跳到主要内容Skip to main content
← Image

FLUX(Black Forest Labs)FLUX (Black Forest Labs)

Black Forest Labs
APIAPI 开源权重Open weights 自托管Self-hosted 端侧On-device 商业Commercial partial lifecycle: active 核验Checked 2026-10-30

图像模型家族的双路线供应商:官方 API/playground 托管生成,部分权重开源可自托管甚至进端侧; 「不主张所有权、但禁训练其他模型」是条款里需要单独核对的条款。

The dual-route supplier of the FLUX model family: hosted generation via the official API/playground, some weights open for self-hosting and even on-device; “claims no ownership, but bans training other models” is the sharpest cut in the terms.

适合与不适合When to use it, when not to

适合要接 FLUX API 做集成、或要自托管开源权重的团队。做模型训练的人先读 1.3(n)—— 用 FLUX 输出训练其他模型被条款禁止。要开箱即用的创作社区形态,对照 Midjourney 再选。

Teams integrating the FLUX API or self-hosting open weights. Anyone doing model training reads 1.3(n) first — training other models on FLUX outputs is prohibited by the terms.

Not a fit for out-of-the-box creative-community workflows — compare Midjourney before choosing.

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 2026-10-20

出品属地Origin
国外出品foreign-made 国外主体(网信办公示名单未见对应备案主体)。foreign entity (no matching filing entity in the CAC registry).
国内使用Direct access
大陆可直连direct OK 官网 bfl.ai 直连实测 200。bfl.ai measured 200 direct.
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(2026-10-20 检索 13 批附件,https://www.cac.gov.cn/2024-04/02/c_1713729983803145.htm);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 (13 batch attachments searched 2026-10-20, https://www.cac.gov.cn/2024-04/02/c_1713729983803145.htm); 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

模型家族官方博客披露迭代线:FLUX 3(2026-07-23「Real World Models」多模态流模型条目)、 FLUX 3 Video Part 1(2026-08-04)、FLUX Video Upscale 2K/4K(2026-08-20)、 FLUX 3 Action(2026-09-23)、FLUX 3 Image Klein(2026-10-15,5.6B 开源权重条目) ——2026-10-21 直抓 bfl.ai/blog 核验;部分权重开源可自托管。

Model family with an iteration line disclosed on the official blog: FLUX 3 (2026-07-23 "Real World Models" multimodal flow-model entry), FLUX 3 Video Part 1 (2026-08-04), FLUX Video Upscale 2K/4K (2026-08-20), FLUX 3 Action (2026-09-23), FLUX 3 Image Klein (2026-10-15, 5.6B open-weights entry) — verified by direct fetch of bfl.ai/blog on 2026-10-21; some weights are open and self-hostable.

运行位置Where it runs

托管线:官方 API 与 playground,生成在云端;开源线:权重可下载自托管, 官方另有端侧条目(FLUX.2 Klein 进 ASUS ProArt 笔记本,2026-06-04)。

Hosted line: official API and playground, generation in the cloud; open line: weights downloadable for self-hosting, plus an official on-device entry (FLUX.2 Klein into ASUS ProArt laptops, 2026-06-04).

本地文件Local files

API/playground 可上传图像作输入;自托管线本机跑模型时数据不出本机。

API/playground can upload images as input; on the self-hosted line, when the model runs locally the data never leaves your machine.

关机后的任务Tasks after shutdown

API 为请求-产出制:任务在云端跑,客户端可断开后轮询取回; 没有「关机续跑」形态的官方声明——模型接口本身不承载跨会话任务。

The API is request-output based: tasks run in the cloud, clients can disconnect and poll to retrieve results; no official statement of a "keep-running-after-shutdown" form — the model interface itself carries no cross-session tasks.

工具与扩展Tools & extensions

主形态是模型接口(API/playground),被第三方应用与平台集成(如 Runway、各类图像工作流); 官方不做终端创作套件。

The main form is a model interface (API/playground) integrated by third-party apps and platforms (e.g. Runway, various image workflows); the official side does not build an end-user creation suite.

上下文与记忆Context & memory

模型接口无状态:每次请求独立,不带跨请求记忆;上下文延续由调用方的应用层 自行组织(多参考图输入是请求内能力,不是会话记忆)。

The model interface is stateless: each request is independent, with no cross-request memory; context continuity is organized by the caller's application layer (multi-reference-image input is an in-request capability, not session memory).

权限与限制Permissions & limits

官方 ToS:默认可能用输入与输出改进模型,退出走 [email protected] 邮件申请; 条款对「用输出训练其他模型」有明确限制(1.3(n))。

Official ToS: inputs and outputs may by default be used to improve the model, opting out requires an email request to [email protected]; the terms explicitly restrict "training other models on Outputs" (1.3(n)).

适合什么任务Best for

要接 FLUX 官方 API、或想自托管开源 FLUX 权重的人/团队。托管与自托管许可路径不同, 商用前先确认走哪条;做模型训练的人注意 1.3(n) 的输出训练禁令。

People/teams that want the official FLUX API, or want to self-host open FLUX weights. Hosted and self-hosted licensing paths differ — confirm which one you are on before commercial use; anyone doing model training should note the 1.3(n) output-training ban.

头号误解Top misconception

  • 把「不主张所有权」读成「随便商用」——商用限于你自己的 Output,且 1.3(n) 禁训练其他模型Reading “claims no ownership” as “commercialize freely” — commercial use is limited to your own Output, and 1.3(n) bans training other models
  • 把托管 API 条款套到自托管开源权重——两套许可证体系Applying hosted-API terms to self-hosted open weights — two license systems
  • 训练退出不是自助开关——要发邮件到 legal@ 申请The training opt-out is not a self-serve switch — email legal@ to request it

未知项清单Known unknowns

  • 托管 API 当前全部模型与档位分流(博客按条目披露,本站未逐档核验)
  • 各开源权重许可证的具体商用边界(逐模型许可证本站未逐份核验)
  • The full current model lineup and tier routing of the hosted API (disclosed entry by entry on the blog; this site has not verified each tier)
  • The concrete commercial boundaries of each open-weight license (per-model licenses not verified one by one by this site)

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 2026-10-21。

类型Type名称Name链接Link
changelogBlack Forest Labs · Blog(官方博客,2026-10-21 直抓核验 FLUX 3/Klein/Video Upscale 等条目与日期)Black Forest Labs · Blog (official blog; FLUX 3/Klein/Video Upscale entries and dates verified by direct fetch 2026-10-21)https://bfl.ai/blog
docsBlack Forest Labs · Terms of Service(官方条款,2026-10-20 直抓 200 核验 1.2(a)/1.3(e)/1.3(n) 与 legal@ 退出路径)Black Forest Labs · Terms of Service (official terms; direct-fetched 200 on 2026-10-20, verifying 1.2(a)/1.3(e)/1.3(n) and the legal@ opt-out path)https://blackforestlabs.ai/terms-of-service/
pricingBlack Forest Labs · Pricing(定价页指针,D9:本站不采价格数字)Black Forest Labs · Pricing (pricing-page pointer; D9: this site does not collect price figures)https://bfl.ai/pricing

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。

This site has not completed measured runs. Image generation does not fit the coding test protocol; capability claims follow the official blog and model cards.

相关条目Related entries

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.