Sky Replacement
A ComfyUI workflow for fast, production-ready sky replacements using Wan models.
下载工作流在 Cloud 中试用 API 文档
00:00
查看工作流内部
源模板的只读预览。在 Cloud 中打开，即可查看节点并进行修改。
Loading the graph
Open full-size workflow preview- 输出
- 视频
- 作者
- ComfyUI
从代码中调用该模型
以下请求使用与体验区相同的预设图和设置。每次提交都可能开始一次付费运行；如果提交结果未知，请勿自动重试。
1 获取 API 密钥 ↗
What you need
- Runs on
- Comfy Cloud
- 密钥
- COMFY_API_KEY
- 你的文件
- 由代码根据链接上传
需要付费的 Cloud 方案和额度。
下载 API JSONAPI 文档 ↗
# 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({
"38": {
"class_type": "WanVideoVAELoader",
"inputs": {
"model_name": "Wan2_1_VAE_fp32.safetensors",
"precision": "fp32",
"use_cpu_cache": False,
"verbose": False
},
"_meta": {
"title": "WanVideo VAE Loader"
}
},
"195": {
"class_type": "WanVideoTorchCompileSettings",
"inputs": {
"allow_unmerged_lora_compile": False,
"backend": "inductor",
"compile_transformer_blocks_only": True,
"dynamic": False,
"dynamo_cache_size_limit": 64,
"dynamo_recompile_limit": 128,
"force_parameter_static_shapes": False,
"fullgraph": False,
"mode": "default"
},
"_meta": {
"title": "WanVideo Torch Compile Settings"
}
},
"891": {
"class_type": "WanVideoLoraSelectMulti",
"inputs": {
"lora_0": "WanAnimate_relight_lora_fp16.safetensors",
"lora_1": "lightx2v_I2V_14B_480p_cfg_step_distill_rank64_bf16.safetensors",
"lora_2": "none",
"lora_3": "none",
"lora_4": "none",
"low_mem_load": False,
"merge_loras": False,
"strength_0": 1,
"strength_1": 1,
"strength_2": 1,
"strength_3": 1,
"strength_4": 1
},
"_meta": {
"title": "WanVideo Lora Select Multi"
}
},
"935": {
"class_type": "CLIPVisionLoader",
"inputs": {
"clip_name": "clip_vision_h.safetensors"
},
"_meta": {
"title": "Load CLIP Vision"
}
},
"936": {
"class_type": "WanVideoClipVisionEncode",
"inputs": {
"clip_vision": [
"935",
0
],
"combine_embeds": "average",
"crop": "center",
"force_offload": True,
"image_1": [
"1295",
0
],
"ratio": 0.5,
"strength_1": 1,
"strength_2": 1,
"tiles": 0
},
"_meta": {
"title": "WanVideo ClipVision Encode"
}
},
"953": {
"class_type": "WanVideoSampler",
"inputs": {
"add_noise_to_samples": False,
"batched_cfg": False,
"cfg": 1,
"denoise_strength": 1,
"end_step": -1,
"force_offload": True,
"image_embeds": [
"1240",
0
],
"model": [
"979",
0
],
"riflex_freq_index": 0,
"rope_function": "comfy",
"scheduler": "dpm++_sde/beta",
"seed": 65346457346,
"shift": 6,
"start_step": 0,
"steps": 6,
"text_embeds": [
"982",
0
],
"uni3c_embeds": [
"1231",
0
]
},
"_meta": {
"title": "WanVideo Sampler"
}
},
"954": {
"class_type": "WanVideoDecode",
"inputs": {
"enable_vae_tiling": False,
"normalization": "default",
"samples": [
"953",
0
],
"tile_stride_x": 144,
"tile_stride_y": 128,
"tile_x": 272,
"tile_y": 272,
"vae": [
"38",
0
]
},
"_meta": {
"title": "WanVideo Decode"
}
},
"979": {
"class_type": "WanVideoModelLoader",
"inputs": {
"attention_mode": "sdpa",
"base_precision": "fp16_fast",
"compile_args": [
"195",
0
],
"load_device": "offload_device",
"lora": [
"891",
0
],
"model": "Wan2_2-Animate-14B_fp8_scaled_e4m3fn_KJ_v2.safetensors",
"quantization": "disabled",
"rms_norm_function": "default"
},
"_meta": {
"title": "WanVideo Model Loader"
}
},
"982": {
"class_type": "WanVideoTextEncodeCached",
"inputs": {
"device": "gpu",
"model_name": "umt5-xxl-enc-bf16.safetensors",
"negative_prompt": "色调艳丽，过曝，静态，细节模糊不清，字幕，风格，作品，画作，画面，静止，整体发灰，最差质量，低质量，JPEG压缩残留，丑陋的，残缺的，多余的手指，画得不好的手部，画得不好的脸部，畸形的，毁容的，形态畸形的肢体，手指融合，静止不动的画面，杂乱的背景，三条腿，背景人很多，倒着走",
"positive_prompt": [
"1201",
0
],
"precision": "bf16",
"quantization": "disabled",
"use_disk_cache": False
},
"_meta": {
"title": "WanVideo TextEncode Cached"
}
},
"1099": {
"class_type": "VHS_VideoInfoLoaded",
"inputs": {
"video_info": [
"1313",
3
]
},
"_meta": {
"title": "Video Info (Loaded) 🎥🅥🅗🅢"
}
},
"1181": {
"class_type": "PreviewImage",
"inputs": {
"images": [
"1295",
0
]
},
"_meta": {
"title": "Preview Image"
}
},
"1201": {
"class_type": "AILab_QwenVL",
"inputs": {
"attention_mode": "sdpa",
"custom_prompt": "Create a video prompt for this image. AI video model Wan2.2. Only respond with the video prompt and nothing else.",
"image": [
"1295",
0
],
"keep_model_loaded": False,
"max_tokens": 512,
"model_name": "Qwen3-VL-2B-Instruct",
"preset_prompt": "🖼️ Detailed Description",
"quantization": "None (FP16)",
"seed": 2074003516
},
"_meta": {
"title": "QwenVL"
}
},
"1231": {
"class_type": "WanVideoUni3C_embeds",
"inputs": {
"controlnet": [
"1232",
0
],
"end_percent": 1,
"offload": True,
"render_latent": [
"1233",
0
],
"start_percent": 0,
"strength": 1
},
"_meta": {
"title": "WanVideo Uni3C Embeds"
}
},
"1232": {
"class_type": "WanVideoUni3C_ControlnetLoader",
"inputs": {
"attention_mode": "sdpa",
"base_precision": "fp16",
"load_device": "main_device",
"model": "Wan21_Uni3C_controlnet_fp16.safetensors",
"quantization": "disabled"
},
"_meta": {
"title": "WanVideo Uni3C Controlnet Loader"
}
},
"1233": {
"class_type": "WanVideoEncode",
"inputs": {
"enable_vae_tiling": False,
"image": [
"1313",
0
],
"latent_strength": 1,
"noise_aug_strength": 0,
"tile_stride_x": 144,
"tile_stride_y": 128,
"tile_x": 272,
"tile_y": 272,
"vae": [
"38",
0
]
},
"_meta": {
"title": "WanVideo Encode"
}
},
"1240": {
"class_type": "WanVideoImageToVideoEncode",
"inputs": {
"augment_empty_frames": 0,
"clip_embeds": [
"936",
0
],
"end_latent_strength": 1,
"force_offload": True,
"fun_or_fl2v_model": False,
"height": [
"1300",
0
],
"noise_aug_strength": 0,
"num_frames": [
"1099",
1
],
"start_image": [
"1295",
0
],
"start_latent_strength": 1,
"tiled_vae": False,
"vae": [
"38",
0
],
"width": [
"1299",
0
]
},
"_meta": {
"title": "WanVideo ImageToVideo Encode"
}
},
"1281": {
"class_type": "GeminiImage2Node",
"inputs": {
"aspect_ratio": "auto",
"images": [
"1314",
0
],
"model": "gemini-3-pro-image-preview",
"prompt": "Replace only the flat, pale sky with a dramatic late-afternoon summer storm rolling in over the steppe. Towering anvil-shaped thunderheads in deep slate blue and bruised violet fill the upper half, their edges lit gold by a low sun breaking through a gap near the horizon. Long shafts of warm light fan down through the clouds, and a faint veil of distant rain hangs over the far left of the horizon. Relight the ground to match the new sky: warm golden light grazing across the cracked earth and grass tufts, with long soft shadows and a cooler, darker tone where cloud shadows fall. Keep the horizon line, the terrain, every pebble and every grass tuft exactly where they are. It must look like a natural photograph, not a composite.",
"resolution": "1K",
"response_modalities": "IMAGE",
"seed": 1085497317911772,
"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)"
}
},
"1285": {
"class_type": "PreviewImage",
"inputs": {
"images": [
"1281",
0
]
},
"_meta": {
"title": "Preview Image"
}
},
"1295": {
"class_type": "ImageScale",
"inputs": {
"crop": "disabled",
"height": [
"1300",
0
],
"image": [
"1281",
0
],
"upscale_method": "nearest-exact",
"width": [
"1299",
0
]
},
"_meta": {
"title": "Upscale Image"
}
},
"1299": {
"class_type": "INTConstant",
"inputs": {
"value": 1280
},
"_meta": {
"title": "width"
}
},
"1300": {
"class_type": "INTConstant",
"inputs": {
"value": 704
},
"_meta": {
"title": "height"
}
},
"1312": {
"class_type": "PreviewAny",
"inputs": {
"source": [
"1201",
0
]
},
"_meta": {
"title": "Preview as Text"
}
},
"1313": {
"class_type": "VHS_LoadVideoFFmpeg",
"inputs": {
"custom_height": [
"1300",
0
],
"custom_width": [
"1299",
0
],
"force_rate": 0,
"format": "Wan",
"frame_load_cap": 0,
"start_time": 0,
"video": "UPLOADED_video_FILENAME"
},
"_meta": {
"title": "Load Video FFmpeg (Upload) 🎥🅥🅗🅢"
}
},
"1314": {
"class_type": "ImageFromBatch",
"inputs": {
"batch_index": 0,
"image": [
"1313",
0
],
"length": 1
},
"_meta": {
"title": "Get Image from Batch"
}
},
"1316": {
"class_type": "VHS_VideoCombine",
"inputs": {
"audio": [
"1313",
2
],
"crf": 22,
"filename_prefix": "SkyReplacement",
"format": "video/h265-mp4",
"frame_rate": [
"1099",
0
],
"images": [
"954",
0
],
"loop_count": 0,
"pingpong": False,
"pix_fmt": "yuv420p10le",
"save_metadata": True,
"save_output": True
},
"_meta": {
"title": "Video Combine 🎥🅥🅗🅢"
}
}
})
workflow.set_input("1313", "video", client.assets.from_url("https://comfy.org/catalog-media/237f4ae8-62d2-483b-a47b-10a35332ffb3/e47c4ffb-b061-4cde-bc56-fdaac2d38528"))
# run() also passes api_key to the partner nodes in this workflow.
job = client.run(workflow, api_key=api_key)
for output in job.get_outputs("1316"):
output.to_file(output.name)试试一个示例
关于此工作流
This ComfyUI workflow is designed for seamless sky replacement in videos, utilizing advanced AI models from Wan. By leveraging nodes such as WanVideoVAELoader and WanVideoModelLoader, the workflow efficiently loads and processes video data. The core of the technique involves using the WanVideoSampler and WanVideoDecode nodes to synthesize and integrate new sky visuals into existing video footage. What sets this workflow apart is its ability to maintain high production quality while ensuring that the new skies blend naturally with the original video content.
The workflow also incorporates the CLIPVisionLoader and WanVideoClipVisionEncode nodes to enhance the semantic understanding of the footage, ensuring that the sky replacements are contextually appropriate. This makes the workflow particularly useful for filmmakers and content creators looking to enhance their video projects with dynamic sky changes without compromising on quality.
常见问答
The workflow supports common video formats such as MP4 and MOV. Ensure your footage is compatible with the WanVideoVAELoader node for optimal results.
Yes, you can customize the sky by using the WanVideoTextEncodeCached node to input specific prompts or select from pre-defined sky models.
The workflow uses the CLIPVisionLoader and WanVideoClipVisionEncode nodes to analyze the video context, ensuring that the sky replacement is semantically consistent with the scene.
Yes, the workflow is designed to handle live-action footage effectively, maintaining high-quality results even in complex scenes.