Nano Banana 2 Lite: Text to Image
Generate images from text descriptions using the ultra-fast Gemini 3.1 Flash-Lite Image model, designed for rapid creation and iteration with no file inputs required. This workflow produces one or more generated images based on your text prompt, with support for interleaved text and image inputs. Ideal for quick concept visualization, rapid prototyping, and iterative design exploration.
试试一个示例
关于此工作流
Nano Banana 2 Lite: Text to Image is a minimal, speed-focused ComfyUI workflow that turns short text prompts into images using Google’s Gemini 3.1 Flash-Lite Image model. It’s built for rapid concepting and quick iterations—no file inputs required—so you can move from idea to visual in seconds. The workflow produces one or more images per prompt, making side-by-side exploration straightforward.
Under the hood, the GeminiNanoBanana2V2 node handles the API call to Gemini 3.1 Flash-Lite Image, taking your text prompt (and optionally interleaved text+image content when supported) and returning generated images to the graph. Outputs are sent directly to SaveImageAdvanced, which writes the results to disk with your chosen folder and filename settings. Because Flash-Lite Image is optimized for low latency, this setup excels at fast iteration loops—adjust the prompt, re-queue, compare, and refine—with minimal overhead.
While it’s streamlined, the node still exposes practical generation controls (such as output count and size, as provided by the node), letting you balance speed and visual detail for early-stage exploration. If you need reference guidance, the workflow supports interleaved content when the node’s interface enables it, but it’s not required for basic text-to-image generation.
常见问答
No. GeminiNanoBanana2V2 calls the Gemini 3.1 Flash-Lite Image API, so generation runs in the cloud. Your machine handles ComfyUI and saving files only.
Follow the GeminiNanoBanana2V2 node’s documentation. Typically, you either paste the key in the node’s settings or set it as an environment/config variable recognized by the node.
Yes, when the node’s interface supports interleaved text+image inputs. You can mix descriptive text with image parts to guide composition or style. If not needed, simple text-only prompts work out of the box.
SaveImageAdvanced handles saving. Set the output directory and filename prefix/pattern there. The node writes each generated image to disk after inference completes.