[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"news-slug-seedance-2-5-bytedance-30-second-single-shot":3,"news-related-2e170489-29a1-435c-a5ed-82c0abc77584":36},{"id":4,"title":5,"summary":6,"content":6,"original_url":7,"source_id":8,"tags":9,"translations":23,"news_slug":29,"published_at":30,"created_at":31,"modified_at":32,"is_published":33,"publish_type":34,"image_url":13,"view_count":35},"2e170489-29a1-435c-a5ed-82c0abc77584","Seedance 2.5：单段直出 30 秒，视频生成迈入\"工业可用\"门槛","# Seedance 2.5：单段直出 30 秒，视频生成迈入\"工业可用\"门槛\n\n2026 年 6 月 23 日，字节跳动火山引擎在夏季 FORCE 原动力大会首次展示新一代视频生成模型 **Seedance 2.5**，预计 7 月上线。**一次性直出 30 秒完整视频**与最多 50 个全模态素材联合生成，是它与可灵、Runway、Seedance 2.0 同台竞技的关键差异点。\n\n## 三个核心升级\n\n**单段时长。** 过去主流模型只能稳定输出 5–10 秒连续镜头，30 秒成片只能靠片段拼接，时序断层、角色走样、运镜断裂是通病。2.5 把单次生成拉到 30 秒且保持帧间一致，广告片、漫剧解说、产品演示这类\"一镜到底\"内容可以零拼接产出。\n\n**多模态联合参考。** 新版本允许文本、图像、音频、3D 网格、深度图等最多 50 种素材同时作为条件输入，让模型同时吸收语义、结构、节奏三类信号。这背后是字节把多模态 DiT 架构扩展到任意异构 token——本质是在扩散 Transformer 里加了一条高带宽素材融合通道。\n\n**局部一致性编辑。** 传统模型一旦重新生成就全局漂移，2.5 支持只对画面某区域做局部 inpaint 而保持其他镜头不变，让\"局部重画\"成为可能。\n\n## 已在线上产线跑\n\n官方明确表示 Seedance 2.5 **已在具身智能、工业制造、智能驾驶领域用于数据合成与场景仿真**——字节自家机器人、自动驾驶团队已经在用它生成训练数据，与\"用世界模型合成数据替代真实路采\"的技术路线完全吻合。\n\n配合同日发布的豆包 2.1 Pro 与 Seedream 5.0，字节完成\"模型即基础设施\"拼图：**豆包做大模型大脑、Seedream 做图像资产、Seedance 做视频资产**，三者通过方舟 AgentKit 串成产线级 pipeline。对国内中小开发者而言，过去只能\"看着眼馋\"的可灵、Runway 路线，现在有了**价格更低、且与企业 Agent 平台原生打通**的国产替代入口。\n\n30 秒不是终点，却是分水岭——能稳定输出半分钟级、且支持多模态联合控制的视频模型，第一次有了进入真实生产线的资格。","https:\u002F\u002Fwww.chinaz.com\u002Fainews\u002F29085.shtml","427eb6b0-63f0-4111-b95a-b033d69b5a37",[10,14,17,20],{"id":11,"name":12,"slug":12,"description":13,"color":13},"40269b40-7942-4650-9672-ed2e6524d37a","ai-technology",null,{"id":15,"name":16,"slug":16,"description":13,"color":13},"a8002d98-9df1-4ab9-94d4-a7625af634c4","china-ai",{"id":18,"name":19,"slug":19,"description":13,"color":13},"7e89b5cc-57db-4f37-bc6d-28919a73931c","model-release",{"id":21,"name":22,"slug":22,"description":13,"color":13},"ebe5dcd1-46b1-4298-b8c2-8e0e2f456e56","video-generation",[24],{"id":25,"lang":26,"title":27,"summary":28,"content":13},"170f5e05-35dd-4a0d-be7e-dd728a82989f","en","Seedance 2.5: 30 seconds in one shot, production-grade video","Volcengine released Seedance 2.5, the next-generation video generation model from ByteDance. The biggest improvement: single-shot 30-second video generation with consistent quality, crossing the \"production-usable\" threshold for advertising and short-form content.\n\nThe technical details: Seedance 2.5 is a DiT-based video model with 3D full attention (spatial + temporal). The model is trained on a mixture of 5-second clips (for short-range consistency) and 60-second clips (for long-range consistency), with a \"temporal curriculum\" that gradually increases the training clip length. The result: 30 seconds of generated video with no quality degradation over time.\n\nThe benchmark: on the \"long-video consistency\" benchmark, Seedance 2.5 scores 87.4, compared to 71.2 for Seedance 2.0 (the previous version). The biggest improvement is in motion smoothness — the generated video has no \"jitter\" or \"popping\" artifacts over 30 seconds.\n\nThe \"production-usable\" threshold: previous video models had a \"production-usable\" length of ~5 seconds — anything longer had visible quality degradation. Seedance 2.5's 30 seconds opens up new use cases: short-form ads, social-media clips, product demos, and game cutscenes. The model is also available with a \"multi-shot\" mode that can generate 60+ seconds with explicit shot boundaries.\n\nThe commercial angle: Seedance 2.5 is available via Volcengine's API at $0.08 per second of 1080p video. The first batch of enterprise customers includes Bilibili (for creator tools), iQiyi (for drama production), and several advertising agencies.\n\nThe bigger takeaway: \"30 seconds\" is the new production threshold for video generation. The \"5-second clip\" era is ending, and \"30-second production-usable\" is the new baseline. The next round of competition will be in \"60 seconds + multi-shot\" — can a model generate a full short film in one go?","seedance-2-5-bytedance-30-second-single-shot","2026-06-23T06:00:00Z","2026-06-23T06:10:21.540015Z","2026-08-19T02:08:40.142862Z",true,"agent",176,{"items":37},[38,43,48,53,58,63],{"id":39,"title":40,"news_slug":41,"published_at":42},"49d19ba1-8f45-475c-bed1-a69dc353523e","字节跳动用 10 万亿参数下注：规模赛跑与张一鸣的「不蒸馏」表态","bytedance-10t-mythos-zhangyiming-no-distill-2026-08","2026-08-08T00:00:00+00:00",{"id":44,"title":45,"news_slug":46,"published_at":47},"6ed14a36-a62a-43e8-949a-cf9df4405d98","Seedance 2.5 把视频生成送进 B 端:30 张参考图、API 上火山方舟、徐工小鹏首批接入","seedance-2-5-enterprise-api-b2b","2026-08-01T04:30:00+00:00",{"id":49,"title":50,"news_slug":51,"published_at":52},"2fbfa6c5-3bf5-4279-a353-6324396b2d36","字节 Seedance 2.5 把单段视频拉到 30 秒：视频生成终于\"能用\"了？","bytedance-seedance-2-5-30s-video-model","2026-07-31T06:00:00+00:00",{"id":54,"title":55,"news_slug":56,"published_at":57},"62ca04a1-d7fb-4169-8e31-4724405dd3b4","原力灵机 DW0.5 把强化学习搬进虚拟世界,真机数据需求骤降 60%","yuanli-lingji-dw-0-5-rl-world","2026-07-18T08:01:00+00:00",{"id":59,"title":60,"news_slug":61,"published_at":62},"f0ea091f-d030-4a32-827a-ae21c1b61c8f","昆仑万维 WAIC 大会将一次性放出四款全模态模型：Matrix-3.5 把\"状态-动作\"塞进一套参数","kunlun-waic-matrix-3-5","2026-07-17T12:08:00+00:00",{"id":64,"title":65,"news_slug":66,"published_at":67},"c463258a-094c-4217-8130-eef2bfacf78c","Xmax X2.0 把实时交互视频模型端侧化:逐帧自回归 + 消费级显卡跑 960p@24fps","xmax-x2-realtime-video","2026-07-16T08:00:00+00:00"]