GenmoGenmo
Genmo 把「开放权重」做成招牌:Mochi 1 以 Apache-2.0 开源(同类海外视频台里少见的宽松档)—— 它同时卖云端产品(playground)和可自跑的模型线,两个面要分开看。
Genmo makes open weights its flagship: Mochi 1 is open-sourced under Apache-2.0 (a rare permissive tier among overseas video benches of this kind) — it sells both a cloud product (playground) and a self-runnable model line; the two surfaces must be viewed separately.
适合与不适合When to use it, when not to
适合:开放权重自跑与微调、研究实验、世界模型叙事跟踪。 不适合:把「世界模型」当可交互环境、要产品级文档/API 公开面(登录墙后)、大陆合规场景。
Fits: open-weights self-running and fine-tuning, research experiments, world-model narrative tracking. Not a fit: reading "world models" as interactive environments, wanting a public product docs/API surface (behind a login wall), mainland compliance scenarios.
中文生态China availability 国产 / 国内可直接用 / 备案状态 · 快照domestic / direct access / CAC filing · snapshot 2026-10-30
- 出品属地Origin
- 国外出品foreign-made Genmo(美国;条款页在册「Genmo, Inc Attn: Legal Department」)。Genmo (US; the terms page carries "Genmo, Inc Attn: Legal Department").
- 国内使用Direct access
- 大陆可直连direct OK genmo.ai 大陆直连实测 200(2026-10-30,代理同 200)。genmo.ai measured 200 direct from mainland (2026-10-30; also 200 via proxy).
- 备案状态CAC filing
- 未见于备案名单not in filing list:未见于网信办公示名单(2026-10-30 检索);该产品未面向大陆公众提供服务,备案不适用。not in the CAC public registry (searched 2026-10-30); the product does not serve the mainland public, so filing does not apply.
备案信息以网信办《生成式人工智能服务已备案信息》公示名单为准(按批次滚动更新,本站只抄编号不解读);「大陆可直连」是本站从大陆网络直连实测的结果,不是厂商承诺。GitHub 类站点当日实测有波动,口径写在依据里。Filing status is quoted from the CAC public registry (updated batch by batch; this site copies the number and does not interpret it). "Direct access" is this site's own reachability measurement from a mainland network, not a vendor promise. GitHub-hosted sites fluctuated on the measurement day — the note says so.
固定坐标系Fixed coordinate system
8 个维度dimensions, 与同分区其他对象逐项可比each comparable item by item against others in the same section.
- 模型与开放条件Model & openness
自研模型线 + 开放权重:Mochi 1 开源文生视频模型(GitHub genmoai/mochi,Apache-2.0, ★3,757、推送 2026-10-22,当日 GitHub API 核验);产品内模型选择面在登录墙后,未逐条核验。
In-house model line + open weights: Mochi 1, an open-source text-to-video model (GitHub genmoai/mochi, Apache-2.0, 3,757 stars, pushed 2026-10-22, GitHub API verified the same day); the in-product model-selection surface sits behind a login wall, not verified item by item.
- 运行位置Where it runs
网页(genmo.ai,/playground 200);产品生成在厂商云端。开放权重+训练/推理代码可本地自跑 (仓内提供)——自跑是仓级事实,产品本身不提供离线客户端。
Web (genmo.ai, /playground 200); product generation on the vendor's cloud. Open weights + training/inference code can run locally (provided in the repo) — self-running is a repo-level fact; the product itself ships no offline client.
- 本地文件Local files
产品作品存账户侧云端;开放权重与代码归本地(Apache-2.0)。未见官方声明的本机数据面。
Product works stored account-side in the cloud; open weights and code go local (Apache-2.0). No officially stated local data surface.
- 关机后的任务Tasks after shutdown
生成即时/排队(按档位波动,未逐条核验);无「关机续跑」任务面(未核验到声明)。
Generation is instant/queued (varies by tier, not verified item by item); no "keep-running-after-shutdown" task surface (no claim verified).
- 工具与扩展Tools & extensions
文生视频(Mochi 1 线);首页主打「Open world models」叙事(当日直抓在册); 产品端编辑/交互面细节未逐条核验(/docs /api 均 307 进登录墙)。
Text-to-video (the Mochi 1 line); the homepage headlines the "Open world models" narrative (direct-fetched the same day); product-side editing/interaction details not verified item by item (/docs and /api both 307 into a login wall).
- 上下文与记忆Context & memory
以账户作品组织;跨会话记忆未核验到官方声明。开放权重可复用是资产复用,不是产品记忆。
Organized via account works; cross-session memory — no official statement verified. Reusable open weights are asset reuse, not product memory.
- 权限与限制Permissions & limits
账号+订阅体系(/pricing 200,D9:价格不采);训练用途开关未逐条核验; 开放权重仓的训练/商用自由度按 Apache-2.0 本身。
Account + subscription system (/pricing 200, D9: prices not collected); the training-use switch not verified line by line; training/commercial freedom of the open-weights repo follows Apache-2.0 itself.
- 适合什么任务Best for
适合:要开放权重自跑的创作者与研究者(Apache-2.0 是同类里少见的宽松档)、 世界模型叙事跟踪、创意实验。 不适合:大陆合规场景(未备案、未面向大陆)、企业级 SLA 主打(未核验到声明)。
Fits: creators and researchers who want open weights to self-run (Apache-2.0 is a rare permissive tier in this category), world-model narrative tracking, creative experiments. Does not fit: mainland compliance scenarios (not filed, not serving the mainland), enterprise SLA as the headline (no claim verified).
头号误解Top misconception
- 「Open world models」是定位叙事——Mochi 1 当前是文生视频模型,别把「世界模型」当成可交互环境理解"Open world models" is positioning narrative — Mochi 1 today is a text-to-video model; do not read "world models" as an interactive environment
- 开放权重线(Mochi 仓 Apache-2.0)与产品(playground)是两个面——仓宽松不等于产品输出宽松The open-weights line (Mochi repo, Apache-2.0) and the product (playground) are two surfaces — a permissive repo does not mean permissive product output
- /docs /api /changelog 全部 307 跳登录墙——公开文档面有限,blog 是公开更新线/docs, /api and /changelog all 307 into a login wall — the public docs surface is limited; the blog is the public update line
未知项清单Known unknowns
- 产品 playground 内的模型清单与差异(登录墙后,未核验)
- API 商业化现状(/api 404,未核验到公开 API 页)
- 训练用途开关(未逐条核验)
- 跨会话记忆机制(未核验到官方声明)
- The model list and differences inside the product playground (behind a login wall, unverified)
- API commercialization status (/api 404; no public API page verified)
- The training-use switch (not verified line by line)
- Cross-session memory mechanism (no official statement verified)
证据来源Evidence sources
判断可回到以下一手源复核。本站核验日Every judgement can be rechecked against the primary sources below. Verified on 2026-10-30,内容更新日content updated on 2026-10-30。
| 类型Type | 名称Name | 链接Link |
|---|---|---|
| official | Genmo · 官网(「Open world models」定位,2026-10-30 直抓在册)Genmo · official site ("Open world models" positioning, direct-fetched 2026-10-30) | https://www.genmo.ai |
| official | Genmo · Terms of Service(主体 Genmo, Inc. 在册,2026-10-30 直抓核验)Genmo · Terms of Service (entity Genmo, Inc. on record, direct-fetched and verified 2026-10-30) | https://www.genmo.ai/terms-of-service |
| changelog | Genmo · Blog(官方更新线,changelog;Mochi 1 SOTA 文在册)Genmo · Blog (official update line, changelog; the Mochi 1 SOTA post is on record) | https://www.genmo.ai/blog |
| repo | Mochi 1 · GitHub 仓(Apache-2.0,★3,757,推送 2026-10-22)Mochi 1 · GitHub repo (Apache-2.0, 3,757 stars, pushed 2026-10-22) | https://github.com/genmoai/mochi |
实测记录Measured runs
本档未跑生成质量与速度测试;结论只覆盖能力面与许可面。
实测建议(三条,都指向官方可读页没给的东西):
- Mochi 1 开放权重在消费级显存上的实际可用性与产物质量
- 开放权重输出与产品端输出的质量差距
- 「世界模型」叙事与当前可交互能力的实际边界
This site has not completed measured runs. The coding-oriented test protocol is at protocols/coding/_protocol.md (the coding task suite does not apply to creative products).
Suggested measurements (three, all pointing at things the official readable pages do not give):
- Mochi 1's real usability and output quality on consumer-grade VRAM
- The quality gap between open-weights output and product-side output
- The actual boundary between the "world model" narrative and today's interactive capability
相关条目Related entries
- RunwayRunway同为海外视频台:Runway 闭源专业工具链,Genmo 带开放权重线。Both overseas video benches: Runway is a closed professional toolchain, Genmo carries an open-weights line.
- LumaLuma (Dream Machine)同为叙事向海外视频台:Luma 权重不开源,Genmo 的 Mochi 开源可自跑。Both narrative-oriented overseas video benches: Luma keeps weights closed, Genmo's Mochi is open and self-runnable.
- FluxFLUX (Black Forest Labs)同「开放权重+可自跑」生态逻辑的图像侧对照:宽松权重养出本地生态。The image-side counterpart of the same "open weights + self-runnable" ecosystem logic: permissive weights grew a local ecosystem.
本页由This page is generated from the ai-handbook 数据层生成(CC BY 4.0)。data layer (CC BY 4.0).
方法论与坐标系定义见仓库内For the methodology and coordinate definitions, see METHODOLOGY.md.