vace-comfyui-native-workflows-image-reference-keyframes-controlnet-video / workflows /translated_segments.json
Download workflows/translated_segments.json from AI-KSK/vace-comfyui-native-workflows-image-reference-keyframes-controlnet-video: direct link, hf CLI and curl.
- Browser
- Download file 18.8 kB
-
https://huggingface.co/AI-KSK/vace-comfyui-native-workflows-image-reference-keyframes-controlnet-video/resolve/main/workflows/translated_segments.json
- Command line
-
hf download hf://AI-KSK/vace-comfyui-native-workflows-image-reference-keyframes-controlnet-video/workflows/translated_segments.json
-
curl -L -o translated_segments.json https://huggingface.co/AI-KSK/vace-comfyui-native-workflows-image-reference-keyframes-controlnet-video/resolve/main/workflows/translated_segments.json
18.8 kB
| [ | |
| { | |
| "start": 0.24, | |
| "end": 11.36, | |
| "text": "These two days, ComfyUI officially added native support for VSC E video editing and model creation. We can see the related documentation and core logic implementation. In short, there hasn't been significant changes in the input fields.", | |
| "source_text": "这两天,ComfyUI 官方原生支持了 VSC E 视频编辑与创建模型,我们可以看到相关的支持文档,核心的逻辑编法,也就是说输入项上并没有特别大的变化。", | |
| "prompt_text": "这两天,ComfyUI 官方原生支持了 VSC E 视频编辑与创建模型,我们可以看到相关的支持文档,核心的逻辑编法,也就是说输入项上并没有特别大的变化。" | |
| }, | |
| { | |
| "start": 11.44, | |
| "end": 16.56, | |
| "text": "Based on the official example, I have updated all seven previous processes.", | |
| "source_text": "以官方的示例为基准,我将之前的七个流程都做了一个流程上的更新。", | |
| "prompt_text": "以官方的示例为基准,我将之前的七个流程都做了一个流程上的更新。" | |
| }, | |
| { | |
| "start": 16.88, | |
| "end": 29.28, | |
| "text": "Single image reference, multiple image references, control plus reference, first and last frames, video expansion, video reference editing, compared to KJ plugins, there may be slight differences in effect, but overall they are not significant.", | |
| "source_text": "单图参考,多图参考,控制加参考,首尾帧,视频扩展,视频的参考编辑和 KJ 的插件相比,效果上面来说可能有细微的差异,但总体来说不大。", | |
| "prompt_text": "单图参考,多图参考,控制加参考,首尾帧,视频扩展,视频的参考编辑和 KJ 的插件相比,效果上面来说可能有细微的差异,但总体来说不大。" | |
| }, | |
| { | |
| "start": 29.6, | |
| "end": 33.92, | |
| "text": "The smooth transition between the first and last frames in VSC E can lead to other processes later on.", | |
| "source_text": "VSC E 首尾帧这个地方的丝滑过渡,后面应该可以衍生出其他的流程。", | |
| "prompt_text": "VSC E 首尾帧这个地方的丝滑过渡,后面应该可以衍生出其他的流程。" | |
| }, | |
| { | |
| "start": 34.389, | |
| "end": 42.949, | |
| "text": "Now let's take a look at the workflow. I'm using the RunningHarb online ComfyUI platform here. Download the corresponding model, which is an official model from the ComfyUI website.", | |
| "source_text": "接着我们这里来看到工作流,我这里使用的是 RunningHarb 在线 comfyui 运行平台,下载对应的模型,那么这个模型使用的是 comfyui 官方的模型。", | |
| "prompt_text": "接着我们这里来看到工作流,我这里使用的是 RunningHarb 在线 comfyui 运行平台,下载对应的模型,那么这个模型使用的是 comfyui 官方的模型。" | |
| }, | |
| { | |
| "start": 43.189, | |
| "end": 53.589, | |
| "text": "Start with single image references. Positive and negative prompts are used. The positive prompt supports Chinese. Here, I'm using an English result for the reverse prompt. All input items are fed into this node.", | |
| "source_text": "先从单图参考看起,正反向提示词,正向提示词是支持中文的,我这里使用的反推是一个英语的结果,输入项都是汇入到这个节点当中去。", | |
| "prompt_text": "先从单图参考看起,正反向提示词,正向提示词是支持中文的,我这里使用的反推是一个英语的结果,输入项都是汇入到这个节点当中去。" | |
| }, | |
| { | |
| "start": 54.149, | |
| "end": 61.589, | |
| "text": "So we've talked about the input for the VAE model many times already—the input is either images or video masks.", | |
| "source_text": "那么对于 VAE 模型的输入项,我们已经聊过很多次了,视频的输入,一个是图像的输入,还有一个视频的遮罩。", | |
| "prompt_text": "那么对于 VAE 模型的输入项,我们已经聊过很多次了,视频的输入,一个是图像的输入,还有一个视频的遮罩。" | |
| }, | |
| { | |
| "start": 61.986, | |
| "end": 66.546, | |
| "text": "In terms of acceleration, we can use Cosvid to speed up the video.", | |
| "source_text": "增益上来说,我们可以使用 Cosvid 加速模型来对视频进行加速。", | |
| "prompt_text": "增益上来说,我们可以使用 Cosvid 加速模型来对视频进行加速。" | |
| }, | |
| { | |
| "start": 66.946, | |
| "end": 71.026, | |
| "text": "For these two nodes, CFG Zero Star and Enhance Video, I think they are suitable.", | |
| "source_text": "CFG Zero Star 和 Enhance Video 这两个节点来说,我觉得是适用的。", | |
| "prompt_text": "CFG Zero Star 和 Enhance Video 这两个节点来说,我觉得是适用的。" | |
| }, | |
| { | |
| "start": 71.426, | |
| "end": 75.906, | |
| "text": "I think T Cache and SART are not applicable in this process; they may reduce the effectiveness.", | |
| "source_text": "T Cache 和 SART 在这个流程当中,我认为不适用,会减损效果的表现。", | |
| "prompt_text": "T Cache 和 SART 在这个流程当中,我认为不适用,会减损效果的表现。" | |
| }, | |
| { | |
| "start": 76.306, | |
| "end": 78.146, | |
| "text": "Those who want to compile the model can do the compilation.", | |
| "source_text": "愿意编译模型的可以做编译。", | |
| "prompt_text": "愿意编译模型的可以做编译。" | |
| }, | |
| { | |
| "start": 78.466, | |
| "end": 81.826, | |
| "text": "The first time takes longer, so the second and third times will be shorter.", | |
| "source_text": "第一次的时间较长,那么第二次、第三次的时间就会变得更短。", | |
| "prompt_text": "第一次的时间较长,那么第二次、第三次的时间就会变得更短。" | |
| }, | |
| { | |
| "start": 82.306, | |
| "end": 85.346, | |
| "text": "The result from the K camera render is a separate reference.", | |
| "source_text": "进 K 摄像机渲染得到结果,那么这就是单独参考。", | |
| "prompt_text": "进 K 摄像机渲染得到结果,那么这就是单独参考。" | |
| }, | |
| { | |
| "start": 85.986, | |
| "end": 90.946, | |
| "text": "If we want to speed up, enable Cosmos, set CFG to one, and reduce the number of steps.", | |
| "source_text": "如果说我们要做加速,开启 Cosvid,将 CFG 设置为一,步数调低。", | |
| "prompt_text": "如果说我们要做加速,开启 Cosvid,将 CFG 设置为一,步数调低。" | |
| }, | |
| { | |
| "start": 91.094, | |
| "end": 98.214, | |
| "text": "Starting from 3 up, resolution can be increased moderately because the quality loss from the accelerated model allows us to bring it back up by increasing the resolution.", | |
| "source_text": "由三往上,分辨率可以适量提高,因为加速模型降低的质量,我们可以通过提高分辨率往回拉一些。", | |
| "prompt_text": "由三往上,分辨率可以适量提高,因为加速模型降低的质量,我们可以通过提高分辨率往回拉一些。" | |
| }, | |
| { | |
| "start": 98.694, | |
| "end": 106.854, | |
| "text": "But in my multiple tests this time, the native process's Cos Vid didn't feel as good as KJ's, and the dynamic performance became weaker.", | |
| "source_text": "但在我这一次多次测试当中,原生流程当中的 Cos Vid 的表现感觉没有 KJ 当中更好用,动态呢会变得偏弱。", | |
| "prompt_text": "但在我这一次多次测试当中,原生流程当中的 Cos Vid 的表现感觉没有 KJ 当中更好用,动态呢会变得偏弱。" | |
| }, | |
| { | |
| "start": 107.094, | |
| "end": 113.494, | |
| "text": "Go to the page for Running Hub or click the link in the explanation. The corresponding process for the video has been updated and is now available online for execution.", | |
| "source_text": "导航至 Running Hub 的页面,或者点击解说的链接,视频相对应的流程已经同步更新,可在线运行。", | |
| "prompt_text": "导航至 Running Hub 的页面,或者点击解说的链接,视频相对应的流程已经同步更新,可在线运行。" | |
| }, | |
| { | |
| "start": 113.494, | |
| "end": 121.494, | |
| "text": "The second process involves multi-image reference. In simple terms, it means combining two input images side by side.", | |
| "source_text": "第二个流程多图参考,多图参考这个地方,简而言之就是将我们的输入图像两张拼接在一块儿。", | |
| "prompt_text": "第二个流程多图参考,多图参考这个地方,简而言之就是将我们的输入图像两张拼接在一块儿。" | |
| }, | |
| { | |
| "start": 121.752, | |
| "end": 132.552, | |
| "text": "Then prompt it with a resolution of 800x800 while maintaining an aspect ratio divisible by 16. Under these conditions, the similarity between the two images remains very high.", | |
| "source_text": "然后对其进行提示,在宽高均为八百,保持纵横比能被十六整除的前提之下,这个双图参考的相似度还是非常的高的。", | |
| "prompt_text": "然后对其进行提示,在宽高均为八百,保持纵横比能被十六整除的前提之下,这个双图参考的相似度还是非常的高的。" | |
| }, | |
| { | |
| "start": 132.872, | |
| "end": 139.672, | |
| "text": "Because in the previous process, I was trying to achieve the result using 1080p and never specifically mentioned the image adjustment step.", | |
| "source_text": "因为在之前流程当中,我是在尝试使用七二零P来得到结果,从来没有具体提过图像调节这个节点。", | |
| "prompt_text": "因为在之前流程当中,我是在尝试使用七二零P来得到结果,从来没有具体提过图像调节这个节点。" | |
| }, | |
| { | |
| "start": 139.992, | |
| "end": 143.672, | |
| "text": "Maintaining the aspect ratio ensures our image doesn't become distorted.", | |
| "source_text": "保持纵横比,那么可以使我们的图像不会变得扭曲。", | |
| "prompt_text": "保持纵横比,那么可以使我们的图像不会变得扭曲。" | |
| }, | |
| { | |
| "start": 143.992, | |
| "end": 148.712, | |
| "text": "If the width and height constraints are within 800, it means there is a restriction on adjusting the longest side.", | |
| "source_text": "宽高约束为八百之内,则意味着对其最长边的调节约束。", | |
| "prompt_text": "宽高约束为八百之内,则意味着对其最长边的调节约束。" | |
| }, | |
| { | |
| "start": 149.112, | |
| "end": 153.352, | |
| "text": "Being divisible by sixteen is a characteristic of the Y2.1 model, regarding size requirements.", | |
| "source_text": "被十六整除是 Y 二点一模型的特性,尺寸上的要求。", | |
| "prompt_text": "被十六整除是 Y 二点一模型的特性,尺寸上的要求。" | |
| }, | |
| { | |
| "start": 153.912, | |
| "end": 164.312, | |
| "text": "Use a reference to cover part of the video. Then use an image as a reference to edit the video, focusing on the input video, the video mask, and the reference image.", | |
| "source_text": "参考编辑遮住视频的某一部分,然后呢,使用一张图像作为参考来编辑视频,还是着眼于输入像视频、视频的遮罩和参考图像。", | |
| "prompt_text": "参考编辑遮住视频的某一部分,然后呢,使用一张图像作为参考来编辑视频,还是着眼于输入像视频、视频的遮罩和参考图像。" | |
| }, | |
| { | |
| "start": 164.552, | |
| "end": 169.672, | |
| "text": "Then as long as you master these three aspects, everyone can make appropriate adjustments to our process.", | |
| "source_text": "那么只要能掌握这三项,大家完全可以在我们的流程上做出适当的改编。", | |
| "prompt_text": "那么只要能掌握这三项,大家完全可以在我们的流程上做出适当的改编。" | |
| }, | |
| { | |
| "start": 169.752, | |
| "end": 181.592, | |
| "text": "VCE is a video editing and creation model. Before, we used techniques like inversion to edit videos, but naturally, it's not as good as a model like VCE.", | |
| "source_text": "VCE 是视频的编辑与创建模型,视频的编辑之前我们都借助于反演等这样的技术来去编辑,自然是不如 VCE 这样的模型。", | |
| "prompt_text": "VCE 是视频的编辑与创建模型,视频的编辑之前我们都借助于反演等这样的技术来去编辑,自然是不如 VCE 这样的模型。" | |
| }, | |
| { | |
| "start": 181.858, | |
| "end": 197.298, | |
| "text": "Do it better by making changes only to this local area. The official documentation doesn't provide an example here, so I tried adjusting it using KJ's node. There's no problem with this approach because I'm only changing the input options without altering any actual conditions.", | |
| "source_text": "做的更好,可以保持只去改变局部手尾针手尾针这个地方官方是没有示例的,我尝试把它给变了一下,还是使用 KJ 的这个节点,用起来没有任何问题,因为这个地方我只是在改变它的输入项,不涉及到真正的条件上的变化。", | |
| "prompt_text": "做的更好,可以保持只去改变局部手尾针手尾针这个地方官方是没有示例的,我尝试把它给变了一下,还是使用 KJ 的这个节点,用起来没有任何问题,因为这个地方我只是在改变它的输入项,不涉及到真正的条件上的变化。" | |
| }, | |
| { | |
| "start": 197.618, | |
| "end": 208.178, | |
| "text": "Extending the video, the official documentation doesn't actually provide an example. So I made some adjustments based on KJ's node setup, added a mask, and then redrew the white areas.", | |
| "source_text": "视频的扩展,官方这个地方其实也没有给出示例,还是借用 KJ 的节点变了一下,将视频进行扩展,做出遮罩,然后进行白色部分的重绘。", | |
| "prompt_text": "视频的扩展,官方这个地方其实也没有给出示例,还是借用 KJ 的节点变了一下,将视频进行扩展,做出遮罩,然后进行白色部分的重绘。" | |
| }, | |
| { | |
| "start": 208.577, | |
| "end": 210.817, | |
| "text": "For images, one goes into the video, and the other into the mask.", | |
| "source_text": "图像来说,一个进视频,一个进遮罩。", | |
| "prompt_text": "图像来说,一个进视频,一个进遮罩。" | |
| }, | |
| { | |
| "start": 211.457, | |
| "end": 212.977, | |
| "text": "Can we use a reference image here?", | |
| "source_text": "那么这个地方可不可以有参考图像?", | |
| "prompt_text": "那么这个地方可不可以有参考图像?" | |
| }, | |
| { | |
| "start": 213.377, | |
| "end": 220.657, | |
| "text": "It's actually possible. We can use the first frame of the video as a reference, or any other image, allowing us to blend different results together.", | |
| "source_text": "其实可以,我们可以取视频的第一帧做参考,也可以取任意的一张图像做参考,允许我们去混合不同的结果。", | |
| "prompt_text": "其实可以,我们可以取视频的第一帧做参考,也可以取任意的一张图像做参考,允许我们去混合不同的结果。" | |
| }, | |
| { | |
| "start": 220.897, | |
| "end": 224.577, | |
| "text": "In terms of this result, I didn't see any seams, and it hasn't been tested multiple times yet.", | |
| "source_text": "就这个结果上而言,我是没有看到接缝的,还未多次测试。", | |
| "prompt_text": "就这个结果上而言,我是没有看到接缝的,还未多次测试。" | |
| }, | |
| { | |
| "start": 224.737, | |
| "end": 227.617, | |
| "text": "Those who have access to the flow can feel it themselves by combining control with reference.", | |
| "source_text": "拿到流的朋友可以自己感悟一下,控制加上参考。", | |
| "prompt_text": "拿到流的朋友可以自己感悟一下,控制加上参考。" | |
| }, | |
| { | |
| "start": 228.097, | |
| "end": 235.617, | |
| "text": "The video controlled here is guided by a pose video, also known as an OpenPose video, and it can be switched.", | |
| "source_text": "这个地方控制的视频是由一个姿态的视频进行引导,也就是一个 OpenPose 视频,可以做切换。", | |
| "prompt_text": "这个地方控制的视频是由一个姿态的视频进行引导,也就是一个 OpenPose 视频,可以做切换。" | |
| }, | |
| { | |
| "start": 235.777, | |
| "end": 237.777, | |
| "text": "I've explained this many times already, just as an example.", | |
| "source_text": "我这里已经讲了很多次了,打个样。", | |
| "prompt_text": "我这里已经讲了很多次了,打个样。" | |
| }, | |
| { | |
| "start": 237.814, | |
| "end": 242.294, | |
| "text": "For example, using depth can lead to significant changes in the outcome.", | |
| "source_text": "比如说使用深度,那么这样引导出来的结果变化将会非常的大。", | |
| "prompt_text": "比如说使用深度,那么这样引导出来的结果变化将会非常的大。" | |
| }, | |
| { | |
| "start": 242.854, | |
| "end": 259.334, | |
| "text": "The last streaming tattoo video shows some strange artifacts. Appropriately reducing the weight of this LoR model is an accelerated process. It uses five steps with a CFG value of one and a resolution of 720p. If you prioritize clarity and don't need high dynamic performance, you can use it.", | |
| "source_text": "最后一个流式纹身视频,可以看到视频当中有一些很奇怪的伪影,适当的去降低这个 LoR 模型的权重,这是一个加速的流程,用了五步 CFG 为一,七二零 P 的分辨率,那么对于清晰度有追求,对动态要求不大的朋友可以去使用它。", | |
| "prompt_text": "最后一个流式纹身视频,可以看到视频当中有一些很奇怪的伪影,适当的去降低这个 LoR 模型的权重,这是一个加速的流程,用了五步 CFG 为一,七二零 P 的分辨率,那么对于清晰度有追求,对动态要求不大的朋友可以去使用它。" | |
| }, | |
| { | |
| "start": 259.654, | |
| "end": 270.054, | |
| "text": "Official support from VSE actually has an important implication: it reduces the VRAM required for the process to run, which is great for users with lower-end hardware. However, it will make the process slower.", | |
| "source_text": "VSE 的官方支持,其实它有一个很重要的内涵,就会降低流程运行所需的显存,对我们的低配用户很好,只不过会让流程的运行变得更慢。", | |
| "prompt_text": "VSE 的官方支持,其实它有一个很重要的内涵,就会降低流程运行所需的显存,对我们的低配用户很好,只不过会让流程的运行变得更慢。" | |
| }, | |
| { | |
| "start": 270.162, | |
| "end": 270.802, | |
| "text": "Hmm.", | |
| "source_text": "嗯。", | |
| "prompt_text": "嗯。" | |
| } | |
| ] |