Pika发布H3 Flex开源模型：六步生成视频与同步音频
Pika-Labs/Pika-H3-Flex: Pika H3 Flex: video and audio generation inference
Pika H3 Flex开源提供了六步快速生成带同步音频视频的能力，对追求创作效率的AI视频开发者极具参考价值，建议关注其多模态参考输入机制。
Pika H3 Flex: video and audio generation inference
Pika H3 Flex：视频与音频生成推理
GitHub repository created 2026-10-29 · 1136 stars in 5.0 days (about 227 a day) · Python
GitHub 仓库创建于 2026-10-29 · 5.0 天内获得 1136 颗星（日均约 227 颗）· Python
Pika H3 Flex Built for professional iteration. Six-step generation · Omni-reference · Native 480p / 768p · Up to 4K via Pika API
Pika H3 Flex 专为专业迭代打造。 六步生成 · 全参考模式 · 原生支持 480p / 768p · 通过 Pika API 最高支持 4K
Try it now · Model weights · Watch samples · Quickstart · Prompt guide · Pika API extensions
立即试用 · 模型权重 · 观看示例 · 快速入门 · 提示词指南 · Pika API 扩展
Pika H3 Flex is a post-trained H3 model, designed to give professional creators the flexibility and speed to explore with precision. Optimized for greater control, it generates video with synchronized audio in six sampling steps and supports up to 12 reference inputs, with shot direction through references and prompts.
Pika H3 Flex 是一个经过后训练的 H3 模型，旨在为专业创作者提供灵活性和速度，以实现精准探索。该模型针对更高的可控性进行了优化，可在六个采样步骤内生成带有同步音频的视频，并支持多达 12 个参考输入，通过参考内容和提示词控制镜头方向。
Pika’s hosted API extends this workflow with 1080p, 2K and 4K video generation, audio stems and prompt enhancement. Local open-weight inference supports native 480p / 768p video with mixed stereo audio.
Pika 托管的 API 在此基础上扩展了工作流程，支持 1080p、2K 和 4K 视频生成、音频分离以及提示词增强。本地开源权重推理支持原生 480p / 768p 视频及混合立体声音频。
Ideal for multi-shot storytelling · Motion graphics · Product films · Reference-guided animation
适用于多镜头叙事 · 动态图形设计 · 产品宣传片 · 参考引导动画
Watch the model
观看模型演示
Explore camera choreography, cinematic storytelling, graphic design and reference-guided animation. Click a preview to watch with sound.
探索镜头调度、电影叙事、平面设计以及参考引导动画。点击预览图可带声音观看。
Frog table-tennis rallyCamera choreography▶ Watch with audio A journey between balconiesText-to-video camera control▶ Watch with audio
青蛙乒乓球对打镜头调度▶ 带声音观看 阳台间的旅程文本到视频镜头控制▶ 带声音观看
The birthday-cake heistMulti-shot storytelling▶ Watch with audio MOVE: kinetic typographyMotion graphics▶ Watch with audio
生日蛋糕劫案多镜头叙事▶ 带声音观看 MOVE：动态排版动态图形设计▶ 带声音观看
Feel the bassProduct film▶ Watch with audio Record-shop danceReference-guided motion▶ Watch with audio
感受低音产品宣传片▶ 带声音观看 唱片店舞蹈参考引导运动▶ 带声音观看
Prompts and input assets for these six showcases
这六个展示案例的提示词和输入素材
These are creative showcases, not release-validation renders or throughput benchmarks. Exact reproduction with the released package is not guaranteed.
这些是创意展示，并非发布验证渲染或吞吐量基准测试。使用发布的包进行精确复现不保证成功。
System overview
系统概览
Pika H3 Flex is one video-and-audio model family with two task adapters: FL2VA for text and endpoint-frame conditioning, and Ref2VA for multimodal references. It builds on MiniMax-H3 and generates video with synchronized stereo audio in six sampling steps.
Pika H3 Flex 是一个视频与音频模型家族，包含两个任务适配器：用于文本和端点帧条件控制的 FL2VA，以及用于多模态参考的 Ref2VA。它基于 MiniMax-H3 构建，可在六个采样步骤内生成带有同步立体声音频的视频。
The inference pipeline encodes the prompt and input media, generates video and audio latents with the selected Pika H3 Flex adapter, then decodes and combines them into an MP4. The package downloads the pinned base components and adapter automatically. Use the prompt guide to describe scenes, motion, dialogue, ambience and music; optional prompt enhancement is available through the Pika API.
推理流程首先编码提示词和输入媒体，然后使用选定的 Pika H3 Flex 适配器生成视频和音频潜在表示（latents），最后对其进行解码并组合成 MP4 文件。该包会自动下载固定的基础组件和适配器。请使用提示词指南来描述场景、动作、对话、环境音和音乐；可通过 Pika API 使用可选的提示词增强功能。
Input and output specifications
输入与输出规范
| Category | Specification | ||
|---|---|---|---|
| Input modalities | Text; first/last images; image, video and audio references, depending on the selected mode | ||
| Native output resolution | 480p / 768p across all 11 supported aspect ratios; exact pixel dimensions follow the selected aspect-ratio bucket | ||
| Output duration | Requested integer durations of 5–15 seconds; actual length follows the model frame grid | ||
| Output aspect ratios | 21:9, 16:9, 3:2, 4:3, 5:4, 1:1, 4:5, 3:4, 2:3, 9:16, 9:21; dimensions use the packaged resolution buckets | ||
| Output frame rate | 24 fps | ||
| Output audio | 32 kHz stereo, mixed into the generated video | ||
| Output files | MP4 with H.264 video and AAC audio, plus JSON generation metadata | ||
| Prompt input | Text or a UTF-8 prompt file; 1–8,000 characters | ||
| Sampling | Six generator evaluations using the released adapter's sampling schedule |
| 类别 | 规格 | ||
|---|---|---|---|
| 输入模态 | 文本；首帧/末帧图像；图像、视频和音频参考，具体取决于所选模式 | ||
| 原生输出分辨率 | 480p / 768p，涵盖所有 11 种支持的宽高比；确切像素尺寸遵循所选的宽高比分桶 | ||
| 输出时长 | 请求的整数时长为 5–15 秒；实际长度遵循模型帧网格 | ||
| 输出宽高比 | 21:9, 16:9, 3:2, 4:3, 5:4, 1:1, 4:5, 3:4, 2:3, 9:16, 9:21；尺寸使用打包的分辨率分桶 | ||
| 输出帧率 | 24 fps | ||
| 输出音频 | 32 kHz 立体声，混合到生成的视频中 | ||
| 输出文件 | MP4（含 H.264 视频和 AAC 音频），以及 JSON 生成元数据 | ||
| 提示词输入 | 文本或 UTF-8 提示词文件；1–8,000 个字符 | ||
| 采样 | 使用已发布适配器的采样计划进行六次生成器评估 |
Durations are rounded up to the supported frame grid: a 5-second request produces 124 frames (about 5.17 seconds), and a 15-second request produces 362 frames (about 15.08 seconds). Native resolution and aspect ratio jointly determine the actual canvas dimensions.
时长会向上舍入至支持的帧网格：5 秒的请求产生 124 帧（约 5.17 秒），15 秒的请求产生 362 帧（约 15.08 秒）。原生分辨率和宽高比共同决定实际的画布尺寸。
Generation modes and reference limits
生成模式和参考限制
| Mode | Inputs | Adapter / CLI task | ||
|---|---|---|---|---|
| Text to video and audio | Prompt, without input media | FL2VA / `t2va` | ||
| First-frame conditioning | Prompt and one opening frame | FL2VA / `i2va` | ||
| Last-frame conditioning | Prompt and one ending frame | FL2VA / `l2va` | ||
| First-and-last-frame conditioning | Prompt and both endpoint frames | FL2VA / `fl2va` | ||
| Multimodal references | Prompt and ordered image/video references, optionally with audio references | Ref2VA / `ref2va` |
| 模式 | 输入 | 适配器 / CLI 任务 | ||
|---|---|---|---|---|
| 文生视频和音频 | 提示词，无输入媒体 | FL2VA / `t2va` | ||
| 首帧条件控制 | 提示词和一个起始帧 | FL2VA / `i2va` | ||
| 末帧条件控制 | 提示词和一个结束帧 | FL2VA / `l2va` | ||
| 首尾帧条件控制 | 提示词和两个端点帧 | FL2VA / `fl2va` | ||
| 多模态参考 | 提示词和有序的图片/视频参考，可选带音频参考 | Ref2VA / `ref2va` |
Ref2VA accepts up to 9 images, 3 videos and 3 standalone audio clips, with 12 reference files total. At least one image or video is required. Each video or standalone audio clip must be 2–15 seconds; combined video duration and combined standalone-audio duration must each be at most 15 seconds. A video's embedded soundtrack stays with that video and does not count as an additional standalone audio file.
Ref2VA 最多接受 9 张图片、3 个视频和 3 个独立音频片段，总共 12 个参考文件。至少需要一张图片或一个视频。每个视频或独立音频片段的时长必须为 2–15 秒；视频总时长和独立音频总时长各自不得超过 15 秒。视频内嵌音轨随视频保留，不计为额外的独立音频文件。
These are accepted input limits, not a guarantee that every maximum-size combination fits GPU memory. The package runs one request on one GPU, with optional CPU offload. See hardware requirements and tested configurations for representative H100 results and memory guidance.
这些是接受的输入限制，并不保证每种最大尺寸的组合都能适应 GPU 内存。该包在单个 GPU 上运行单个请求，并支持可选的 CPU 卸载。请参阅硬件要求和经过测试的配置，以获取代表性的 H100 结果和内存指导。
Pika API extensions
Pika API 扩展
Extend your generation workflow with Pika’s hosted endpoints:
通过 Pika 的主机端点扩展您的生成工作流：
| Feature | Support | Endpoint | ||
|---|---|---|---|---|
| Higher-resolution video | Generate 1080p, 2K and 4K video from a draft video | Draft to video | ||
| Audio stems | Obtain separate audio stems | Audio stems | ||
| Prompt enhancement | Enhance generation prompts | Prompt enhancement |
| 功能 | 支持情况 | 端点 | ||
|---|---|---|---|---|
| 高分辨率视频 | 从草稿视频生成 1080p、2K 和 4K 视频 | 草稿转视频 | ||
| 音频分离 | 获取独立的音频轨道 | 音频分离 | ||
| 提示词增强 | 增强生成提示词 | 提示词增强 |
These optional hosted services are separate from the downloadable adapters and local inference package. The specifications above describe local model inference; see each endpoint’s documentation for its API capabilities and usage.
这些可选的托管服务与可下载的适配器和本地推理包是分开的。上述规格描述的是本地模型推理；有关其 API 功能和用法，请参阅每个端点的文档。
Install
安装
The command and Python distribution are named `pika-h3-pro`.
命令和 Python 发行版的名称为 `pika-h3-pro`。
Clone the repository, then enter its directory:
克隆仓库，然后进入其目录：
For pip installation, install system `git`, `patch`, `ffmpeg` and a C compiler with development headers first (for example, `build-essential` on Debian/Ubuntu). Triton needs the compiler at runtime. Docker and Conda recipes supply these prerequisites.
对于 pip 安装，请先安装系统级的 `git`、`patch`、`ffmpeg` 以及带有开发头文件的 C 编译器（例如 Debian/Ubuntu 上的 `build-essential`）。Triton 在运行时需要编译器。Docker 和 Conda 配方提供了这些先决条件。
Use Python 3.12, an NVIDIA GPU and a compatible driver. Representative cases are tested on one H100 80GB per request; H200 and B300 remain untested. Longer clips and large reference sets can exceed GPU memory. Downloading both tasks acquired approximately 210 GB of base files plus 5.5 GB of adapters; allow additional space for dependencies and download staging. See tested cases and memory measurements.
使用 Python 3.12、NVIDIA GPU 及兼容的驱动程序。代表性用例在每个请求基于单块 H100 80GB 进行测试；H200 和 B300 尚未测试。较长的片段和大型参考集可能会超出 GPU 内存限制。下载这两个任务共获取了约 210 GB 的基础文件加上 5.5 GB 的适配器；请为依赖项和下载暂存预留额外空间。请参阅已测试的用例和内存测量数据。
更进一步：量化金融体系
看懂新闻只是起点——沿量化金融路径，把它变成能交付的工程能力