在线Stable Diffusion

什么是提示？

提示是一个通常不超过350个字符（75个标记）的短语，描述你想要生成的图像。

负面提示是一种简单的方法，用于指定生成的图像中不应出现的内容。

What is a negative prompt ?

A negative prompt is a list of words or concepts you want Stable Diffusion to avoid in the generated image. While the regular prompt describes what you want, the negative prompt acts as a filter — steering the model away from unwanted elements such as artifacts, deformations, or specific styles. Common entries include terms like blurry, bad anatomy, extra fingers, watermark, low quality, ugly.

Negative prompts work through classifier-free guidance: the model simultaneously moves toward your prompt and away from the negative prompt, with the CFG scale controlling how strongly both signals are applied. Note that Flux models largely make negative prompts obsolete — their improved architecture handles quality and anatomy natively, so prompt adherence alone is usually sufficient.

How does mentioning an artist influence the result ?

Adding an artist's name to your prompt biases the model toward their visual signature — brushwork, color palette, composition, lighting style, and level of detail. This works because Stable Diffusion was trained on large labeled datasets that include artworks attributed to specific artists. For example, adding Greg Rutkowski tends to produce highly detailed fantasy illustrations, while Monet shifts the output toward soft impressionist textures.

Not all artists carry the same weight in the model — some are strongly represented in the training data, others barely influence the output. The effect also varies between model versions: an artist well-recognized by SD 1.5 may be less prominent in SDXL or Flux. Explore hundreds of artist style comparisons on our dedicated artist reference page to find the styles that best match your creative vision.

'seed'是什么意思？

'seed'（种子）是启动创作过程的数字。你不必自己创建这个数字，如果你不选择一个（通常选择-1），它会随机创建。

但如果你控制种子，你可以重新创建相同的图像，尝试不同的参数或修改提示。

什么是LoRA？

LoRA代表'低秩适应'。它是一组修改基础模型的小型扩展。

你可以用它来调整Stable Diffusion以适应某种风格或主题。你可以在一个提示中混合多个LoRA，并赋予不同的权重。这为创作开启了无限可能。

我可以在本地安装Stable Diffusion吗？

可以！需要至少6GB内存（NVRAM）的GPU。

你可以使用社区训练的自定义模型，用LoRA微调结果，以及更多功能。

How to install Stable Diffusion locally ?

What is ComfyUI ?

ComfyUI is a node-based graphical interface for Stable Diffusion. Rather than a traditional form, it lets you build image generation pipelines visually by connecting nodes — each node handles one specific step: loading a model, encoding a prompt, sampling, or decoding the image. This makes the entire process transparent and fully customizable.

Compared to Automatic1111, ComfyUI offers more granular control over the generation pipeline and is generally faster, at the cost of a steeper learning curve. It supports all major models: SD 1.5, SDXL, Stable Diffusion 3, and Flux. The ecosystem is extended via ComfyUI Manager, which lets you install hundreds of community custom nodes directly from the interface. Download it from the official ComfyUI GitHub repository.

What is Flux ?

Flux is a state-of-the-art text-to-image model released in 2024 by Black Forest Labs, founded by former Stability AI researchers. It significantly outperforms SDXL in prompt adherence, photorealism, and — notably — the ability to render legible text inside generated images, a long-standing weakness of previous diffusion models.

Flux comes in three variants: FLUX.1-schnell (fastest, Apache 2.0 license), FLUX.1-dev (higher quality, open weights for non-commercial use), and FLUX.1-pro (best quality, API only). The open variants are available on Hugging Face and run locally via ComfyUI or Automatic1111 with the appropriate extension.