LesenForschungRadarAnlageframework
Anmelden / Registrieren
← Zurück zur Lektüre
NewsThema · Pika H3 FlexWesentlich1 Min. LesezeitDisconfirmAI Redaktion

Pika Labs Releases Open-Weights Video Model Pika H3 Flex, Claiming #1 Among Open Models on DesignArena Leaderboard

Pika Labs has launched Pika H3 Flex, an open-weights video generation model post-trained on MiniMax H3. The model supports multi-shot storytelling, complex camera moves, and audio stem exports. Officially, it ranks #1 among open-weights models on DesignArena’s text-to-video leaderboard and can generate a 15-second 768p clip with synchronized audio in as little as seven seconds.

2026-10-30 02:10 GMT+8

Pika Labs officially released its new video generation model, Pika H3 Flex, on October 29, 2026. This open-weights model is post-trained on the MiniMax H3 architecture and is designed to give creatives greater speed and control during iteration. According to the official description, the model allows users to bring up to 12 image, video, and audio references, direct complex camera moves, build multi-shot stories, and explore new visual looks.

In terms of performance metrics, Pika Labs claims that Pika H3 Flex can generate a 15-second clip with synchronized audio at 768p resolution in as little as seven seconds. Additionally, the model supports exporting five separate audio stems and finishing videos in up to 4K resolution. The company also states that the model debuted at #1 among open-weights models and #7 overall on DesignArena’s text-to-video leaderboard.

As an open-source project, code and setup instructions for Pika H3 Flex are available on GitHub, while the model weights are hosted on Hugging Face. This release means developers can not only use the model for content creation but also build upon and fine-tune the model itself. Although specific benchmark methodologies and independent reproduction results were not detailed in the announcement, this product release provides a new open-source option and technical reference points for the AI video generation field.

Quellen

Anmelden / Registrieren
LesenForschungRadarAnlageframework
  • youtube.com↗
Thema

Pika H3 Flex

Pika Labs于2026年10月29日正式发布开源视频生成模型Pika H3 Flex。该模型基于MiniMax H3进行后训练,通过蒸馏将采样步骤从50步减少至6步,实现了比实时更快的视频生成速度(官方称7秒内可生成15秒768p带音…

3 Schritte1 BeiträgeAktualisiert 2026-10-31 18:33

Wie es dazu kam

  1. 2026-10-29Pika Labs发布博客文章《Introducing Pika H3 Flex》,正式推出基于MiniMax H3后训练的开源视频模型,宣称其在DesignArena榜单中开源排名第一。pika.art ↗
  2. 2026-10-29Pika Labs发布技术深度解析《Pika H3 Flex Technical Deep Dive》,详细说明六步生成机制、音频分离技术及与竞品(如HeyGen, Seedance)的价格和能力对比。pika.art ↗
  3. 2026-10-30第三方媒体或社区报道Pika H3 Flex发布,确认其支持多镜头故事构建及复杂运镜,并再次引用DesignArena排名数据。youtube.com ↗

Was noch ungeklärt ist

  • 本地推理在消费级GPU上的实际显存占用和生成延迟表现如何?
  • 五轨音频分离功能在非API托管环境下的可用性及质量稳定性?
  • 相比基础版MiniMax H3,Pika H3 Flex在特定艺术风格或物理模拟场景下的具体优势边界在哪里?
Vollständiges Thema lesen →
Recherche fortsetzen →