Character × 4 Scene Explorer
Upload a reference image and generate a 1x4 grid. Select desired image and upscale/refine with Nano Banana Pro.
ワークフローをダウンロードCloudで試す API ドキュメント
概要
The 'Character × 4 Scene Explorer' workflow is designed to help users generate a variety of character images based on a single reference image. This workflow leverages the power of Google Gemini Image and Nano Banana Pro models to create a 1x4 grid of character variations. By using nodes such as LoadImage and GeminiImage2Node, the workflow first processes the input image to produce four distinct character scenes. Once generated, users can select their preferred image for further refinement and upscaling using the Nano Banana Pro model, ensuring high-quality output.
Technically, this workflow utilizes several custom nodes, including those from comfyui-kjnodes and comfyui_essentials, which streamline the image processing and refinement tasks. Nodes like Reroute, PrimitiveInt, and RegexReplace are employed to manage data flow and transform inputs efficiently. This workflow is particularly useful for artists and designers looking to explore multiple character design options quickly, as it automates the generation and refinement process, saving time and enhancing creativity.
入力するもの
- ワークフローを読み込み中
ComfyUI で実行できます。 このワークフローはこのページでは実行できません。すべてが読み込まれた状態でComfy Cloudで開くか、ご自身のComfyUI用にダウンロードしてください。
Cloudで試す例
ワークフローの内部
ノードグラフの読み取り専用プレビューです。ノードを編集するには、Comfy Cloudで開くかダウンロードしてください。
Loading the graph
Open full-size workflow preview- 作成者
- Rob
カスタムノードが必要です
- ワークフローを読み込み中
コードからこのワークフローを使用する
可能な場合はワークフローグラフと入力から生成されます。それ以外の場合は保存された例が表示されます。料金が発生する場合があるため、コードを実行する前に認証情報とエンドポイントを確認してください。
1 API キーを取得 ↗
What you need
- Runs on
- Comfy Cloud
- キー
- COMFY_API_KEY
有料の Cloud プランとクレジットが必要です。
API JSONをダウンロードAPI ドキュメント ↗
# Python 3.10+: pip install comfy-sdk==0.4.0
import os
from comfy_sdk import Comfy
api_key = os.environ.get("COMFY_API_KEY", "").strip()
if not api_key: raise RuntimeError("Set COMFY_API_KEY")
os.environ.setdefault("COMFY_BASE_URL", "https://cloud.comfy.org")
client = Comfy(api_key=api_key)
workflow = client.workflows.from_json({
"1": {
"class_type": "LoadImage",
"inputs": {
"image": "cube_robot.jpg"
},
"_meta": {
"title": "Load Image"
}
},
"2": {
"class_type": "GeminiImage2Node",
"inputs": {
"aspect_ratio": "9:16",
"images": [
"1",
0
],
"model": "gemini-3-pro-image-preview",
"prompt": [
"24",
0
],
"resolution": "4K",
"response_modalities": "IMAGE+TEXT",
"seed": 123,
"system_prompt": "You are an expert image-generation engine. You must ALWAYS produce an image.\nInterpret all user input—regardless of format, intent, or abstraction—as literal visual directives for image composition.\nIf a prompt is conversational or lacks specific visual details, you must creatively invent a concrete visual scenario that depicts the concept.\nPrioritize generating the visual representation above any text, formatting, or conversational requests."
},
"_meta": {
"title": "Nano Banana Pro (Google Gemini Image)"
}
},
"3": {
"class_type": "SaveImage",
"inputs": {
"filename_prefix": "ComfyUI-split_stack",
"images": [
"2",
0
]
},
"_meta": {
"title": "Save Image"
}
},
"10": {
"class_type": "SaveImage",
"inputs": {
"filename_prefix": "ComfyUI-image_1",
"images": [
"7:29",
0
]
},
"_meta": {
"title": "Save Image"
}
},
"11": {
"class_type": "SaveImage",
"inputs": {
"filename_prefix": "ComfyUI-image_2",
"images": [
"7:101",
0
]
},
"_meta": {
"title": "Save Image"
}
},
"12": {
"class_type": "SaveImage",
"inputs": {
"filename_prefix": "ComfyUI-image_2",
"images": [
"7:187",
0
]
},
"_meta": {
"title": "Save Image"
}
},
"13": {
"class_type": "SaveImage",
"inputs": {
"filename_prefix": "ComfyUI-image_1",
"images": [
"7:145",
0
]
},
"_meta": {
"title": "Save Image"
}
},
"15": {
"class_type": "GeminiImage2Node",
"inputs": {
"aspect_ratio": "16:9",
"images": [
"14:78",
0
],
"model": "gemini-3-pro-image-preview",
"prompt": "Upscale image 1 and add subtle details. Crop if necessary.",
"resolution": "4K",
"response_modalities": "IMAGE+TEXT",
"seed": 321,
"system_prompt": "You are an expert image-generation engine. You must ALWAYS produce an image.\nInterpret all user input—regardless of format, intent, or abstraction—as literal visual directives for image composition.\nIf a prompt is conversational or lacks specific visual details, you must creatively invent a concrete visual scenario that depicts the concept.\nPrioritize generating the visual representation above any text, formatting, or conversational requests."
},
"_meta": {
"title": "Nano Banana Pro (Google Gemini Image)"
}
},
"16": {
"class_type": "SaveImage",
"inputs": {
"filename_prefix": "ComfyUI-selected_image-upscaled",
"images": [
"15",
0
]
},
"_meta": {
"title": "Save Upscaled Image"
}
},
"18": {
"class_type": "SaveImage",
"inputs": {
"filename_prefix": "ComfyUI-selected_image",
"images": [
"14:78",
0
]
},
"_meta": {
"title": "Save Selected Image"
}
},
"21": {
"class_type": "PrimitiveInt",
"inputs": {
"value": 1
},
"_meta": {
"title": "Select Image Number"
}
},
"22": {
"class_type": "PrimitiveStringMultiline",
"inputs": {
"value": "The subject is inside a giant computer motherboard landscape - towering capacitors, circuit trace pathways like roads, and LED lights creating a Tron-like microverse where the robot fits the scale perfectly. Dark eerie atmosphere."
},
"_meta": {
"title": "Describe the Scene"
}
},
"23": {
"class_type": "PrimitiveStringMultiline",
"inputs": {
"value": "Use the input image. Divide the frame in 4 different stacked 16:9 frames.\nFor each frame choose the angle and type of shot (close up, extreme close up, dutch, over-shoulder, wide shot, Cowboy shot, full shot, etc).\n\nEnsure that subject is present in each frame and the environment is consistent between frames:\n{scene}"
},
"_meta": {
"title": "Prompt Template"
}
},
"24": {
"class_type": "RegexReplace",
"inputs": {
"case_insensitive": True,
"count": 0,
"dotall": False,
"multiline": False,
"regex_pattern": "{scene}",
"replace": [
"22",
0
],
"string": [
"23",
0
]
},
"_meta": {
"title": "Replace Text (Regex)"
}
},
"14:78": {
"class_type": "ImageFromBatch",
"inputs": {
"batch_index": [
"14:89",
0
],
"image": [
"7:333",
0
],
"length": 1
},
"_meta": {
"title": "Get Image from Batch"
}
},
"14:83": {
"class_type": "ResizeAndPadImage",
"inputs": {
"image": [
"1",
0
],
"interpolation": "area",
"padding_color": "white",
"target_height": [
"14:86",
1
],
"target_width": [
"14:86",
0
]
},
"_meta": {
"title": "Resize And Pad Image"
}
},
"14:86": {
"class_type": "GetImageSize",
"inputs": {
"image": [
"14:78",
0
]
},
"_meta": {
"title": "Get Image Size"
}
},
"14:87": {
"class_type": "ImageBatch",
"inputs": {
"image1": [
"14:78",
0
],
"image2": [
"14:83",
0
]
},
"_meta": {
"title": "Batch Images (DEPRECATED)"
}
},
"14:89": {
"class_type": "SimpleMath+",
"inputs": {
"a": [
"21",
0
],
"b": 0,
"value": "a - 1"
},
"_meta": {
"title": "Image # as index"
}
},
"7:101": {
"class_type": "ImageCrop",
"inputs": {
"height": [
"7:66",
0
],
"image": [
"2",
0
],
"width": [
"7:65",
0
],
"x": 0,
"y": [
"7:63",
0
]
},
"_meta": {
"title": "2"
}
},
"7:109": {
"class_type": "PreviewImage",
"inputs": {
"images": [
"7:101",
0
]
},
"_meta": {
"title": "Preview Image"
}
},
"7:136": {
"class_type": "PreviewImage",
"inputs": {
"images": [
"7:145",
0
]
},
"_meta": {
"title": "Preview Image"
}
},
"7:144": {
"class_type": "SimpleMath+",
"inputs": {
"a": [
"7:66",
0
],
"b": 0,
"value": "2*a"
},
"_meta": {
"title": "Row 3 - Y val"
}
},
"7:145": {
"class_type": "ImageCrop",
"inputs": {
"height": [
"7:66",
0
],
"image": [
"2",
0
],
"width": [
"7:65",
0
],
"x": 0,
"y": [
"7:144",
0
]
},
"_meta": {
"title": "3"
}
},
"7:164": {
"class_type": "PreviewImage",
"inputs": {
"images": [
"7:187",
0
]
},
"_meta": {
"title": "Preview Image"
}
},
"7:172": {
"class_type": "SimpleMath+",
"inputs": {
"a": [
"7:66",
0
],
"b": 0,
"value": "3*a"
},
"_meta": {
"title": "Row 4 - Y val"
}
},
"7:18": {
"class_type": "GetImageSize",
"inputs": {
"image": [
"2",
0
]
},
"_meta": {
"title": "Get Image Size"
}
},
"7:187": {
"class_type": "ImageCrop",
"inputs": {
"height": [
"7:66",
0
],
"image": [
"2",
0
],
"width": [
"7:65",
0
],
"x": 0,
"y": [
"7:172",
0
]
},
"_meta": {
"title": "4"
}
},
"7:29": {
"class_type": "ImageCrop",
"inputs": {
"height": [
"7:66",
0
],
"image": [
"2",
0
],
"width": [
"7:65",
0
],
"x": 0,
"y": 0
},
"_meta": {
"title": "1"
}
},
"7:32": {
"class_type": "PreviewImage",
"inputs": {
"images": [
"7:29",
0
]
},
"_meta": {
"title": "Preview Image"
}
},
"7:333": {
"class_type": "ImageBatchMulti",
"inputs": {
"image_1": [
"7:29",
0
],
"image_2": [
"7:101",
0
],
"image_3": [
"7:145",
0
],
"image_4": [
"7:187",
0
],
"inputcount": 4
},
"_meta": {
"title": "Image Batch Multi"
}
},
"7:353": {
"class_type": "SimpleMath+",
"inputs": {
"a": 0,
"b": 0,
"value": "(2*a) + (3*b)"
},
"_meta": {
"title": "Row 3 - Y val"
}
},
"7:63": {
"class_type": "SimpleMath+",
"inputs": {
"a": [
"7:66",
0
],
"b": 0,
"value": "a"
},
"_meta": {
"title": "Row 2 - Y val"
}
},
"7:65": {
"class_type": "SimpleMath+",
"inputs": {
"a": [
"7:18",
0
],
"b": 0,
"value": "a"
},
"_meta": {
"title": "WIDTH"
}
},
"7:66": {
"class_type": "SimpleMath+",
"inputs": {
"a": [
"7:18",
1
],
"b": 0,
"value": "a / 4"
},
"_meta": {
"title": "HEIGHT"
}
}
})
# run() also passes api_key to the partner nodes in this workflow.
job = client.run(workflow, api_key=api_key)よくある質問
Yes, you can upload any image as a reference, but results may vary based on the quality and content of the image.
You can rerun the workflow with different parameters or a new reference image to generate new variations.
The Nano Banana Pro model refines and upscales images, improving details and resolution for a polished final output.
Yes, the workflow is designed to be user-friendly, with step-by-step instructions to guide users through the process.