Krea 2: Style Reference
Upload a style reference image and input a text prompt. Generate an image matching the prompt while applying the aesthetic style from your reference.
试试一个示例
关于此工作流
This ComfyUI workflow demonstrates style-conditioned text-to-image generation using the Krea v2 API. You provide one or more style reference images and a text prompt; the graph extracts a style representation from your references with Krea2StyleReferenceNode and uses it to guide image synthesis in Krea2ImageNode. The output is a new image that fulfills your prompt while inheriting the aesthetic from your reference images. LoadImage supplies the style inputs, Krea2StyleReferenceNode encodes them into a style signal, Krea2ImageNode performs the generation request to Krea’s API, and SaveImage writes the result to disk.
A MarkdownNote is included to annotate the graph with usage tips.
Technically, the data flows as follows: one or more images from LoadImage feed into Krea2StyleReferenceNode, which produces a style reference output (an embedding/handle) that conditions the generator. Krea2ImageNode combines this style reference with your text prompt and generation settings (e.g., size and other parameters exposed in the node) to create the final image on the Krea backend. You can chain multiple input images through the style node(s) to blend influences from different sources, enabling nuanced control over palette, texture, and overall look while keeping the content determined by your prompt.
常见问答
Chaining multiple inputs lets the workflow blend influences from each reference. Earlier references often set broad tone and palette, while later ones can nudge details. If styles conflict, reduce the number of references or adjust their order to prioritize the look you want.
Krea2 nodes call the Krea API. Enter your API key where the node UI requests credentials or follow your ComfyUI/Krea integration’s setup instructions. Without valid credentials, Krea2ImageNode will not generate results.
If the nodes expose a style strength or weighting parameter, lower it for a looser influence and raise it for a closer match to the reference. You can also use fewer or more focused references (e.g., a cropped texture or palette) to fine-tune the effect.
This workflow transfers style, not content. Ensure your text prompt clearly describes the subject and composition you want. Use style references that emphasize look (palette, brushwork, lighting) rather than content elements you expect to be copied.