11/2026
First Seen
This claim has expired. It is kept for the record and no longer describes the current state.
Valid until: 9/25/2026(expired)
Sources
Related Entities
Related Claims
VerifiedWith the accelerator enabled, MiniMax H3 takes approximately 2 minutes to generate 1 second of video, approximately 14 minutes for a 480P ten-second video, and approximately 20–30 minutes for a 768P ten-second video.74% similarUnverifiedThe author states that this long-video workflow is not intended to replace the existing 30-second solution, as generating 60-second videos on a 12 GB GPU is more time-consuming.74% similarUnverifiedVideo real-time processing typically requires completing single-frame processing within 33 milliseconds, whereas VLM inference latency far exceeds this requirement, and a pure VLM solution may be 10 to 100 times slower than real-time.74% similarUnverified传统制作中，为口播视频手动添加文字包装单条至少需要10-15分钟73% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/908171API
curl https://kongchang.com/api/v1/knowledge/claims/908171MCP
get_claim(id=908171)