VFX 分層特效轉影片
將一張參考圖像製作成可混合的 VFX 關鍵影格,加入色差、發光、鏡頭光斑、膠片質感,並用 Seedance 2.0 生成電影感影片。
關於
本 ComfyUI 工作流程可將單張參考圖像製作成分層 VFX 關鍵影格,並以此驅動 Seedance 2.0 電影感影片。Generated VFX Passes 群組中,各特效層(色差、發光/泛光、鏡頭光斑、膠片質感)分別以 ImageBlend 節點混合。發光用 ImageBlur 製作並以 Screen 混合,鏡頭光斑以黑底疊加 Screen,色差為全畫面 Normal 混合。各層強度由 PrimitiveFloat 節點控制,可用 PreviewAny 預覽各階段或用 SaveImage 儲存靜態圖。
當合成關鍵影格調整完成後,將其輸入 Seedance 2.0 Output 群組的 ByteDance2ReferenceNode,產生繼承關鍵影格風格的動態片段。GeminiImage2Node 與 GeminiNode 用於構建與細修圖像特效層,ImageBlend 控制各層疊加。最後 SaveVideo 輸出 Seedance 2.0 影片。此流程可讓你在靜態圖上非破壞性調整電影感,逐層預覽,並將一致風格延伸至影片,僅需一張參考圖。
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": "LoadImage",
"inputs": {
"image": "0a7de162b039a84850dfbff6b319c67bea78a97aaff0c1d2fc979d68914586f8.png"
},
"_meta": {
"title": "Input reference image"
}
},
"2": {
"class_type": "GeminiImage2Node",
"inputs": {
"aspect_ratio": "auto",
"images": [
"1",
0
],
"model": "Nano Banana 2 (Gemini 3.1 Flash Image)",
"prompt": "Using the uploaded image as a strict reference, create a full-frame chromatic aberration pass. Preserve the exact subject, composition, lighting, geometry, identity, colors, and aspect ratio. Add only subtle optical RGB edge separation, prism fringing on high-contrast edges, slight red/cyan fringe near the image borders, and restrained glass dispersion. Do not add text, logos, watermarks, new objects, duplicate subjects, or change the scene.",
"resolution": "2K",
"response_modalities": "IMAGE",
"seed": 947627340135294,
"system_prompt": "You are a precise VFX image-pass generator. Preserve the input image exactly except for the requested optical effect. Return an image only."
},
"_meta": {
"title": "Pass 1 - Chromatic aberration"
}
},
"3": {
"class_type": "ImageBlur",
"inputs": {
"blur_radius": 19,
"image": [
"1",
0
],
"sigma": 5
},
"_meta": {
"title": "Pass 2 - Bloom blur source"
}
},
"4": {
"class_type": "GeminiImage2Node",
"inputs": {
"aspect_ratio": "auto",
"images": [
"1",
0
],
"model": "Nano Banana 2 (Gemini 3.1 Flash Image)",
"prompt": "Create a same-aspect-ratio black-background lens flare overlay matched to the uploaded reference image. The output should mostly be pure black with only realistic optical elements: soft anamorphic streaks, highlight glints, flare ghosts, subtle halation around existing bright light directions, and a few faint dust/sensor sparkle artifacts. Do not reproduce the scene itself. Do not add text, logos, UI, captions, or new objects. This will be blended with Screen/Add mode over the original image.",
"resolution": "2K",
"response_modalities": "IMAGE",
"seed": 148534497297643,
"system_prompt": "You generate optical overlay passes for compositing. Keep non-flare areas black so the pass can be screen-blended."
},
"_meta": {
"title": "Pass 3 - Lens flare overlay"
}
},
"5": {
"class_type": "GeminiImage2Node",
"inputs": {
"aspect_ratio": "auto",
"images": [
"1",
0
],
"model": "Nano Banana 2 (Gemini 3.1 Flash Image)",
"prompt": "Using the uploaded image as a strict reference, create a film finishing pass version of the same frame. Preserve all content, identity, framing, and object layout exactly. Add subtle highlight rolloff, soft halation, restrained film grain, very mild vignette, faint gate texture, and a cinematic color finish. Do not add text, logos, captions, watermarks, new people, new props, or change the scene.",
"resolution": "2K",
"response_modalities": "IMAGE",
"seed": 685209943797967,
"system_prompt": "You are a finishing artist producing a restrained full-frame grade pass. Preserve scene content exactly and return an image only."
},
"_meta": {
"title": "Pass 4 - Film finish"
}
},
"6": {
"class_type": "PrimitiveFloat",
"inputs": {
"value": 0.18
},
"_meta": {
"title": "Chromatic strength"
}
},
"7": {
"class_type": "ImageBlend",
"inputs": {
"blend_factor": [
"6",
0
],
"blend_mode": "normal",
"image1": [
"1",
0
],
"image2": [
"2",
0
]
},
"_meta": {
"title": "Blend chromatic"
}
},
"8": {
"class_type": "PrimitiveFloat",
"inputs": {
"value": 0.28
},
"_meta": {
"title": "Glow strength"
}
},
"9": {
"class_type": "ImageBlend",
"inputs": {
"blend_factor": [
"8",
0
],
"blend_mode": "screen",
"image1": [
"7",
0
],
"image2": [
"3",
0
]
},
"_meta": {
"title": "Blend bloom/glow"
}
},
"10": {
"class_type": "PrimitiveFloat",
"inputs": {
"value": 0.6
},
"_meta": {
"title": "Lens flare strength"
}
},
"11": {
"class_type": "ImageBlend",
"inputs": {
"blend_factor": [
"10",
0
],
"blend_mode": "screen",
"image1": [
"9",
0
],
"image2": [
"4",
0
]
},
"_meta": {
"title": "Blend lens flare"
}
},
"12": {
"class_type": "PrimitiveFloat",
"inputs": {
"value": 0.25
},
"_meta": {
"title": "Film finish strength"
}
},
"13": {
"class_type": "ImageBlend",
"inputs": {
"blend_factor": [
"12",
0
],
"blend_mode": "soft_light",
"image1": [
"11",
0
],
"image2": [
"5",
0
]
},
"_meta": {
"title": "Final VFX composite"
}
},
"14": {
"class_type": "SaveImage",
"inputs": {
"filename_prefix": "vfx_passes_seedance2/final_vfx_reference",
"images": [
"13",
0
]
},
"_meta": {
"title": "Save final reference"
}
},
"16": {
"class_type": "GeminiNode",
"inputs": {
"images": [
"13",
0
],
"model": "gemini-3-1-flash-lite",
"prompt": "Analyze the final VFX reference image and write one vivid Seedance 2.0 reference-to-video motion prompt that makes the shot feel cinematic, dimensional, and high contrast rather than flat. Preserve the exact subject, identity, composition, object layout, and baked VFX look, including chromatic aberration, bloom/glow, lens flare glints, halation, film grain, vignette, and polished cinematic color. Infer the strongest scene-appropriate camera move and describe it with concrete film language: foreground/background parallax, dolly or orbital movement, slight roll or tilt where useful, rack focus, lens compression, motion blur, volumetric light, shadow sweeps, highlight pulses, atmospheric particles, and layered depth. Add purposeful movement in the scene elements so the frame has scale and energy, but avoid chaotic motion. Include strict constraints against text, logo changes, object morphing, extra subjects, warped geometry, scene redesign, and washed-out low-contrast lighting. Return only one 80-120 word prompt.",
"seed": 824781862334978,
"system_prompt": "You are a senior cinematic motion prompt writer for Seedance 2.0. Convert a still VFX reference into a film-trailer-quality video direction with strong depth, lighting contrast, camera motivation, and layered motion while preserving the reference frame exactly. Do not write generic or flat motion such as only subtle drift; choose a controlled cinematic move that reveals scale and parallax. Return exactly one prompt with no markdown."
},
"_meta": {
"title": "Write Seedance motion prompt"
}
},
"17": {
"class_type": "PreviewAny",
"inputs": {
"source": [
"16",
0
]
},
"_meta": {
"title": "Review generated prompt"
}
},
"20": {
"class_type": "ByteDance2ReferenceNode",
"inputs": {
"model": "Seedance 2.0",
"model.auto_downscale": False,
"model.auto_upscale": False,
"model.duration": 6,
"model.generate_audio": True,
"model.prompt": [
"17",
0
],
"model.ratio": "adaptive",
"model.reference_images.image_1": [
"13",
0
],
"model.reference_images.image_2": [
"1",
0
],
"model.resolution": "720p",
"seed": 20481295,
"watermark": False
},
"_meta": {
"title": "Seedance 2.0 reference to video"
}
},
"21": {
"class_type": "SaveVideo",
"inputs": {
"filename_prefix": "video/vfx_passes_seedance2",
"format": "mp4",
"format.codec": "h264",
"video": [
"20",
0
]
},
"_meta": {
"title": "Save Seedance MP4"
}
},
"23": {
"class_type": "SaveImage",
"inputs": {
"filename_prefix": "vfx_passes_seedance2/chromatic_aberration_pass",
"images": [
"2",
0
]
},
"_meta": {
"title": "Save chromatic pass"
}
},
"24": {
"class_type": "SaveImage",
"inputs": {
"filename_prefix": "vfx_passes_seedance2/glow_blur_pass",
"images": [
"3",
0
]
},
"_meta": {
"title": "Save glow pass"
}
},
"25": {
"class_type": "SaveImage",
"inputs": {
"filename_prefix": "vfx_passes_seedance2/lens_flare_overlay_pass",
"images": [
"4",
0
]
},
"_meta": {
"title": "Save lens flare pass"
}
},
"26": {
"class_type": "SaveImage",
"inputs": {
"filename_prefix": "vfx_passes_seedance2/film_finish_pass",
"images": [
"5",
0
]
},
"_meta": {
"title": "Save film finish pass"
}
}
})
# run() also passes api_key to the partner nodes in this workflow.
job = client.run(workflow, api_key=api_key)問答
將該層 PrimitiveFloat 強度設為 0,或將 ImageBlend alpha 設為 0。也可暫時斷開該層輸入。
Screen 混合容易讓亮部過曝。降低發光 PrimitiveFloat,減少 ImageBlur 半徑讓泛光更集中,並以 100% 檢視高光過渡。
降低色差層 PrimitiveFloat 強度,並保持 ImageBlend 模式為 Normal。小幅度即可營造效果,避免邊緣干擾。
先完成合成關鍵影格(可用 SaveImage 與 PreviewAny),再將該輸出送入 ByteDance2ReferenceNode。特效強度保持適中且一致,調整後重新 SaveVideo。