# Black Forest Labs 发布开源文生图模型 FLUX 3 Image Klein(5.6B)
> 原标题:FLUX 3 Image Klein
- 来源:Black Forest Labs:Blog
- 发布时间:2026-10-15T15:00:00.000Z
- AIHOT:https://aihot.news/items/ybc0etkws56yxmrveyttxz3lm
- 原文:https://bfl.ai/blog/flux-3-image-klein
## 摘要
Black Forest Labs 发布 5.6B 参数开源权重模型 FLUX 3 Image Klein,支持文生图、图像编辑和最多 10 张参考图的多参考生成,采用 FLUX Kommunity License。
## 推荐理由
官方公布了参数规模、显存门槛、生成速度和评测对比,读者可以评估它在本地部署与微调场景中的适配性。
## 正文 · 中文译文
今天我们发布 FLUX 3 Image Klein,这是一个 5.6B 参数的开源权重模型,用于文本到图像生成、图像编辑和多参考生成,最多支持 10 张参考图像。它与完整的 FLUX 3 Image 模型共享核心能力:原生 4K 生成、像素级精确编辑,以及控制元素放置位置或将编辑限制在指定区域的边界框。
FLUX 3 Image Klein 建立在 FLUX 3 跨图像、视频和音频的共享学习成果之上。图像贡献空间细节,视频增加运动和物理动态,音频提供有关事件和交互的信息。这些信号共同帮助模型生成贴近真实世界的图像。
本次发布包括一个用于微调的基础模型和一个用于低延迟生成的步数蒸馏模型,以及在与 NVIDIA 支持下开发的 FP8、INT8 和 NVFP4 版本,均采用 FLUX Kommunity License。
[在 BFL API 中试用](https://docs.bfl.ai/api-reference/utility/generate-an-image-with-flux-3-klein) · [下载权重](https://huggingface.co/black-forest-labs/flux-3-image-klein-base)

## 以更少参数实现领先的开源权重质量
在内部评估中,FLUX 3 Image Klein 在文本到图像、多参考生成和图像编辑方面排名第一,领先于更大的开源权重模型。在对比中,它的评分也高于闭源模型 Ideogram V4.5 和 FLUX.2 \[pro\]。它以 5.6B 参数成为对比中最小的开源模型,而 Qwen-Image 2.1 为 7B,Krea 2 为 12.9B。与 FLUX.2 \[klein\] 9B 相比,它在文本到图像上提升 91 Elo,在图像编辑上提升 70 Elo,同时参数减少 38%。

FLUX 3 Image Klein 还针对消费级 GPU 进行了优化。更少的参数使其便于本地运行,也更便宜地适配,而有数据主权或本地部署要求的团队可以在自己的基础设施上评估和适配该模型。这对于设备端图像编辑也至关重要,让消费者可以直接在他们常用的笔记本电脑上编辑图像。生成器权重在 BF16 下需要 11.2 GB,在 FP8 下需要 5.6 GB。借助 INT8 版本,FLUX 3 Image Klein 可通过 ComfyUI 或 Diffusers 在显存低至 8GB 的 NVIDIA GeForce RTX GPU 上运行,将 FLUX.2 \[klein\] 开创的能力扩展到更小的硬件上。使用蒸馏模型和 INT8 版本,它通过 ComfyUI 在 NVIDIA GeForce RTX 5090 上约 4 秒生成一张 1MP 图像,在 RTX 4070 上不到 9 秒。

按 GPU 划分的生成时间:步数蒸馏模型,16 步。每张图像秒数,每次运行一张图像。越低越好。由 NVIDIA 测量;RTX 结果在 ComfyUI 中测得;数据结果是每种配置下表现最佳的框架。每个 GPU 显示其最快的版本。生成时间为 GPU 本身的时间;第 02 节中的 API 延迟还包括排队、网络和提示词上采样。

按 GPU 划分的生成时间:基础模型,50 步。每张图像秒数,每次运行一张图像。越低越好。由 NVIDIA 测量;RTX 结果在 ComfyUI 中测得。每个 GPU 显示其最快的版本。
基础检查点专为微调而构建:Diffusers 和 Ostris 的 AI Toolkit 从第一天起就支持 LoRA 训练,NVIDIA 的 NeMo AutoModel 提供了可直接基于已发布检查点运行的配方。
对于需要更低延迟的更快生产工作负载,我们还与团队合作进行特定定制,帮助您实现目标。[联系我们](https://bfl.ai/contact)。
## 在速度与质量的前沿
FLUX 3 Image Klein 还扩展了速度与质量的帕累托前沿。通过 BFL API,它的延迟低于 Qwen-Image 2.1、FLUX.2 \[pro\]、Ideogram V4.5 和 Krea 2,并在其延迟及更高延迟下获得最高评分。在下方图表中,紫色线描绘了每个延迟下可获得的最高评分,灰色线则显示了此前的前沿。
延迟为端到端测量,从请求提交到生成完成的 1 MP 图像,包括排队时间。FLUX 3 Image Klein 在我们的服务器上计时。其他模型从客户端计时,包含轮询和网络时间。Krea 2 按其 Medium 层级计时;其 Large 层级文生图耗时 31.1 秒。

## 开始使用
FLUX 3 Image Klein 今日起以开放权重形式提供,并可通过 [BFL API](https://docs.bfl.ai/api-reference/utility/generate-an-image-with-flux-3-klein) 使用。
### 试用
- BFL API:蒸馏模型以 flux-3-image-klein 提供,支持提示词上采样和像素还原,可精确保留图像中未编辑的区域。定价为每 1K 图像 $0.009 起。
- [Playground](https://dashboard.bfl.ai/playground) 或 [Hugging Face](https://huggingface.co/spaces/black-forest-labs/flux-3-image-klein)。
### 用它构建
- Hugging Face 上的权重:[base](https://huggingface.co/black-forest-labs/flux-3-image-klein-base) 和 [distilled](https://huggingface.co/black-forest-labs/flux-3-image-klein)。
- [参考推理代码](https://github.com/black-forest-labs/flux-3-image-klein),包括用于边界框控制的 Gradio 演示。
- 在 [Hugging Face Spaces](https://huggingface.co/spaces/black-forest-labs/flux-3-image-klein) 上试用。
- 首日支持 [ComfyUI](https://comfy.org/flux-3-image-klein)、[Hugging Face Diffusers](https://huggingface.co/blog/flux-3)、TensorRT-LLM、[SGLang Diffusion](https://docs.sglang.io/cookbook/diffusion/FLUX/FLUX.3) 和 [vLLM-Omni](https://recipes.vllm.ai/black-forest-labs/flux-3-image-klein)。
- 使用 [Ostris 的 AI Toolkit](https://github.com/ostris/ai-toolkit) 进行 LoRA 训练,并在 [NVIDIA NeMo AutoModel](https://github.com/NVIDIA-NeMo/Automodel/blob/main/docs/model-coverage/diffusion/black-forest-labs/flux3-image-klein.mdx) 中提供配方。
- 在 NVIDIA GB200 NVL72 上训练,并针对 NVIDIA 加速计算平台优化,通过 NVIDIA TensorRT LLM、[vLLM-Omni](https://recipes.vllm.ai/black-forest-labs/flux-3-image-klein) 和 [SGLang Diffusion](https://docs.sglang.io/cookbook/diffusion/FLUX/FLUX.3) 提供服务。
- 在 NVIDIA GB300 上,蒸馏模型以 NVFP4 在 0.6 秒内生成 1MP 图像。服务一个 LoRA 会使生成时间增加约 5%。
### 许可证
权重以 FLUX Kommunity License 提供,允许年收入低于 $5M 的个人和公司免费商用,同时允许研究和非商业用途。更大的组织可通过我们的自托管许可选项或 API 进行商业部署。
### 安全
为了让 AI 技术既开放又安全,我们在模型开发生命周期中实施了多重保障措施以防止滥用。这些措施包括在预训练期间使用数据集过滤器以限制接触有害内容,以及在训练后阶段进行针对性微调以抑制有害行为,使诱导有害内容变得越来越困难。在发布之前,FLUX 3 Image Klein 检查点经过了内部严格的红队测试,并由我们的外部安全合作伙伴 Cinder 评估,结果显示其漏洞显著少于其他流行的开放权重模型。有关我们安全方法的更多信息,请见[此处](https://bfl.ai/blog/capable-open-and-safe-combating-ai-misuse)。
FLUX 3 Image Klein 是 FLUX 3 系列继 FLUX 3 Action 之后的第二个开放权重版本。后续还会有更多。
### 了解更多
- [FLUX 3: Real World Models](https://bfl.ai/blog/flux-3),FLUX 3 系列背后的研究
- [FLUX 3 Action](https://bfl.ai/models/flux-3-action),我们用于机器人控制的开放权重 7B 模型
### 引用
如果您在工作中使用 FLUX 3 Image Klein,请引用:
```
@misc{bfl2026flux3image,
author = {{Black Forest Labs}},
title = {{FLUX} 3 Image Klein},
year = {2026},
url = {https://bfl.ai/blog/flux-3-image-klein},
}
```
## 正文 · 原文
Today we are releasing FLUX 3 Image Klein, a 5.6B parameter open weights model for text-to-image generation, image editing and multi-reference generation with up to 10 reference images. It shares core capabilities with the full FLUX 3 Image model: native 4K generation, pixel-perfect editing, and bounding boxes that control where elements are placed or confine edits to a defined area.
FLUX 3 Image Klein builds on FLUX 3's shared learnings across images, video and audio. Images contribute spatial detail, video adds motion and physical dynamics, and audio provides information about events and interactions. Together, these signals help the model generate images that feel grounded in the real world.
The release includes a base model for fine-tuning and a step distilled model for low-latency generation, along with FP8, INT8 and NVFP4 builds developed with NVIDIA support, under the FLUX Kommunity License.
[Try it in the BFL API](https://docs.bfl.ai/api-reference/utility/generate-an-image-with-flux-3-klein) · [Download the weights](https://huggingface.co/black-forest-labs/flux-3-image-klein-base)

## Leading open weights quality with fewer parameters
In internal evaluations, FLUX 3 Image Klein ranks first in text-to-image, multi-reference generation and image editing, ahead of larger open weights models. It also rates above the closed models in the comparison, Ideogram V4.5 and FLUX.2 \[pro\]. At 5.6B parameters, it is the smallest open model in the comparison, against 7B for Qwen-Image 2.1 and 12.9B for Krea 2. Compared with FLUX.2 \[klein\] 9B, it gains 91 Elo in text-to-image and 70 in image editing, with 38% fewer parameters.

FLUX 3 Image Klein was also optimized for consumer GPUs. Fewer parameters make it practical to run locally and cheaper to adapt, and teams with data sovereignty or local deployment requirements can evaluate and adapt the model on their own infrastructure. This is also key for on-device image editing, allowing consumers to edit images directly on their favorite laptops. The generator weights require 11.2 GB in BF16 and 5.6 GB in FP8. With the INT8 build, FLUX 3 Image Klein runs on NVIDIA GeForce RTX GPUs with as little as 8GB of VRAM, through ComfyUI or Diffusers, extending what FLUX.2 \[klein\] started to even smaller hardware. With the distilled model and the INT8 build, it generates a 1MP image in about 4 seconds on an NVIDIA GeForce RTX 5090 and under 9 seconds on an RTX 4070 through ComfyUI.

Generation time by GPU: Step-distilled model, 16 steps. Seconds per image, one image per run. Lower is better. Measured by NVIDIA; RTX results in ComfyUI; data results are the best-performing framework for each configuration. Each GPU shows its quickest build. Generation time on the GPU itself; API latency in Section 02 also includes queueing, network and prompt upsampling.

Generation time by GPU: Base model, 50 steps. Seconds per image, one image per run. Lower is better. Measured by NVIDIA; RTX results in ComfyUI. Each GPU shows its quickest build.
The base checkpoint is built for fine-tuning: Diffusers and Ostris's AI Toolkit support LoRA training from day one, and NVIDIA's NeMo AutoModel ships recipes that work directly from the released checkpoints.
For even faster production workloads that need lower latency, we also work with teams on specific customizations that get you there. [Get in touch](https://bfl.ai/contact).
## On the speed–quality frontier
FLUX 3 Image Klein also extends the speed–quality Pareto frontier. Through the BFL API, it delivers lower latency than Qwen-Image 2.1, FLUX.2 \[pro\], Ideogram V4.5 and Krea 2, and the highest rating at its latency and above. In the charts below, the violet line traces the highest rating available at each latency and the gray line shows the previous frontier.
Latency is measured end to end, from request submission to the finished 1 MP image, including queueing. FLUX 3 Image Klein was timed on our servers. Other models were timed from the client, which includes polling and network time. Krea 2 was timed on its Medium tier; its Large tier took 31.1 seconds for text-to-image.

## Get started
FLUX 3 Image Klein is available today as open weights and through the [BFL API](https://docs.bfl.ai/api-reference/utility/generate-an-image-with-flux-3-klein).
### Try it
- BFL API: the distilled model is available as flux-3-image-klein, with prompt upsampling and pixel restoration, which preserves unedited regions of an image exactly. Pricing starts at $0.009 per 1K image.
- [Playground](https://dashboard.bfl.ai/playground) or [Hugging Face](https://huggingface.co/spaces/black-forest-labs/flux-3-image-klein).
### Build with it
- Weights on Hugging Face: [base](https://huggingface.co/black-forest-labs/flux-3-image-klein-base) and [distilled](https://huggingface.co/black-forest-labs/flux-3-image-klein).
- [Reference inference code](https://github.com/black-forest-labs/flux-3-image-klein), including a Gradio demo for bounding-box control.
- Try it out on [Hugging Face Spaces](https://huggingface.co/spaces/black-forest-labs/flux-3-image-klein).
- Day-one support in [ComfyUI](https://comfy.org/flux-3-image-klein), [Hugging Face Diffusers](https://huggingface.co/blog/flux-3), TensorRT-LLM, [SGLang Diffusion](https://docs.sglang.io/cookbook/diffusion/FLUX/FLUX.3) and [vLLM-Omni](https://recipes.vllm.ai/black-forest-labs/flux-3-image-klein).
- LoRA training with [Ostris's AI Toolkit](https://github.com/ostris/ai-toolkit) and recipes in [NVIDIA NeMo AutoModel](https://github.com/NVIDIA-NeMo/Automodel/blob/main/docs/model-coverage/diffusion/black-forest-labs/flux3-image-klein.mdx).
- Trained on NVIDIA GB200 NVL72 and optimized for NVIDIA's accelerated computing platform, serving through NVIDIA TensorRT LLM, [vLLM-Omni](https://recipes.vllm.ai/black-forest-labs/flux-3-image-klein) and [SGLang Diffusion](https://docs.sglang.io/cookbook/diffusion/FLUX/FLUX.3).
- On NVIDIA GB300, the distilled model generates a 1MP image in under 0.6 seconds in NVFP4. Serving a LoRA adds around 5% to generation time.
### License
The weights are available under the FLUX Kommunity License, which allows free commercial use for individuals and companies with annual revenue under $5M, alongside research and non-commercial use. Larger organizations can deploy commercially through our self-hosted licensing options or API.
### Safety
In an effort to make AI technology both open and safe, we implement multiple safeguards throughout the model development lifecycle to combat misuse. These include dataset filters during pre-training to limit exposure to harmful content, and targeted fine-tuning during post-training to suppress harmful behaviors, making it progressively harder to elicit harmful content. Ahead of the release, the FLUX 3 Image Klein checkpoints were rigorously red teamed internally and evaluated by our external safety partner Cinder, demonstrating substantially fewer vulnerabilities than other popular open weights models. More on our safety methodology is available [here](https://bfl.ai/blog/capable-open-and-safe-combating-ai-misuse).
FLUX 3 Image Klein is the second open weights release from the FLUX 3 family, following FLUX 3 Action. More will follow.
### Learn more
- [FLUX 3: Real World Models](https://bfl.ai/blog/flux-3), the research behind the FLUX 3 family
- [FLUX 3 Action](https://bfl.ai/models/flux-3-action), our open weights 7B model for robot control
### Citation
If you use FLUX 3 Image Klein in your work, please cite:
```
@misc{bfl2026flux3image,
author = {{Black Forest Labs}},
title = {{FLUX} 3 Image Klein},
year = {2026},
url = {https://bfl.ai/blog/flux-3-image-klein},
}
```