阿里开源视频生成巅峰之作!Wan2.1-VACE-14B实战全解析:从动作迁移到多图融合

        在AI视频生成领域,阿里云开源的Wan2.1-VACE-14B模型凭借其突破性技术架构与多模态能力,正掀起一场创作革命。本文将从核心优势、硬件配置四大核心功能(动作迁移、视频扩展、视频局部编辑、多图融合) 进行深度解析,助你全面掌握这一前沿工具。

VACE github项目地址:GitHub - ali-vilab/VACE: Official implementations for paper: VACE: All-in-One Video Creation and Editinghttps://github.com/ali-vilab/VACE

1 核心优势:技术架构与性能标杆

  • SOTA生成质量:在权威评测集Vbench中,Wan2.1-VACE-14B以86.22%总分超越Sora、Luma等国内外模型,尤其在复杂运动生成物理建模上表现卓越。其采用自研的因果3D VAE架构视频Diffusion Transformer,通过时空压缩与长时程依赖建模,实现无限时长1080P视频编解码,压缩效率提升40%。

  • 多模态全能覆盖:支持五大核心功能: 文本/图像到视频(T2V/I2V) 、视频编辑图像生成(T2I) 及 视频到音频(V2A) ,覆盖创作全流程。尤其突出的是动态中英文字幕生成,显著提升信息传达效率。

  • 硬件兼容性与效率:专业版14B模型需16GB以上显存(如RTX 4090),而极速版1.3B仅需8.2GB显存,可在消费级GPU上4分钟生成5秒480P视频。硬件推荐(需支持 CUDA 11.8+):

        14B专业版:RTX 4090(24GB显存)+ 32GB内存 + 100GB SSD

        1.3B极速版:RTX 3060 Ti(8GB显存)即可运行

        本地显卡配置不足的,可以在onethingai网心云平台使用,新人首次注册送5元代金券,完成学生认证可领20元大额代金券无门槛使用(OneThingAI算力云 - 热门GPU算力平台)。本次实战用到的镜像和GPU参数如下:

2 实战功能解析

2.1 动作迁移:跨主体动态复刻

         技术原理:基于跨注意力机制3D因果VAE,模型通过捕捉源动作的时空特征,结合目标主体的外观一致性约束(如RefAdapter模块),实现自然迁移。

操作案例:将舞蹈动作从专业演员迁移至普通用户,保持肢体协调与角色特征。

        上传动作参考视频与目标图像,调整frame_num(帧数)与sample_guide_scale(动作强度),生成动态一致的新视频。手部细节、面部表情、姿势动态保持的都非常好。

2.2 视频扩展:无限时长与高清增强

         算法创新:结合Flow Matching框架自研VAE,支持从静态图像生成720P高清视频,并通过时间插值实现视频时长扩展。实验显示,14B模型在RTX 4090上生成18秒720P视频仅需19分钟。

操作案例:将电影画面进行扩展,可以保持较高的一致性,并且富有质感

        相较于传统插帧算法(如SuperSloMo),Wan2.1的PSNR提升1.34dB,显著减少画面撕裂。

2.3 视频局部编辑:精准可控的修改操作流程

 通过遮罩工具划定编辑区域,调整参数--sample_shift(噪声调度)与guide_scale(提示词权重),生成局部修改后的连贯视频。

 案例展示:在沙滩边走路的视频中,仅修改人物,背景动态保持自然过

2.4 多图融合:风格与信息的协同创作

输入形式:支持多图输入(如人物图+商品图),通过T5编码器解析语义,在潜空间进行特征融合。像素级融合:采用 离散小波变换(DWT) 与自适应权重分配,保留多图关键细节.

案例展示:图中的男人拿着图中的包包

        画面略微不协调,但人物和物品一致性保持的很好。可以进一步修改重绘、动态参数进一步调整。欢迎大家在评论区反馈生图效果~


模型下载地址:Kijai/WanVideo_comfy at main

!VACE需搭配万相T2V模型使用,例如:VACE-14B必须搭配T2V-14B模型使用!

节点下载地址(请更新到最新版5.12):GitHub - kijai/ComfyUI-WanVideoWrapper

注册平台后,私信免费获取完整工作流

# ComfyUI Error Report ## Error Details - **Node ID:** 42 - **Node Type:** WanVideoSampler - **Exception Type:** AttributeError - **Exception Message:** 'WanModel' object has no attribute 'vace_patch_embedding' ## Stack Trace ``` File "F:\ComfyUI_Mie_V6.0\ComfyUI\execution.py", line 361, in execute output_data, output_ui, has_subgraph = get_output_data(obj, input_data_all, execution_block_cb=execution_block_cb, pre_execute_cb=pre_execute_cb) ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ File "F:\ComfyUI_Mie_V6.0\ComfyUI\execution.py", line 236, in get_output_data return_values = _map_node_over_list(obj, input_data_all, obj.FUNCTION, allow_interrupt=True, execution_block_cb=execution_block_cb, pre_execute_cb=pre_execute_cb) ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ File "F:\ComfyUI_Mie_V6.0\ComfyUI\execution.py", line 208, in _map_node_over_list process_inputs(input_dict, i) File "F:\ComfyUI_Mie_V6.0\ComfyUI\execution.py", line 197, in process_inputs results.append(getattr(obj, func)(**inputs)) ^^^^^^^^^^^^^^^^^^^^^^^^^^^^ File "F:\ComfyUI_Mie_V6.0\ComfyUI\custom_nodes\ComfyUI-WanVideoWrapper\nodes.py", line 3325, in process noise_pred, self.cache_state = predict_with_cfg( ^^^^^^^^^^^^^^^^^ File "F:\ComfyUI_Mie_V6.0\ComfyUI\custom_nodes\ComfyUI-WanVideoWrapper\nodes.py", line 2617, in predict_with_cfg noise_pred_cond, cache_state_cond = transformer( ^^^^^^^^^^^^ File "F:\ComfyUI_Mie_V6.0\python_embeded\Lib\site-packages\torch\nn\modules\module.py", line 1751, in _wrapped_call_impl return self._call_impl(*args, **kwargs) ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ File "F:\ComfyUI_Mie_V6.0\python_embeded\Lib\site-packages\torch\nn\modules\module.py", line 1762, in _call_impl return forward_call(*args, **kwargs) ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ File "F:\ComfyUI_Mie_V6.0\ComfyUI\custom_nodes\ComfyUI-WanVideoWrapper\wanvideo\modules\model.py", line 1617, in forward vace_hints = self.forward_vace(x, data["context"], data["seq_len"], kwargs) ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ File "F:\ComfyUI_Mie_V6.0\ComfyUI\custom_nodes\ComfyUI-WanVideoWrapper\wanvideo\modules\model.py", line 1196, in forward_vace c = [self.vace_patch_embedding(u.unsqueeze(0).float()).to(x.dtype) for u in vace_context] ^^^^^^^^^^^^^^^^^^^^^^^^^ File "F:\ComfyUI_Mie_V6.0\python_embeded\Lib\site-packages\diffusers\models\modeling_utils.py", line 291, in __getattr__ return super().__getattr__(name) ^^^^^^^^^^^^^^^^^^^^^^^^^ File "F:\ComfyUI_Mie_V6.0\python_embeded\Lib\site-packages\torch\nn\modules\module.py", line 1940, in __getattr__ raise AttributeError( ``` ## System Information - **ComfyUI Version:** 0.3.43 - **Arguments:** ComfyUI\main.py --windows-standalone-build --listen 0.0.0.0 --enable-cors-header * - **OS:** nt - **Python Version:** 3.12.10 (tags/v3.12.10:0cc8128, Apr 8 2025, 12:21:36) [MSC v.1943 64 bit (AMD64)] - **Embedded Python:** true - **PyTorch Version:** 2.7.1+cu128 ## Devices - **Name:** cuda:0 NVIDIA GeForce RTX 3090 : cudaMallocAsync - **Type:** cuda - **VRAM Total:** 25769279488 - **VRAM Free:** 24347803648 - **Torch VRAM Total:** 67108864 - **Torch VRAM Free:** 50200576 ## Logs ``` 2025-07-24T17:17:54.612419 - 2025-07-24T17:17:54.663419 - INFO: IPAdapter model loaded from F:\ComfyUI_Mie_V6.0\ComfyUI\models\ipadapter\ip-adapter-plus_sd15.safetensors2025-07-24T17:17:54.663419 - 2025-07-24T17:17:54.663419 - end_vram - start_vram: 2537542306 - 2537542306 = 02025-07-24T17:17:54.663419 - 2025-07-24T17:17:54.664419 - #11 [IPAdapterUnifiedLoader]: 0.05s - vram 0b2025-07-24T17:17:54.664419 - 2025-07-24T17:17:54.924153 - end_vram - start_vram: 2639727650 - 2537542306 = 1021853442025-07-24T17:17:54.924153 - 2025-07-24T17:17:54.924670 - #10 [IPAdapterAdvanced]: 0.26s - vram 102185344b2025-07-24T17:17:54.924670 - 2025-07-24T17:17:54.931880 - end_vram - start_vram: 2733990564 - 2635766434 = 982241302025-07-24T17:17:54.931880 - 2025-07-24T17:17:54.931880 - #14 [RescaleCFG]: 0.01s - vram 98224130b2025-07-24T17:17:54.931880 - 2025-07-24T17:17:54.933882 - Requested to load BaseModel 2025-07-24T17:17:55.233473 - loaded completely 19197.46322517395 1639.406135559082 True 2025-07-24T17:18:23.076833 - 100%|██████████████████████████████████████████████████████████████████████████████████| 30/30 [00:27<00:00, 1.07it/s]2025-07-24T17:18:23.076833 - 100%|██████████████████████████████████████████████████████████████████████████████████| 30/30 [00:27<00:00, 1.08it/s]2025-07-24T17:18:23.076833 - 2025-07-24T17:18:23.078833 - end_vram - start_vram: 4711053742 - 2733990562 = 19770631802025-07-24T17:18:23.078833 - 2025-07-24T17:18:23.078833 - #3 [KSampler]: 28.15s - vram 1977063180b2025-07-24T17:18:23.078833 - 2025-07-24T17:18:23.711202 - end_vram - start_vram: 8100460554 - 3605411690 = 44950488642025-07-24T17:18:23.711239 - 2025-07-24T17:18:23.711239 - #8 [VAEDecode]: 0.63s - vram 4495048864b2025-07-24T17:18:23.711754 - 2025-07-24T17:18:23.864142 - end_vram - start_vram: 3605411690 - 3605411690 = 02025-07-24T17:18:23.864142 - 2025-07-24T17:18:23.864142 - #15 [SaveImage]: 0.15s - vram 0b2025-07-24T17:18:23.865143 - 2025-07-24T17:18:23.865143 - comfyui lumi batcher overwrite task done2025-07-24T17:18:23.865143 - 2025-07-24T17:18:23.866143 - Prompt executed in 31.05 seconds 2025-07-24T17:18:26.480125 - got prompt 2025-07-24T17:18:53.685302 - 100%|██████████████████████████████████████████████████████████████████████████████████| 30/30 [00:27<00:00, 1.14it/s]2025-07-24T17:18:53.686303 - 100%|██████████████████████████████████████████████████████████████████████████████████| 30/30 [00:27<00:00, 1.11it/s]2025-07-24T17:18:53.686303 - 2025-07-24T17:18:53.686303 - end_vram - start_vram: 4689507218 - 3605411690 = 10840955282025-07-24T17:18:53.686303 - 2025-07-24T17:18:53.687302 - #3 [KSampler]: 27.17s - vram 1084095528b2025-07-24T17:18:53.687302 - 2025-07-24T17:18:54.232028 - end_vram - start_vram: 8100460554 - 3605411690 = 44950488642025-07-24T17:18:54.232028 - 2025-07-24T17:18:54.233027 - #8 [VAEDecode]: 0.54s - vram 4495048864b2025-07-24T17:18:54.233027 - 2025-07-24T17:18:54.380027 - end_vram - start_vram: 3605411690 - 3605411690 = 02025-07-24T17:18:54.380027 - 2025-07-24T17:18:54.380027 - #15 [SaveImage]: 0.15s - vram 0b2025-07-24T17:18:54.380027 - 2025-07-24T17:18:54.381027 - comfyui lumi batcher overwrite task done2025-07-24T17:18:54.382027 - 2025-07-24T17:18:54.385027 - Prompt executed in 27.90 seconds 2025-07-24T17:24:25.352704 - got prompt 2025-07-24T17:24:25.445704 - end_vram - start_vram: 3507187562 - 3507187562 = 02025-07-24T17:24:25.446704 - 2025-07-24T17:24:25.446704 - #20 [LoraLoader]: 0.05s - vram 0b2025-07-24T17:24:25.446704 - 2025-07-24T17:24:25.447704 - Requested to load SD1ClipModel 2025-07-24T17:24:25.792290 - loaded completely 18695.929541397094 235.84423828125 True 2025-07-24T17:24:25.800290 - end_vram - start_vram: 3521486046 - 3507187562 = 142984842025-07-24T17:24:25.800290 - 2025-07-24T17:24:25.800290 - #7 [CLIPTextEncode]: 0.35s - vram 14298484b2025-07-24T17:24:25.800290 - 2025-07-24T17:24:25.808290 - end_vram - start_vram: 3519593078 - 3507187562 = 124055162025-07-24T17:24:25.808290 - 2025-07-24T17:24:25.809290 - #6 [CLIPTextEncode]: 0.01s - vram 12405516b2025-07-24T17:24:25.809290 - 2025-07-24T17:24:25.810290 - end_vram - start_vram: 3507187562 - 3507187562 = 02025-07-24T17:24:25.810290 - 2025-07-24T17:24:25.810290 - #11 [IPAdapterUnifiedLoader]: 0.00s - vram 0b2025-07-24T17:24:25.810290 - 2025-07-24T17:24:26.053290 - end_vram - start_vram: 3609372906 - 3507187562 = 1021853442025-07-24T17:24:26.053290 - 2025-07-24T17:24:26.054290 - #10 [IPAdapterAdvanced]: 0.24s - vram 102185344b2025-07-24T17:24:26.054290 - 2025-07-24T17:24:26.059290 - end_vram - start_vram: 3703635820 - 3605411690 = 982241302025-07-24T17:24:26.059290 - 2025-07-24T17:24:26.060290 - #14 [RescaleCFG]: 0.01s - vram 98224130b2025-07-24T17:24:26.060290 - 2025-07-24T17:24:26.061290 - Requested to load BaseModel 2025-07-24T17:24:26.644908 - loaded completely 19912.1512134552 1639.406135559082 True 2025-07-24T17:24:53.885378 - 100%|██████████████████████████████████████████████████████████████████████████████████| 30/30 [00:27<00:00, 1.08it/s]2025-07-24T17:24:53.885378 - 100%|██████████████████████████████████████████████████████████████████████████████████| 30/30 [00:27<00:00, 1.10it/s]2025-07-24T17:24:53.885378 - 2025-07-24T17:24:53.886378 - end_vram - start_vram: 4689507218 - 3703635818 = 9858714002025-07-24T17:24:53.886378 - 2025-07-24T17:24:53.886378 - #3 [KSampler]: 27.83s - vram 985871400b2025-07-24T17:24:53.887378 - 2025-07-24T17:24:54.482797 - end_vram - start_vram: 8100460554 - 3605411690 = 44950488642025-07-24T17:24:54.482797 - 2025-07-24T17:24:54.482797 - #8 [VAEDecode]: 0.60s - vram 4495048864b2025-07-24T17:24:54.482797 - 2025-07-24T17:24:54.640169 - end_vram - start_vram: 3605411690 - 3605411690 = 02025-07-24T17:24:54.640169 - 2025-07-24T17:24:54.640169 - #15 [SaveImage]: 0.16s - vram 0b2025-07-24T17:24:54.640169 - 2025-07-24T17:24:54.641169 - comfyui lumi batcher overwrite task done2025-07-24T17:24:54.641169 - 2025-07-24T17:24:54.643410 - Prompt executed in 29.28 seconds 2025-07-24T17:25:09.991282 - got prompt 2025-07-24T17:25:10.085282 - end_vram - start_vram: 3507187562 - 3507187562 = 02025-07-24T17:25:10.085282 - 2025-07-24T17:25:10.085282 - #20 [LoraLoader]: 0.05s - vram 0b2025-07-24T17:25:10.085282 - 2025-07-24T17:25:10.086282 - Requested to load SD1ClipModel 2025-07-24T17:25:10.369134 - loaded completely 18695.929541397094 235.84423828125 True 2025-07-24T17:25:10.377134 - end_vram - start_vram: 3521486046 - 3507187562 = 142984842025-07-24T17:25:10.377134 - 2025-07-24T17:25:10.377134 - #7 [CLIPTextEncode]: 0.29s - vram 14298484b2025-07-24T17:25:10.377134 - 2025-07-24T17:25:10.385134 - end_vram - start_vram: 3519593078 - 3507187562 = 124055162025-07-24T17:25:10.385134 - 2025-07-24T17:25:10.385134 - #6 [CLIPTextEncode]: 0.01s - vram 12405516b2025-07-24T17:25:10.385134 - 2025-07-24T17:25:10.386135 - end_vram - start_vram: 3507187562 - 3507187562 = 02025-07-24T17:25:10.386135 - 2025-07-24T17:25:10.386135 - #11 [IPAdapterUnifiedLoader]: 0.00s - vram 0b2025-07-24T17:25:10.386135 - 2025-07-24T17:25:10.631046 - end_vram - start_vram: 3609372906 - 3507187562 = 1021853442025-07-24T17:25:10.631046 - 2025-07-24T17:25:10.631046 - #10 [IPAdapterAdvanced]: 0.24s - vram 102185344b2025-07-24T17:25:10.631046 - 2025-07-24T17:25:10.637046 - end_vram - start_vram: 3703635820 - 3605411690 = 982241302025-07-24T17:25:10.637046 - 2025-07-24T17:25:10.637046 - #14 [RescaleCFG]: 0.01s - vram 98224130b2025-07-24T17:25:10.637046 - 2025-07-24T17:25:10.639047 - Requested to load BaseModel 2025-07-24T17:25:11.180881 - loaded completely 19912.1512134552 1639.406135559082 True 2025-07-24T17:25:39.272121 - 100%|██████████████████████████████████████████████████████████████████████████████████| 30/30 [00:28<00:00, 1.08it/s]2025-07-24T17:25:39.272121 - 100%|██████████████████████████████████████████████████████████████████████████████████| 30/30 [00:28<00:00, 1.07it/s]2025-07-24T17:25:39.272121 - 2025-07-24T17:25:39.273121 - end_vram - start_vram: 4689507218 - 3703635818 = 9858714002025-07-24T17:25:39.273121 - 2025-07-24T17:25:39.274121 - #3 [KSampler]: 28.64s - vram 985871400b2025-07-24T17:25:39.274121 - 2025-07-24T17:25:39.888702 - end_vram - start_vram: 8100460554 - 3605411690 = 44950488642025-07-24T17:25:39.888702 - 2025-07-24T17:25:39.888702 - #8 [VAEDecode]: 0.61s - vram 4495048864b2025-07-24T17:25:39.888702 - 2025-07-24T17:25:40.051703 - end_vram - start_vram: 3605411690 - 3605411690 = 02025-07-24T17:25:40.051703 - 2025-07-24T17:25:40.051703 - #15 [SaveImage]: 0.16s - vram 0b2025-07-24T17:25:40.051703 - 2025-07-24T17:25:40.052703 - comfyui lumi batcher overwrite task done2025-07-24T17:25:40.052703 - 2025-07-24T17:25:40.060703 - Prompt executed in 30.06 seconds 2025-07-24T17:43:56.476860 - got prompt 2025-07-24T17:43:57.039587 - end_vram - start_vram: 106744016 - 106744016 = 02025-07-24T17:43:57.039587 - 2025-07-24T17:43:57.039587 - #7 [WanVideoVAELoader]: 0.48s - vram 0b2025-07-24T17:43:57.040588 - 2025-07-24T17:43:57.040588 - end_vram - start_vram: 106744016 - 106744016 = 02025-07-24T17:43:57.040588 - 2025-07-24T17:43:57.040588 - #12 [WanVideoExperimentalArgs]: 0.00s - vram 0b2025-07-24T17:43:57.040588 - 2025-07-24T17:43:57.041588 - end_vram - start_vram: 106744016 - 106744016 = 02025-07-24T17:43:57.041588 - 2025-07-24T17:43:57.041588 - #10 [WanVideoEnhanceAVideo]: 0.00s - vram 0b2025-07-24T17:43:57.041588 - 2025-07-24T17:43:57.042588 - end_vram - start_vram: 106744016 - 106744016 = 02025-07-24T17:43:57.042588 - 2025-07-24T17:43:57.042588 - #21 [WanVideoVACEModelSelect]: 0.00s - vram 0b2025-07-24T17:43:57.042588 - 2025-07-24T17:44:00.282807 - Detected model in_channels: 36 2025-07-24T17:44:00.282807 - Model type: fl2v, num_heads: 40, num_layers: 40 2025-07-24T17:44:00.283807 - Model variant detected: i2v_720 2025-07-24T17:44:00.596667 - model_type FLOW 2025-07-24T17:44:00.597701 - Using accelerate to load and assign model weights to device... 2025-07-24T17:44:00.801188 - Loading transformer parameters to cpu: 83%|██████████████████████████████▋ | 1080/1304 [00:00<00:00, 5458.86it/s]2025-07-24T17:44:00.839187 - Loading transformer parameters to cpu: 100%|█████████████████████████████████████| 1304/1304 [00:00<00:00, 5446.33it/s]2025-07-24T17:44:00.839187 - 2025-07-24T17:44:00.844187 - end_vram - start_vram: 106744016 - 106744016 = 02025-07-24T17:44:00.844187 - 2025-07-24T17:44:00.844187 - #22 [WanVideoModelLoader]: 3.80s - vram 0b2025-07-24T17:44:00.844187 - 2025-07-24T17:44:02.947231 - end_vram - start_vram: 106744016 - 106744016 = 02025-07-24T17:44:02.948230 - 2025-07-24T17:44:02.948230 - #6 [LoadWanVideoT5TextEncoder]: 2.10s - vram 0b2025-07-24T17:44:02.948230 - 2025-07-24T17:44:02.948230 - Moving video model to cpu 2025-07-24T17:45:21.569551 - end_vram - start_vram: 11728619736 - 106744016 = 116218757202025-07-24T17:45:21.569558 - 2025-07-24T17:45:21.569558 - #8 [WanVideoTextEncode]: 78.62s - vram 11621875720b2025-07-24T17:45:21.569558 - 2025-07-24T17:45:21.624655 - end_vram - start_vram: 115132624 - 115132624 = 02025-07-24T17:45:21.625215 - 2025-07-24T17:45:21.625215 - #34 [LoadImage]: 0.05s - vram 0b2025-07-24T17:45:21.625215 - 2025-07-24T17:45:21.626235 - end_vram - start_vram: 115132624 - 115132624 = 02025-07-24T17:45:21.626242 - 2025-07-24T17:45:21.626242 - #19 [INTConstant]: 0.00s - vram 0b2025-07-24T17:45:21.626242 - 2025-07-24T17:45:21.671234 - end_vram - start_vram: 115132624 - 115132624 = 02025-07-24T17:45:21.671234 - 2025-07-24T17:45:21.671234 - #24 [ImageResizeKJ]: 0.04s - vram 0b2025-07-24T17:45:21.671234 - 2025-07-24T17:45:21.738079 - end_vram - start_vram: 115132624 - 115132624 = 02025-07-24T17:45:21.738592 - 2025-07-24T17:45:21.739108 - #30 [ImageResizeKJ]: 0.07s - vram 0b2025-07-24T17:45:21.739108 - 2025-07-24T17:45:21.782786 - end_vram - start_vram: 115132624 - 115132624 = 02025-07-24T17:45:21.782786 - 2025-07-24T17:45:21.782786 - #35 [ImageResizeKJ]: 0.04s - vram 0b2025-07-24T17:45:21.782786 - 2025-07-24T17:45:21.915384 - end_vram - start_vram: 115132624 - 115132624 = 02025-07-24T17:45:21.915902 - 2025-07-24T17:45:21.915902 - #29 [WanVideoVACEStartToEndFrame]: 0.13s - vram 0b2025-07-24T17:45:21.915902 - 2025-07-24T17:45:21.916414 - end_vram - start_vram: 115132624 - 115132624 = 02025-07-24T17:45:21.916918 - 2025-07-24T17:45:21.917002 - #40 [GetImageRangeFromBatch]: 0.00s - vram 0b2025-07-24T17:45:21.917002 - 2025-07-24T17:45:21.981944 - end_vram - start_vram: 115132624 - 115132624 = 02025-07-24T17:45:21.981944 - 2025-07-24T17:45:21.981944 - #25 [WanVideoVACEStartToEndFrame]: 0.06s - vram 0b2025-07-24T17:45:21.981944 - 2025-07-24T17:45:21.995235 - end_vram - start_vram: 115132624 - 115132624 = 02025-07-24T17:45:21.995235 - 2025-07-24T17:45:21.995235 - #41 [MaskBatchMulti]: 0.01s - vram 0b2025-07-24T17:45:21.995235 - 2025-07-24T17:45:22.039203 - end_vram - start_vram: 115132624 - 115132624 = 02025-07-24T17:45:22.039203 - 2025-07-24T17:45:22.039203 - #37 [ImageBatchMulti]: 0.04s - vram 0b2025-07-24T17:45:22.039708 - 2025-07-24T17:45:22.039713 - end_vram - start_vram: 115132624 - 115132624 = 02025-07-24T17:45:22.040227 - 2025-07-24T17:45:22.040227 - #26 [GetImageSizeAndCount]: 0.00s - vram 0b2025-07-24T17:45:22.040227 - 2025-07-24T17:45:22.883828 - input_masks shape2025-07-24T17:45:22.883828 - 2025-07-24T17:45:22.884341 - torch.Size([81, 496, 896])2025-07-24T17:45:22.884341 - 2025-07-24T17:45:34.863567 - VAE encoding: 100%|██████████████████████████████████████████████████████████████████████| 9/9 [00:11<00:00, 1.30s/it]2025-07-24T17:45:34.863567 - VAE encoding: 100%|██████████████████████████████████████████████████████████████████████| 9/9 [00:11<00:00, 1.32s/it]2025-07-24T17:45:34.864077 - 2025-07-24T17:45:46.589102 - VAE encoding: 100%|██████████████████████████████████████████████████████████████████████| 9/9 [00:11<00:00, 1.30s/it]2025-07-24T17:45:46.589612 - VAE encoding: 100%|██████████████████████████████████████████████████████████████████████| 9/9 [00:11<00:00, 1.30s/it]2025-07-24T17:45:46.589612 - 2025-07-24T17:45:46.600743 - refs shape2025-07-24T17:45:46.600743 - 2025-07-24T17:45:46.600743 - torch.Size([1, 3, 1, 496, 896])2025-07-24T17:45:46.600743 - 2025-07-24T17:45:46.715123 - VAE encoding: 56%|██████████████████████████████████████▉ | 5/9 [00:00<00:00, 44.31it/s]2025-07-24T17:45:46.805051 - VAE encoding: 100%|██████████████████████████████████████████████████████████████████████| 9/9 [00:00<00:00, 44.38it/s]2025-07-24T17:45:46.805583 - 2025-07-24T17:45:46.870662 - end_vram - start_vram: 2252473238 - 115132624 = 21373406142025-07-24T17:45:46.870662 - 2025-07-24T17:45:46.871171 - #16 [WanVideoVACEEncode]: 24.83s - vram 2137340614b2025-07-24T17:45:46.871171 - 2025-07-24T17:45:46.873216 - timesteps: tensor([1000.0000, 983.7124, 964.5367, 941.6301, 913.7864, 879.2166, 835.1485, 777.0445, 696.9283, 579.3684, 390.1099, 34.7826], device='cuda:0') 2025-07-24T17:45:46.887141 - Seq len: 38192 2025-07-24T17:49:31.510530 - Sampling 85 frames at 896x496 with 12 steps 2025-07-24T17:49:37.849674 - 0%| | 0/12 [00:00<?, ?it/s]2025-07-24T17:49:44.296375 - 0%| | 0/12 [00:06<?, ?it/s]2025-07-24T17:49:44.297376 - 2025-07-24T17:49:44.747159 - !!! Exception during processing !!! 'WanModel' object has no attribute 'vace_patch_embedding' 2025-07-24T17:49:44.783076 - Traceback (most recent call last): File "F:\ComfyUI_Mie_V6.0\ComfyUI\execution.py", line 361, in execute output_data, output_ui, has_subgraph = get_output_data(obj, input_data_all, execution_block_cb=execution_block_cb, pre_execute_cb=pre_execute_cb) ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ File "F:\ComfyUI_Mie_V6.0\ComfyUI\execution.py", line 236, in get_output_data return_values = _map_node_over_list(obj, input_data_all, obj.FUNCTION, allow_interrupt=True, execution_block_cb=execution_block_cb, pre_execute_cb=pre_execute_cb) ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ File "F:\ComfyUI_Mie_V6.0\ComfyUI\execution.py", line 208, in _map_node_over_list process_inputs(input_dict, i) File "F:\ComfyUI_Mie_V6.0\ComfyUI\execution.py", line 197, in process_inputs results.append(getattr(obj, func)(**inputs)) ^^^^^^^^^^^^^^^^^^^^^^^^^^^^ File "F:\ComfyUI_Mie_V6.0\ComfyUI\custom_nodes\ComfyUI-WanVideoWrapper\nodes.py", line 3325, in process noise_pred, self.cache_state = predict_with_cfg( ^^^^^^^^^^^^^^^^^ File "F:\ComfyUI_Mie_V6.0\ComfyUI\custom_nodes\ComfyUI-WanVideoWrapper\nodes.py", line 2617, in predict_with_cfg noise_pred_cond, cache_state_cond = transformer( ^^^^^^^^^^^^ File "F:\ComfyUI_Mie_V6.0\python_embeded\Lib\site-packages\torch\nn\modules\module.py", line 1751, in _wrapped_call_impl return self._call_impl(*args, **kwargs) ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ File "F:\ComfyUI_Mie_V6.0\python_embeded\Lib\site-packages\torch\nn\modules\module.py", line 1762, in _call_impl return forward_call(*args, **kwargs) ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ File "F:\ComfyUI_Mie_V6.0\ComfyUI\custom_nodes\ComfyUI-WanVideoWrapper\wanvideo\modules\model.py", line 1617, in forward vace_hints = self.forward_vace(x, data["context"], data["seq_len"], kwargs) ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ File "F:\ComfyUI_Mie_V6.0\ComfyUI\custom_nodes\ComfyUI-WanVideoWrapper\wanvideo\modules\model.py", line 1196, in forward_vace c = [self.vace_patch_embedding(u.unsqueeze(0).float()).to(x.dtype) for u in vace_context] ^^^^^^^^^^^^^^^^^^^^^^^^^ File "F:\ComfyUI_Mie_V6.0\python_embeded\Lib\site-packages\diffusers\models\modeling_utils.py", line 291, in __getattr__ return super().__getattr__(name) ^^^^^^^^^^^^^^^^^^^^^^^^^ File "F:\ComfyUI_Mie_V6.0\python_embeded\Lib\site-packages\torch\nn\modules\module.py", line 1940, in __getattr__ raise AttributeError( AttributeError: 'WanModel' object has no attribute 'vace_patch_embedding'. Did you mean: 'patch_embedding'? 2025-07-24T17:49:44.785980 - end_vram - start_vram: 34183609272 - 173795536 = 340098137362025-07-24T17:49:44.785980 - 2025-07-24T17:49:44.785980 - #42 [WanVideoSampler]: 237.91s - vram 34009813736b2025-07-24T17:49:44.785980 - 2025-07-24T17:49:44.798561 - comfyui lumi batcher overwrite task done2025-07-24T17:49:44.798968 - 2025-07-24T17:49:44.868694 - Prompt executed in 348.38 seconds 2025-07-24T18:17:32.104770 - [ComfyUI-Manager] The ComfyRegistry cache update is still in progress, so an outdated cache is being used.2025-07-24T18:17:32.105296 - 2025-07-24T18:17:33.147891 - FETCH DATA from: https://raw.githubusercontent.com/ltdrdata/ComfyUI-Manager/main/custom-node-list.json2025-07-24T18:17:33.147891 - 2025-07-24T18:17:36.036752 - [DONE]2025-07-24T18:17:36.036752 - 2025-07-24T18:17:36.120191 - FETCH DATA from: https://raw.githubusercontent.com/ltdrdata/ComfyUI-Manager/main/github-stats.json2025-07-24T18:17:36.120191 - 2025-07-24T18:17:37.292226 - [DONE]2025-07-24T18:17:37.292226 - 2025-07-24T18:17:37.313785 - FETCH DATA from: https://raw.githubusercontent.com/ltdrdata/ComfyUI-Manager/main/extras.json2025-07-24T18:17:37.313785 - 2025-07-24T18:17:38.420206 - [DONE]2025-07-24T18:17:38.420897 - 2025-07-24T18:17:38.602146 - FETCH DATA from: https://raw.githubusercontent.com/ltdrdata/ComfyUI-Manager/main/extension-node-map.json2025-07-24T18:17:38.602146 - 2025-07-24T18:17:39.726647 - [DONE]2025-07-24T18:17:39.726647 - 2025-07-24T18:21:32.720524 - got prompt 2025-07-24T18:21:36.403235 - timesteps: tensor([1000.0000, 983.7124, 964.5367, 941.6301, 913.7864, 879.2166, 835.1485, 777.0445, 696.9283, 579.3684, 390.1099, 34.7826], device='cuda:0') 2025-07-24T18:21:36.571239 - Seq len: 38192 2025-07-24T18:21:36.578239 - Sampling 85 frames at 896x496 with 12 steps 2025-07-24T18:21:36.834814 - 0%| | 0/12 [00:00<?, ?it/s]2025-07-24T18:21:40.522225 - 0%| | 0/12 [00:03<?, ?it/s]2025-07-24T18:21:40.522225 - 2025-07-24T18:21:40.567079 - !!! Exception during processing !!! 'WanModel' object has no attribute 'vace_patch_embedding' 2025-07-24T18:21:40.572236 - Traceback (most recent call last): File "F:\ComfyUI_Mie_V6.0\ComfyUI\execution.py", line 361, in execute output_data, output_ui, has_subgraph = get_output_data(obj, input_data_all, execution_block_cb=execution_block_cb, pre_execute_cb=pre_execute_cb) ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ File "F:\ComfyUI_Mie_V6.0\ComfyUI\execution.py", line 236, in get_output_data return_values = _map_node_over_list(obj, input_data_all, obj.FUNCTION, allow_interrupt=True, execution_block_cb=execution_block_cb, pre_execute_cb=pre_execute_cb) ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ File "F:\ComfyUI_Mie_V6.0\ComfyUI\execution.py", line 208, in _map_node_over_list process_inputs(input_dict, i) File "F:\ComfyUI_Mie_V6.0\ComfyUI\execution.py", line 197, in process_inputs results.append(getattr(obj, func)(**inputs)) ^^^^^^^^^^^^^^^^^^^^^^^^^^^^ File "F:\ComfyUI_Mie_V6.0\ComfyUI\custom_nodes\ComfyUI-WanVideoWrapper\nodes.py", line 3325, in process noise_pred, self.cache_state = predict_with_cfg( ^^^^^^^^^^^^^^^^^ File "F:\ComfyUI_Mie_V6.0\ComfyUI\custom_nodes\ComfyUI-WanVideoWrapper\nodes.py", line 2617, in predict_with_cfg noise_pred_cond, cache_state_cond = transformer( ^^^^^^^^^^^^ File "F:\ComfyUI_Mie_V6.0\python_embeded\Lib\site-packages\torch\nn\modules\module.py", line 1751, in _wrapped_call_impl return self._call_impl(*args, **kwargs) ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ File "F:\ComfyUI_Mie_V6.0\python_embeded\Lib\site-packages\torch\nn\modules\module.py", line 1762, in _call_impl return forward_call(*args, **kwargs) ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ File "F:\ComfyUI_Mie_V6.0\ComfyUI\custom_nodes\ComfyUI-WanVideoWrapper\wanvideo\modules\model.py", line 1617, in forward vace_hints = self.forward_vace(x, data["context"], data["seq_len"], kwargs) ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ File "F:\ComfyUI_Mie_V6.0\ComfyUI\custom_nodes\ComfyUI-WanVideoWrapper\wanvideo\modules\model.py", line 1196, in forward_vace c = [self.vace_patch_embedding(u.unsqueeze(0).float()).to(x.dtype) for u in vace_context] ^^^^^^^^^^^^^^^^^^^^^^^^^ File "F:\ComfyUI_Mie_V6.0\python_embeded\Lib\site-packages\diffusers\models\modeling_utils.py", line 291, in __getattr__ return super().__getattr__(name) ^^^^^^^^^^^^^^^^^^^^^^^^^ File "F:\ComfyUI_Mie_V6.0\python_embeded\Lib\site-packages\torch\nn\modules\module.py", line 1940, in __getattr__ raise AttributeError( AttributeError: 'WanModel' object has no attribute 'vace_patch_embedding'. Did you mean: 'patch_embedding'? 2025-07-24T18:21:40.574784 - end_vram - start_vram: 34183609272 - 32966353744 = 12172555282025-07-24T18:21:40.574784 - 2025-07-24T18:21:40.575297 - #42 [WanVideoSampler]: 7.68s - vram 1217255528b2025-07-24T18:21:40.575297 - 2025-07-24T18:21:40.575811 - comfyui lumi batcher overwrite task done2025-07-24T18:21:40.575811 - 2025-07-24T18:21:40.577866 - Prompt executed in 7.84 seconds ``` ## Attached Workflow Please make sure that workflow does not contain any sensitive information such as API keys or passwords. ``` Workflow too large. Please manually upload the workflow from local file system. ``` ## Additional Context (Please add any additional context or steps to reproduce the error here) 什么问题,该怎么解决
最新发布
07-25
<think>嗯,用户遇到了在使用VACE、MoviiGen和14B模型的生视频工流中出现的MissingNode Types问题,特别是RHHiddenNodes节点缺失的问题。我需要先理解这个问题的可能原因,然后给出解决方案。首先,RHHiddenNodes节点缺失通常是因为缺少相应的自定义节点包。可能用户的工流依赖了一些第三方节点,而这些节点没有被正确安装。比如,RHHiddenNodes可能属于某个特定的节点包,比如ComfyUI的管理器或者某个自定义扩展。接下来,我应该检查用户的ComfyUI版本是否最新,因为旧版本可能不兼容某些节点。然后,建议用户通过ComfyUI Manager安装缺失的节点,或者手动安装所需的扩展包。此外,可能需要检查节点包的依赖关系,比如PyTorch版本或者Python库是否满足要求。另外,用户提到的VACE和MoviiGen可能是指特定的工流程或模型,需要确认这些组件是否与14B模型兼容。如果节点是工者自定义的,可能需要联系者获取支持。同时,参考之前的引用中提到Wan2.1模型在阿里云上的应用,可能需要检查是否有相关的环境配置问题,比如CUDA版本或显存不足,但用户的问题更是节点缺失,所以重点应放在节点安装上。最后,应该分步骤给出解决方案,比如更新ComfyUI、通过管理器安装节点、手动安装、检查依赖、重新导入工流等,并提醒用户备份以防万一。同时生成相关问题,帮助用户进一步解决可能遇到的类似问题。</think>### VACE/MoviiGen/14B生视频工流中RHHiddenNodes节点缺失的解决方案 ### 一、问题定位 1. **节点类型定义缺失**:RHHiddenNodes属于自定义节点类型,通常由扩展包提供 2. **工流依赖关系断裂**:工者使用的第三方节点包未正确安装 3. **版本兼容性问题**:ComfyUI版本与节点包要求不匹配 ### 二、分步解决方案 #### 步骤1:检查基础环境 ```bash # 确认ComfyUI版本要求 python main.py --version # 验证PyTorch环境 python -c "import torch; print(torch.__version__)" ``` 建议使用CUDA 11.8+和PyTorch 2.0+环境[^1] #### 步骤2:安装缺失节点包 通过ComfyUI Manager执行: 1. 打开ComfyUI界面点击"Manager" 2. 在搜索栏输入"RHHiddenNodes" 3. 找到对应节点包点击"Install" #### 步骤3:手动安装扩展(备选方案) ```bash cd ComfyUI/custom_nodes git clone https://github.com/RockOfFire/ComfyUI-RH-HiddenNodes # 重启ComfyUI服务 ``` #### 步骤4:验证依赖关系 检查`requirements.txt`中的关键依赖: ``` torch>=2.0.1 torchvision>=0.15.2 opencv-python>=4.7.0 ``` ### 三、典型错误排查 | 错误类型 | 解决方案 | |---------|----------| | `ModuleNotFoundError` | 执行`pip install -r requirements.txt` | | CUDA内存不足 | 改用14B极速版模型[^1] | | 节点接口不匹配 | 检查工流JSON文件版本 | ### 四、工流修复技巧 1. 使用`Ctrl+Shift+V`导入工流时选择"自动修复节点" 2. 在节点属性面板点击"尝试加载备用实现" 3. 通过Alt+右键点击缺失节点进行替换
评论
添加红包

请填写红包祝福语或标题

红包个数最小为10个

红包金额最低5元

当前余额3.43前往充值 >
需支付:10.00
成就一亿技术人!
领取后你会自动成为博主和红包主的粉丝 规则
hope_wisdom
发出的红包
实付
使用余额支付
点击重新获取
扫码支付
钱包余额 0

抵扣说明:

1.余额是钱包充值的虚拟货币,按照1:1的比例进行支付金额的抵扣。
2.余额无法直接购买下载,可以购买VIP、付费专栏及课程。

余额充值