分鏡到影片 - Seedance 2.0
用簡單文字提示生成 8 格分鏡,再用該分鏡作為 Seedance 2.0 的動畫基礎。分鏡圖提供視覺藍圖,讓你能更精確控制動作、鏡頭、構圖與故事節奏。
關於
此工作流程會把白話的場景描述轉成 8 格分鏡圖,接著用 Seedance 2.0 依照分鏡來做動畫。在 Previs Storyboard Generator 群組中,你的文字會經過 LLM/影像節點(GeminiNode 與 OpenAIGPTImageNodeV2)處理,並由 StringConcatenate 組裝成每格分鏡的結構化提示詞。生成的影格會用 ImageStitch 排成一張接觸表(contact sheet),透過 PreviewAny 預覽,並由 SaveImage 儲存。這能在正式渲染影片前,先快速取得視覺藍圖——鏡頭順序、機位角度與構圖。
當你對分鏡滿意後,切換到 Seedance Video 群組。在那裡,ByteDance2ReferenceNode 會使用分鏡圖(透過 Image 輸入選取)作為視覺引導,以生成影片的動作與風格。你可以用 PreviewAny 監看進度,並用 SaveVideo 匯出最終結果。此工作流程依賴 ComfyUI 的群組旁路切換,讓你先迭代分鏡,再把成果乾淨地交給 Seedance 2.0 進行動畫製作——每一步都保留創作控制。
您提供的內容
- 正在讀取工作流程
在 ComfyUI 中執行。 此工作流程無法在此頁面上執行。請在已載入所有內容的 Comfy Cloud 中開啟,或下載到您自己的 ComfyUI。
在 Cloud 中試用工作流程內部
節點圖的唯讀預覽。請在 Comfy Cloud 中開啟,或下載後編輯節點。
Loading the graph
Open full-size workflow preview- 作者
- ComfyUI
需要自訂節點
- 正在讀取工作流程
從程式碼使用此工作流程
可行時會根據工作流程圖和輸入內容產生;否則會顯示儲存的範例。執行可能產生費用的程式碼前,請檢查認證資訊和端點。
What you need
- Runs on
- Comfy Cloud
- 金鑰
- COMFY_API_KEY
需要付費 Cloud 方案與點數。
# 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": "OpenAIGPTImageNodeV2",
"inputs": {
"model": "gpt-image-2",
"model.background": "auto",
"model.custom_height": 1024,
"model.custom_width": 1536,
"model.quality": "low",
"model.size": "1536x1024",
"n": 1,
"prompt": [
"4",
0
],
"seed": 1096883914
},
"_meta": {
"title": "OpenAI GPT Image 2.5"
}
},
"2": {
"class_type": "SaveImage",
"inputs": {
"filename_prefix": "ComfyUI",
"images": [
"1",
0
]
},
"_meta": {
"title": "Save Image"
}
},
"4": {
"class_type": "GeminiNode",
"inputs": {
"model": "gemini-3-1-pro",
"prompt": "A ninja sprints across a rooftop beneath the moonlit night sky before launching high into the air. He descends with precision, crashing down onto an unsuspecting enemy and driving them to the ground.\n\nAs the enemy falls, the ninja immediately slips behind the corner of a nearby structure, disappearing from sight. Cut to a close-up of his eyes peering around the edge of the wall—focused, alert, and tense as he scans for danger.\n\nSpotting another enemy nearby, he quietly reaches for a ninja star. Without hesitation, he flicks his wrist and sends the weapon flying across the scene. The spinning star cuts through the air and strikes the second enemy, who stumbles backward from the impact before collapsing to the ground.\n\nFast-paced action, dynamic camera angles, dramatic moonlight, stealth, tension, and fluid ninja choreography.\n",
"seed": 486422127652111,
"system_prompt": "You are an elite action-film previs storyboard artist and cinematic director.\n\nYour task is to convert the user's idea into a highly detailed image-generation prompt for a single storyboard sheet containing 8 continuous sequential action frames.\n\nThe goal is not realism or finished concept art.\n\nThe goal is to communicate:\n\n- action choreography\n- movement\n- camera language\n- timing\n- momentum\n- staging\n- visual storytelling\n- dynamic composition\n\nVISUAL STYLE\n\nCreate a rough director-style previs storyboard.\n\nUse only:\n\n- minimalist stick figures\n- simple gesture drawings\n- construction-line anatomy\n- motion arcs\n- speed lines\n- camera framing indicators\n- directional arrows\n- momentum trails\n- impact marks\n- environmental movement cues\n\nDo NOT include:\n\n- realistic anatomy\n- muscles\n- clothing details\n- costumes\n- shoes\n- facial rendering\n- textures\n- shading\n- polished illustration\n- comic-book rendering\n- finished character design\n\nCharacters should appear as simple action mannequins whose purpose is only to communicate movement.\n\nSTORYBOARD STRUCTURE\n\nGenerate:\n\n- 8 continuous action panels\n- arranged in a storyboard grid\n- each panel representing the next moment in time\n\nEvery frame should feel connected to the previous frame.\n\nShow clear progression of:\n\n- movement\n- momentum\n- action escalation\n- camera movement\n- visual rhythm\n\nCAMERA DIRECTION\n\nUse highly varied cinematic camera language.\n\nPossible examples:\n\n- extreme wide shot\n- wide shot\n- medium shot\n- close-up\n- extreme close-up\n- over-the-shoulder\n- low angle\n- high angle\n- bird's-eye view\n- worm's-eye view\n- Dutch angle\n- tracking shot\n- handheld energy\n- push-in framing\n- pull-back framing\n\nAvoid repetitive framing.\n\nEvery panel should introduce a meaningful camera choice.\n\nACTION VISUALIZATION\n\nEmphasize:\n\n- body mechanics\n- weight shifts\n- momentum\n- anticipation\n- follow-through\n- directional force\n- action readability\n\nVisualize movement using:\n\n- motion trails\n- gesture overlaps\n- action arcs\n- trajectory lines\n- force indicators\n- impact symbols\n- environmental reactions\n\nThe action should feel energetic and easy to understand at a glance.\n\nOBJECT INTERACTION\n\nWhen the user's idea includes props, weapons, tools, vehicles, powers, or environmental interaction:\n\n- make them visually dominant\n- clearly show how they move through space\n- exaggerate motion paths\n- show cause and effect\n- emphasize silhouettes and readability\n\nEFFECTS\n\nEffects should be represented as rough previs graphics:\n\n- ink splashes\n- smoke trails\n- debris\n- energy streaks\n- particles\n- geometric motion shapes\n- environmental reactions\n\nKeep effects simple and readable.\n\nUse them only to reinforce movement and direction.\n\nTEXT USAGE\n\nUse very little text.\n\nOnly include:\n\n- occasional camera notes\n- simple action notes\n- short directional labels\n\nNever use dialogue.\n\nNever use captions that dominate the image.\n\nDRAWING STYLE\n\nThe final image should resemble:\n\n- film previs\n- stunt choreography planning\n- action blocking sheets\n- animation planning boards\n- director thumbnail sketches\n- cinematic motion studies\n\nPrioritize clarity of action over artistic polish.\n\nThe storyboard should feel like it was created by an action director planning a complex sequence before production.\n\nOUTPUT FORMAT\n\nCreate a single image containing all 8 storyboard panels.\n\nEnsure strong visual variety across the panels while maintaining continuous action flow.\n\nThe image should communicate movement, timing, camera choreography, and cinematic energy immediately, even when viewed from a distance."
},
"_meta": {
"title": "Google Gemini"
}
},
"5": {
"class_type": "ByteDance2ReferenceNode",
"inputs": {
"model": "Seedance 2.0",
"model.auto_downscale": False,
"model.auto_upscale": False,
"model.duration": 10,
"model.generate_audio": True,
"model.prompt": [
"20",
0
],
"model.ratio": "16:9",
"model.reference_images.image_1": [
"10",
0
],
"model.reference_images.image_2": [
"7",
0
],
"model.resolution": "720p",
"seed": 275927675,
"watermark": False
},
"_meta": {
"title": "ByteDance Seedance 2.5 Reference to Video (Legacy)"
}
},
"6": {
"class_type": "GeminiNode",
"inputs": {
"images": [
"19",
0
],
"model": "gemini-3-1-pro",
"prompt": "A ninja sprints across a rooftop beneath the moonlit night sky before launching high into the air. He descends with precision, crashing down onto an unsuspecting enemy and driving them to the ground.\n\nAs the enemy falls, the ninja immediately slips behind the corner of a nearby structure, disappearing from sight. Cut to a close-up of his eyes peering around the edge of the wall—focused, alert, and tense as he scans for danger.\n\nSpotting another enemy nearby, he quietly reaches for a ninja star. Without hesitation, he flicks his wrist and sends the weapon flying across the scene. The spinning star cuts through the air and strikes the second enemy, who stumbles backward from the impact before collapsing to the ground.\n\nFast-paced action, dynamic camera angles, dramatic moonlight, stealth, tension, and fluid ninja choreography.\n10 seconds",
"seed": 1050544927289873,
"system_prompt": "Storyboard Camera Interpretation\n\nWhen a storyboard is provided:\n\n* Analyze each panel to determine the intended framing, perspective, shot size, camera position, camera movement, and progression of the sequence.\n* Treat storyboard panels as a shot plan. A change in panel composition may represent either a camera move or a cut depending on what best serves the sequence.\n* Generate dynamic, cinematic shot progression inspired by professional film editing and action choreography.\n* Use exciting, visually engaging shot variety when appropriate, including wide shots, medium shots, close-ups, extreme close-ups, low angles, high angles, tracking shots, push-ins, pull-backs, and perspective-driven compositions.\n* Prioritize strong visual storytelling, pacing, impact, and readability over literal recreation of storyboard drawings.\n* Use cuts when they improve pacing, emphasize important actions, reveal information, increase intensity, or enhance dramatic impact.\n* Ensure every cut feels motivated by the action and preserves continuity of movement, screen direction, subject identity, lighting, environment, and spatial relationships.\n* Maintain clear audience orientation at all times so the subject remains easy to follow throughout the sequence.\n* Translate rough storyboard framing into polished cinematic coverage while preserving the intended sequence of events.\n* Favor cinematic shot progression that escalates in energy and visual interest over time when appropriate.\n* Never reproduce storyboard panel borders, arrows, labels, notes, annotations, or text. Use them only to understand the intended action and camera language.\n* The final video should appear as a professionally shot cinematic sequence, not as an animated storyboard.\n"
},
"_meta": {
"title": "Google Gemini"
}
},
"7": {
"class_type": "LoadImage",
"inputs": {
"image": "34854d164971d4da64fba423cc05f17520192f53926162c7e8380867d485d018.png [output]"
},
"_meta": {
"title": "Load Image"
}
},
"9": {
"class_type": "SaveVideo",
"inputs": {
"filename_prefix": "video/ComfyUI",
"format": "auto",
"format.codec": "auto",
"video": [
"5",
0
]
},
"_meta": {
"title": "Save Video"
}
},
"10": {
"class_type": "LoadImage",
"inputs": {
"image": "b00101f87d3beb138186a1534e0c9ac6a4a95e20d2c1f9a02d59563b165bf816.png"
},
"_meta": {
"title": "Load Image"
}
},
"11": {
"class_type": "PreviewAny",
"inputs": {
"source": [
"20",
0
]
},
"_meta": {
"title": "Preview as Text"
}
},
"19": {
"class_type": "ImageStitch",
"inputs": {
"direction": "right",
"image1": [
"10",
0
],
"image2": [
"7",
0
],
"match_image_size": True,
"spacing_color": "white",
"spacing_width": 0
},
"_meta": {
"title": "Stitch Images"
}
},
"20": {
"class_type": "StringConcatenate",
"inputs": {
"delimiter": ", ",
"string_a": [
"6",
0
],
"string_b": "RULES:\n@image_1 is the character, maintain facial and outfit identity in the generation.\n@image_2 is only for rough scene composition. DO NOT generate arrows or text from this image."
},
"_meta": {
"title": "Concatenate Prompt"
}
}
})
# run() also passes api_key to the partner nodes in this workflow.
job = client.run(workflow, api_key=api_key)問答
需要。OpenAIGPTImageNodeV2 與 GeminiNode 需要依其節點指示設定有效的 API key。Seedance 2.0 步驟透過 ByteDance2ReferenceNode 執行,必須已安裝並正確設定。執行前請先確認所有節點都能無錯誤載入。
可以。用 LoadImage 載入自製分鏡圖(例如接觸表),並將其送入 Seedance 2.0 群組的 Image 輸入。此時可旁路 Previs Storyboard Generator 群組。
在場景描述與任何送入 StringConcatenate 的提示詞文字中寫得更明確:加入每格分鏡的指示,例如「Panel 1:遠景建立鏡頭」、「Panel 2:中景向右跟拍」等。高對比、構圖清晰的分鏡格更利於引導。如有提供,可在 ByteDance2ReferenceNode 提高參考影響程度,以更貼近分鏡。
確認 Seedance 群組的 Image 輸入指向你生成的 8 格分鏡拼接圖。加強分鏡中的視覺線索(清楚輪廓、風格一致),精修提示詞讓構圖更明確,並在可用時提高 ByteDance2ReferenceNode 的參考引導。反覆迭代直到構圖與動作對齊。
相關工作流程
- Nano Banana Pro, Gemini3 Pro Image Preview, Google
Create a character turnaround
- Qwen-Image-Edit
See your subject from a new angle
- 合作夥伴節點
1 張圖像輸入分割堆疊 - Nano Banana 2 + Kling 3.0
- Qwen-Image-Edit, Wan2.2
1個圖像輸入分割堆疊 - Qwen多角度 + Wan 2.2
- Kling 3.0, Nano Banana 2
3.1 進階:多鏡頭影片
- Seedance 2.0
4K Seedance 2.0 - 參考圖像轉影片
- 合合作夥伴節點
Comfy 廣告分鏡腳本產生
- Kling 3.0
Kling 3.0:一鍵多拍
- Qwen-Image-Edit
使用3D相機節點產生多個角度
- Nano Banana Pro
單張圖像產生多張一致鏡頭
- Wan2.2
多關鍵幀影片銜接
- 合作夥伴節點
快速發想到分鏡到動畫影片(GPT Image 2 + Seedance 2.0)