How does the workflow keep character and outfit consistency when I recreate a single image?

The Select Image group reuses the same subject/reference image(s) and prompt you used for the grid. By feeding those back into the GeminiNanoBanana2V2 generator, the standalone image aligns closely with the chosen cell’s look and outfit.

Where are my images saved?

By default, SaveImage writes to ComfyUI/output (or the folder set in the SaveImage node). The contact sheet and the final single image will be saved according to that configuration.

My 3x3 results look too similar (or too random). How can I adjust variation?

Use the controls inside the 3x3 Contact Sheet group (e.g., seed or randomness parameters inside the CustomCombo/GeminiNanoBanana2V2 node) to increase or decrease variation. Keep the core prompt stable for consistency, and use StringReplace to quickly test small prompt tweaks across the grid.

Can I swap models or references?

Yes. You can replace the subject/reference images in LoadImage to change identity/outfit cues. GeminiNanoBanana2V2 is the model used here; if you have compatible alternatives, you can substitute them in the CustomCombo section, then test a new grid.

3x3 小样图 - ComfyUI 工作流

This workflow builds a fast 3x3 contact sheet of image ideas from a single prompt and a subject reference image, then lets you turn your favorite cell into a standalone image with consistent character and outfit details. The 3x3 Contact Sheet group fans out nine variations through the GeminiNanoBanana2V2 generator (wrapped in a CustomCombo), then assembles them into a grid and saves the result. StringReplace is used to keep text inputs flexible and to label or organize outputs, while LoadImage and SaveImage handle your references and final files. Helpful notes inside the graph are provided via MarkdownNote nodes.

After reviewing the grid, the Select Image group makes it easy to target one frame for re-generation at higher fidelity while reusing the same subject/reference inputs and prompt. This preserves the identity and outfit cues you established up front, so the single image matches what you saw in the grid. It’s a quick, reliable way to explore pose, angle, and styling options without iterating each image manually.

API

通过代码使用此工作流

每个 Comfy 工作流都是一个 JSON 图。下面的内容就是此工作流的完整数据，ComfyUI 运行时也是如此——你可以从该 URL 获取、放入版本控制，或在 ComfyUI 中加载并逐节点运行。

6c3fa01b5d73.json

正在获取工作流 JSON…

使用 Comfy SDK 从 TypeScript 或 Python 运行。相同代码可用于 Comfy Cloud 或你自托管的 ComfyUI——只需更改基础 URL。

// Install (beta)
npm i @comfyorg/sdk

// Run this workflow (TypeScript)
import { Comfy } from "@comfyorg/sdk";

const client = new Comfy({ apiKey: "comfyui-..." });
const wf = await client.workflows.fromFile("workflow_api.json");
const job = await client.run(wf);
await job.getOutputs("<output-node-id>")[0].toFile("output.png");

SDK 需要 API 格式的工作流：在 ComfyUI 中打开此工作流，使用文件 → 导出工作流（API）。

Comfy Cloud API 访问需要包含 API 密钥的套餐。

SDK 文档获取 API 密钥

常见问题