動態動效設計
用 Seedance-ready 提示詞製作精緻動態設計影片。
關於
動態動效設計是一個圖片轉影片的 ComfyUI 工作流程,可把靜態 UI 截圖與品牌素材轉成完成度較高的動效短片。它使用 GeminiNode 或 GeminiNanoBanana2,將你的需求與參考資料整理成一段可直接用於 Seedance 的動效提示詞,接著把提示詞與你上傳的圖片一起送入 ByteDance2ReferenceNode 生成動畫。你可以用 PreviewImage 與 PreviewAny 監看進度,並用 SaveVideo 匯出結果。為了清晰易讀,工作流程分成三個群組:Generate Seedance Motion Designer Prompt(由 LLM 驅動的提示詞整理)、Generate UI Frames(快速預覽與影格檢查)、Generate Video(最終渲染與匯出)。
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({
"150": {
"class_type": "ByteDance2ReferenceNode",
"inputs": {
"model": "Seedance 2.0 Fast",
"model.auto_downscale": False,
"model.auto_upscale": False,
"model.duration": 7,
"model.generate_audio": True,
"model.prompt": [
"177",
0
],
"model.ratio": "16:9",
"model.reference_images.image_1": [
"189:10",
0
],
"model.reference_images.image_2": [
"189:15",
0
],
"model.reference_images.image_3": [
"189:26",
0
],
"model.reference_images.image_4": [
"189:30",
0
],
"model.resolution": "720p",
"seed": 1468536166,
"watermark": False
},
"_meta": {
"title": "ByteDance Seedance 2.5 Reference to Video (Legacy)"
}
},
"151": {
"class_type": "GeminiNode",
"inputs": {
"images": [
"189:137",
0
],
"model": "gemini-3-1-pro",
"prompt": "7 second duration, product launch",
"seed": 445440905297116,
"system_prompt": "You are a Seedance Prompt Architect specializing in motion design videos. Convert every user request into a single polished Seedance-ready video generation prompt.\n\nAlways return only the final prompt. Do not include explanations, greetings, headings, markdown commentary, analysis, alternatives, questions, or notes. Do not use emojis.\n\nIf the user provides minimal detail, infer sensible creative defaults. Use a 7 to 15 second duration unless the user specifies otherwise. Use 16:9 for cinematic or brand films, 9:16 for social/mobile videos, and 1:1 only when requested or clearly implied.\n\nWhen images are provided, treat them as visual references. Preserve the subject, composition, colors, materials, logo placement, typography style, product shape, and identity-relevant details. Clearly distinguish between character identity references, product references, texture references, and storyboard/composition references.\n\nWrite prompts with strong motion design direction: kinetic typography, graphic transitions, particle systems, morphing shapes, fluid simulations, glass, chrome, fabric, paper, liquid, smoke, light trails, UI motion, parallax, depth layers, camera choreography, rhythmic cuts, easing, anticipation, follow-through, overshoot, inertia, and seamless transformation logic.\n\nStructure the prompt as one coherent generation prompt containing: duration, aspect ratio, visual style, subject, scene, motion system, time-coded sequence, camera movement, lighting, material behavior, transitions, sound design when useful, and constraints.\n\nUse precise time-coded beats for multi-shot videos. Each beat must have one dominant visual action, clear camera behavior, and a physical or graphic transition into the next beat. Prioritize motion clarity over visual clutter.\n\nMake motion feel designed, not random. Specify animation principles such as smooth acceleration, elastic easing, magnetic snapping, layered parallax, orbital motion, synchronized pulses, frame-filling reveals, match cuts, masked wipes, morph transitions, and final hero-frame resolution.\n\nFor brand, product, UI, and logo videos, maintain product geometry and visual identity. Avoid distorted labels, invented text, random logos, illegible typography, melted packaging, off-brand colors, and uncontrolled camera chaos.\n\nFor abstract motion design, define the visual system: palette, materials, shapes, spatial rules, movement rhythm, transformation mechanism, density curve, and ending composition.\n\nFor loops, explicitly make the final frame return to the first-frame composition with matching object positions, lighting, scale, and motion direction.\n\nEnd every prompt with relevant negative constraints: no subtitles unless requested, no random text, no watermark, no malformed typography, no flicker, no jitter, no distorted geometry, no inconsistent lighting, no unintended logos, no visual noise, no identity drift, no extra objects, and no chaotic cuts."
},
"_meta": {
"title": "Google Gemini"
}
},
"176": {
"class_type": "SaveVideo",
"inputs": {
"codec": "auto",
"filename_prefix": "video/ComfyUI",
"format": "auto",
"format.codec": "auto",
"video": [
"150",
0
]
},
"_meta": {
"title": "Save Video"
}
},
"177": {
"class_type": "PreviewAny",
"inputs": {
"source": [
"151",
0
]
},
"_meta": {
"title": "Preview as Text"
}
},
"190": {
"class_type": "GeminiNanoBanana2",
"inputs": {
"aspect_ratio": "16:9",
"images": [
"208",
0
],
"model": "Nano Banana 2 (Gemini 3.1 Flash Image)",
"prompt": [
"191",
0
],
"resolution": "4K",
"response_modalities": "IMAGE",
"seed": 137159869185568,
"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.",
"thinking_level": "HIGH"
},
"_meta": {
"title": "Nano Banana 2"
}
},
"191": {
"class_type": "PreviewAny",
"inputs": {
"source": [
"192",
0
]
},
"_meta": {
"title": "Preview as Text"
}
},
"192": {
"class_type": "GeminiNode",
"inputs": {
"images": [
"208",
0
],
"model": "gemini-3-1-pro",
"prompt": "",
"seed": 744635533672075,
"system_prompt": "You are a high-end UI motion design prompt engine for landing page visuals, always reenforce the original product shape.\n\nYour sole job is to transform any user input into one polished, production-ready prompt for a short motion design clip featuring premium UI graphics for a landing page.\n\nAlways treat the user’s input as a request for a landing-page-ready UI motion visual.\n\nReturn prompt only.\nDo not explain.\nDo not ask questions.\nDo not provide options.\nDo not use headings.\nDo not output analysis.\nDo not output negative prompts unless explicitly requested.\n\nCore objective:\nGenerate one excellent prompt for visually refined UI graphics intended for short motion clips on a landing page. The result should feel premium, modern, clear, minimal, conversion-oriented, and suitable for high-end SaaS, AI, fintech, productivity, or digital product branding.\n\nDefault visual constraints:\n\n* 2x2 grid\n* no gutters\n* 0px frame\n\nThis layout should be treated as the default composition unless the user explicitly overrides it.\n\nVisual priorities:\n\n* premium UI design\n* clean composition\n* strong visual hierarchy\n* crisp interface elements\n* elegant spacing\n* believable product UI\n* minimal clutter\n* high readability\n* modern motion-design sensibility\n* landing-page-ready polish\n* visually cohesive graphic treatment\n\nMotion direction:\nAssume the graphic is intended for a short motion design clip, typically 3–12 seconds, with subtle, smooth, premium animation language. Even when generating a still-image prompt, the visual should imply motion-design readiness and modular animation potential.\n\nPreferred qualities:\n\n* sleek\n* intentional\n* minimal\n* editorial\n* premium\n* futuristic but usable\n* product-focused\n* visually bold but controlled\n\nAvoid:\n\n* mockup screens\n* device frames\n* bezels\n* browser windows unless explicitly requested\n* clutter\n* generic dashboards\n* floating random icons\n* tiny illegible text\n* chaotic layout\n* overdesigned neon effects\n* excessive glow\n* stock-like visuals\n* cheap UI\n* decorative elements without function\n\nPrompt behavior:\nAlways output one strong final prompt that is directly usable in an image or video generation model.\nThe prompt should clearly describe:\n\n* composition\n* style\n* UI density\n* layout behavior\n* motion-design feel\n* lighting and finish\n* overall brand tone\n\nDefault assumptions when the user is vague:\n\n* premium landing page visual\n* short motion-design clip\n* modular UI graphic system\n* 2x2 grid composition\n* no gutters\n* 0px frame\n* no screen\n* modern product-brand aesthetic\n* clean background\n* high-end commercial polish\n\nThe output must always be a single prompt only.\n"
},
"_meta": {
"title": "Google Gemini"
}
},
"194": {
"class_type": "PreviewImage",
"inputs": {
"images": [
"189:137",
0
]
},
"_meta": {
"title": "Preview Image"
}
},
"208": {
"class_type": "LoadImage",
"inputs": {
"image": "86b1c162b1fba87aa0c95c45ebc536c6d5db163e34a7348518ede58ccc0ef7d5.jpg"
},
"_meta": {
"title": "Load Image"
}
},
"189:10": {
"class_type": "ImageCropV2",
"inputs": {
"crop_region": [
"189:12",
0
],
"image": [
"190",
0
]
},
"_meta": {
"title": "Crop Image"
}
},
"189:12": {
"class_type": "PrimitiveBoundingBox",
"inputs": {
"height": [
"189:13",
1
],
"width": [
"189:7",
1
],
"x": 0,
"y": 0
},
"_meta": {
"title": "Bounding Box"
}
},
"189:13": {
"class_type": "ComfyMathExpression",
"inputs": {
"expression": "a/b",
"values.a": [
"189:8",
1
],
"values.b": [
"189:9",
0
]
},
"_meta": {
"title": "Math Expression"
}
},
"189:137": {
"class_type": "BatchImagesNode",
"inputs": {
"images.image0": [
"189:10",
0
],
"images.image1": [
"189:26",
0
],
"images.image2": [
"189:15",
0
],
"images.image3": [
"189:30",
0
]
},
"_meta": {
"title": "Batch Images"
}
},
"189:15": {
"class_type": "ImageCropV2",
"inputs": {
"crop_region": [
"189:16",
0
],
"image": [
"190",
0
]
},
"_meta": {
"title": "Crop Image"
}
},
"189:16": {
"class_type": "PrimitiveBoundingBox",
"inputs": {
"height": [
"189:13",
1
],
"width": [
"189:7",
1
],
"x": 0,
"y": [
"189:13",
1
]
},
"_meta": {
"title": "Bounding Box"
}
},
"189:25": {
"class_type": "PrimitiveBoundingBox",
"inputs": {
"height": [
"189:13",
1
],
"width": [
"189:7",
1
],
"x": [
"189:7",
1
],
"y": 0
},
"_meta": {
"title": "Bounding Box"
}
},
"189:26": {
"class_type": "ImageCropV2",
"inputs": {
"crop_region": [
"189:25",
0
],
"image": [
"190",
0
]
},
"_meta": {
"title": "Crop Image"
}
},
"189:30": {
"class_type": "ImageCropV2",
"inputs": {
"crop_region": [
"189:32",
0
],
"image": [
"190",
0
]
},
"_meta": {
"title": "Crop Image"
}
},
"189:32": {
"class_type": "PrimitiveBoundingBox",
"inputs": {
"height": [
"189:13",
1
],
"width": [
"189:7",
1
],
"x": [
"189:7",
1
],
"y": [
"189:13",
1
]
},
"_meta": {
"title": "Bounding Box"
}
},
"189:7": {
"class_type": "ComfyMathExpression",
"inputs": {
"expression": "a/b",
"values.a": [
"189:8",
0
],
"values.b": [
"189:9",
0
]
},
"_meta": {
"title": "Math Expression"
}
},
"189:8": {
"class_type": "GetImageSize",
"inputs": {
"image": [
"190",
0
]
},
"_meta": {
"title": "Get Image Size"
}
},
"189:9": {
"class_type": "PrimitiveInt",
"inputs": {
"value": 2
},
"_meta": {
"title": "Int"
}
}
})
# run() also passes api_key to the partner nodes in this workflow.
job = client.run(workflow, api_key=api_key)問答
Seedance-ready prompt 是一段為 Seedance Motion Designer 量身整理過的精簡、結構化文字描述。此工作流程使用 GeminiNode 或 GeminiNanoBanana2,把你的需求與視覺參考轉成清楚、不含糊的提示詞,聚焦在動效意圖(鏡頭運動、節奏、焦點元素)。這能提高一致性,並降低用 ByteDance2ReferenceNode 驅動或把提示詞拿去 Seedance 使用時的試誤成本。
ByteDance2ReferenceNode 是圖片轉影片引擎,會根據你的 Seedance 提示詞與上傳的基礎圖片、風格/品牌參考來解讀並生成動畫序列,並受其參數控制(例如解析度、動作強度、時長/影格數與 seed)。
若你有網路連線並希望使用完整的雲端 LLM 以獲得更豐富的提示詞整理,使用 GeminiNode。若你偏好本機、較輕量的選項或需要離線工作,使用 GeminiNanoBanana2。兩者都能產出 Seedance-ready prompt;依連線狀況與資源限制選擇即可。
先使用乾淨、高解析度的基礎圖片。加入與目標風格一致的品牌/風格參考。Seedance 提示詞保持簡短且指令明確——指定鏡頭運動、焦點元素、節奏與轉場。在 ByteDance2ReferenceNode 中迭代動作強度、時長/影格數與 seed。渲染最終影片前,先用 Generate UI Frames 的預覽確認構圖。
相關工作流程
- LTX-2.3
Add camera motion first last frame
- Wan Animate
Copy movement from a video
- ElevenLabs
ElevenLabs:文生音效
- 合作夥伴節點
Gemini Omni Flash
- Kling, Kling 1.6
Kling 1.6:影片特效
- Kling 3.0
Kling 3.0:動作控制
- Kling 2.6, Kling
Kling2.6:運動控制
- Kling 1.6, Kling
Kling:雙角色互動特效
- 合作夥伴節點
LTX 2.3 Cameraman IC-LoRA
- 合作夥伴節點
LTX 乾淨底板製作(VFX)
- LTX-2.3
LTX-2.3 風格過渡
- MiniMax
Minimax H3 影片重新風格化