原创 SiliconCloud 2025-08-15 20:59 北京
电影级质感视频生成效果。
硅基流动大模型云服务平台 SiliconCloud 已上线阿里通义万相团队最新开源的视频生成基础模型 Wan2.2,包括文生视频模型 Wan2.2-T2V-A14B、图生视频模型 Wan2.2-I2V-A14B,价格均为 2 元 / Video。
该系列模型是业界首个基于 MoE 架构的视频模型,总参数量为 27B,激活参数 14B。Wan2.2 首创“电影级美学控制系统”,将光影密码、构图法则、色彩心理学编码成 60 多个直观的参数,将光影、色彩、镜头语言装进生成模型,实现电影级质感视频生成。从实际生成效果看,Wan2.2 无疑是当前成片质量最优的开源视频生成模型。
欢迎通过以下方式使用 Wan2.2,国内站与国际站新用户可分别自动获取 14 元或 1 美元赠金体验。
国内站在线体验
https://cloud.siliconflow.cn/models
国际站在线体验
https://cloud.siliconflow.com/models
第三方应用接入教程
https://docs.siliconflow.cn/cn/usercases/
开发者 API 文档
https://docs.siliconflow.cn/cn/api-reference/chat-completions/
模型效果
运动控制
提示词:On the wing of a high-altitude jet soaring through the sky, a gymnast clad in a red-and-white leotard advances slowly in a cat-walk posture, her hair and fluttering outfit swept back by the fierce wind. Suddenly, she leaps into the air, executing a flawless aerial cartwheel before landing steadily on the metal wing’s edge. Then, amid the howling gusts, she performs two consecutive aerial flips with twists, her arms slicing through the wind like windmills, tracing extreme arcs. Finally, she stabilizes herself on one foot, fingertips lightly grazing the wing’s edge, completing this impossible moment against the roar of the engines. The wide shot pulls back—the entire aircraft pierces through layered clouds, as if the whole world holds its breath for her focus and balance.
镜头特写
提示词:A man with graying hair, a beard, and a gray shirt looks down and to his right, then turns his head to the left. The camera angle is a close-up, focused on the man's face. The lighting is dim, with a greenish tint. The scene appears to be real-life footage.
人物情绪
提示词:In the dark room, only the faint glow of a phone screen illuminated the face of a young woman. Her pupils were dilated with sheer terror, and her lips parted slightly, yet no sound escaped. Cold beads of sweat trickled from her temples, sliding slowly down her rigid cheeks.
视效调控
提示词: A purely visual and atmospheric video piece focusing on the interplay of light and shadow, with a corn train as the central motif. Imagine a stage bathed in dramatic, warm spotlights, where a corn train, rendered as a stark silhouette, moves slowly across space. The video explores the dynamic interplay of light and shadow cast by the train, creating abstract patterns, shapes, and illusions that dance across the stage. The soundtrack should be ambient and minimalist, enhancing the atmospheric and abstract nature of the piece.
基础运镜
提示词:The waves crash against the jagged rocks of the shoreline, sending spray high into the air. The rocks are a dark gray color, with sharp edges and deep crevices. The water is a clear blue-green, with white foam where the waves break against the rocks. The sky is a light gray, with a few white clouds dotting the horizon.
模型特点及性能
通义万相 2.2 率先在视频生成扩散模型中引入 MoE 架构,有效解决视频生成处理 Token 过长导致的计算资源消耗大问题。Wan2.2-T2V-A14B、Wan2.2-I2V-A14B 两款模型均由高噪声专家模型和低噪专家模型组成,分别负责视频的整体布局和细节完善,在同参数规模下,可节省约 50% 的计算资源消耗。模型能力上,通义万相 2.2 在复杂运动生成、人物交互、美学表达、复杂运动等维度上也取得了显著提升。
Wan2.2 主要引入了以下创新:
- 高效 MoE 架构:Wan2.2 将混合专家模型(MoE)架构引入视频扩散模型。该架构通过专业化且强大的专家模型对跨时间步的去噪过程进行解耦处理,在保持相同计算成本的同时显著提升了整体模型容量。
- 电影级美学呈现:Wan2.2 整合了精心筛选的美学数据,包含光线、构图、对比度、色调等详细标注标签。这使得电影风格生成更精准可控,便于创作符合自定义审美偏好的视频内容。
- 复杂动作生成:相较 Wan2.1 版本,Wan2.2 训练数据量大幅提升,图像数据增加 65.6%,视频数据增加 83.2%。这一扩展显著增强了模型在动作、语义和美学等多维度的泛化能力,在所有开源与闭源模型中实现了顶尖性能表现。
根据官方公布的新版 Wan-Bench 2.0 基准数据,Wan2.2 在关键指标上的性能表现比肩 Seedance1.0、Sora 等领先的商业模型。
Token 工厂 SiliconCloud
Qwen3-8B 等免费用
作为一站式大模型云服务平台,SiliconCloud 致力于为开发者提供极速响应、价格亲民、品类齐全、稳定丝滑的大模型 API。
除了 Wan2.2,SiliconCloud 已上架包括 GLM-4.5V、Step3、Qwen3-Coder、Qwen3-30B-A3B、MOSS-TTSD-V0.5、GLM-4.5、Qwen3-235B-A22B、Kimi K2 Instruct、Qwen3-Embedding & Reranker、DeepSeek-R1-0528、CosyVoice2 在内的上百款模型。其中,DeepSeek-R1 蒸馏版(8B、7B、1.5B)、Qwen3-8B 等多款大模型 API 免费使用,让开发者实现“Token 自由”。
开发者可在 SiliconCloud 平台自由对比、组合各类大模型,只需在开发应用时轻松调用更易用、高效的 API,为你的生成式 AI 应用选择最佳实践。
近期更新
SiliconCloud上线智谱GLM-4.5V
SiliconCloud上线阿里Qwen3-Coder
硅基流动上线DeepSeek-R1 & V3企业服务
爆改Gemini-CLI,用DeepSeek跑同款命令行
提效300%,大模型一体机加速药企报告撰写
