🧩 ComfyUI Workflows: Bring a Cartoon to LifeComfyUI 工作流：让卡通活起来

1 A workflow is a recipe (again!)工作流是一张配方卡（又来了！）

If you did the Dockerfile lesson, this will feel familiar: a Dockerfile is a recipe for packing software boxes, and a ComfyUI workflow is a recipe for making art. Same superpower, different kitchen:

如果你学过 Dockerfile 课程，这里会似曾相识：Dockerfile 是打包软件盒子的配方，ComfyUI 工作流是创作艺术的配方。同样的超能力，不同的厨房：

We're using the same ComfyUI you met in Part 1 — these screenshots happen to come from a small GPU server running the heavier video models, but the canvas, the nodes and every idea below are identical on your machine.

我们用的就是你在第一部分认识的那个 ComfyUI——这些截图恰好来自一台跑着较重视频模型的小型 GPU 服务器，但画布、节点和下面的每一个概念，在你的电脑上完全一样。

2 The empty studio空荡荡的工作室

Start a new workflow (Workflow menu → New, or the + tab) and you get a blank dark canvas. Quick tour of what surrounds it:

新建一个工作流（Workflow 菜单 → New，或 + 标签页），你会得到一块空白的深色画布。先认认它周围都是什么：

• Pan: drag empty canvas (or hold Space). Zoom: scroll wheel.
• Fit everything on screen: press . (the period key) — you'll use this constantly.
• Undo works: Ctrl/⌘ + Z, just like a document.

• 平移：拖动空白画布（或按住空格）。缩放：滚轮。
• 全图适配屏幕：按 .（句号键）——你会一直用它。
• 撤销有效：Ctrl/⌘ + Z，和写文档一样。

3 Adding nodes: double-click and type添加节点：双击，然后打字

The fastest way to add any node: double-click empty canvas and start typing. The search understands partial words — type wan image and every node related to Wan (the video model family we'll use) appears, with a preview of its inputs on the right:

添加任何节点最快的方法：双击空白画布，直接打字。搜索支持部分匹配——输入 wan image，所有和 Wan（我们要用的视频模型家族）相关的节点都会出现，右侧还带输入输出预览：

4 The cartoon-motion workflow at a glance卡通动画工作流总览

Here is the complete workflow we'll run — 11 nodes that turn one still image into a video clip. Read it like a sentence, left to right:

这就是我们要运行的完整工作流——11 个节点，把一张静图变成一段视频。像读句子一样，从左到右读：

Eleven nodes sounds like a lot, but they form just four zones:

11 个节点听起来很多，但它们只构成四个区域：

5 The magic node: anchoring frame 1 to your cartoon魔法节点：把第 1 帧锚定到你的卡通

One node makes this image-to-video instead of text-to-video: Wan22ImageToVideoLatent. Your picture goes in as start_image — and becomes the literal first frame of the video. Everything the AI dreams must flow out of that frame, so your character's design survives.

有一个节点让这一切变成图生视频而不是文生视频：Wan22ImageToVideoLatent。你的图片作为 start_image 进去——成为视频名副其实的第一帧。AI 梦出的一切都必须从那一帧流出，所以你的角色设计得以保留。

6 Using it: press Run and watch the line light up使用它：按 Run，看流水线亮起来

Press the blue Run button. Your workflow joins the queue, and ComfyUI executes node by node — each one highlights green as it works, and the top bar shows total progress:

按下蓝色 Run 按钮。你的工作流加入队列，ComfyUI 逐个节点执行——正在干活的节点亮绿色，顶栏显示总进度：

• You can keep editing the canvas while it runs — the queue took a snapshot of your graph.
• Queue several runs with different seeds, walk away, come back to a litter of videos.
• The X button next to Run cancels the current job.

• 运行时你可以继续编辑画布——队列已经对你的图拍了快照。
• 用不同的种子排队好几个任务，去干别的，回来收获一窝视频。
• Run 旁边的 X 按钮取消当前任务。

7 The payoff: paws on the heart收获：爪子放在心口

When the last node finishes, the Video Combine node grows a little player showing your clip, and the MP4 lands in ComfyUI's output/ folder:

最后一个节点完成时，Video Combine 节点会长出一个小播放器显示你的片段，MP4 落进 ComfyUI 的 output/ 文件夹：

And here is the real before → after from this exact session — one still picture in, and out comes a cat who stands up, presses her paws to her heart and beams at you:

这就是本次操作的真实前后对比——一张静图进去，出来一只站起来、把爪子贴在心口、对你灿烂微笑的猫：

8 Editing: where the craft lives修改：手艺所在

A downloaded workflow is a starting point — editing it is how you make it yours. The two places you'll edit most, side by side:

下载来的工作流只是起点——把它改成你的才是真本事。最常修改的两个地方，正好并排：

✍️ Edit 1 — the motion prompt修改 1——动作提示词

Our positive prompt asked her to "stand up and walk forward, hind legs clearly lifting off the ground…" — and the model had a warmer idea: she stood up, pressed her paws to her heart and smiled. We kept it (full story in Edit 2). Change the motion words and re-run: jumps with joy, waves at the camera, turns her head and blinks. Describe the motion, not the character — the character comes from your image.

我们的正向提示词让她"站起来向前走，后腿明显抬离地面……"——而模型有个更暖的主意：她站起来，把爪子贴在心口，笑了。我们留下了这个镜头（完整故事见修改 2）。换掉动作词再跑一次：开心地跳起来、对镜头挥手、转头眨眼。描述动作，别描述角色——角色来自你的图片。

🚫 Edit 2 — the negative prompt (two true stories)修改 2——负向提示词（两个真实故事）

The negative box has oddly specific lines like extra leg, third leg, multiple legs… When we first ran walking clips, the model kept growing a ghostly third leg mid-stride. Naming the exact failure in the negative prompt fixed it. That's what negative prompts are for: when the model makes the same mistake twice, describe the mistake and it learns to steer away.

负向框里有些特别具体的词，比如 extra leg, third leg, multiple legs…（多余的腿、第三条腿……）。我们最早跑走路视频时，模型总在迈步中途长出一条幽灵般的第三条腿。把这个失败原原本本写进负向提示词，就治好了。负向提示词就是干这个的：模型把同一个错犯了两次，就把错误描述出来，它会学着绕开。

And the clip in step 7? Take 1 used a plain walking prompt — the cat just sat there paddling her front paws, back legs never leaving the ground. So take 2 added "stands up" to the prompt and banned the failure in the negative ("legs glued to the ground, motionless back legs"). She stood up alright — then, instead of walking, she put her paws on her heart and beamed. It was better than what we'd asked for, so it became the goal. Two lessons in one: name the failure you saw, and when the model improvises a better take than your plan — you're the director. Keep it.

那第 7 步的片段呢？第一镜用了普通的走路提示词——猫坐在原地划前爪，后腿根本不离地。于是第二镜在提示词里加上"站起来"，并把看到的失败禁进负向（"腿粘在地上、后腿不动"）。她确实站起来了——然后没有走路，而是把爪子放在心口，笑容满面。这比我们要求的更好，于是它成了目标。一举两课：把你看到的失败说出名字；当模型即兴发挥出比你计划更好的镜头——你是导演，留下它。

🎛️ Edit 3 — the KSampler dials修改 3——KSampler 的旋钮

9 Saving & sharing workflows保存与分享工作流

Your workflow deserves to survive. The workflow-actions menu (click the workflow name / Graph button) has everything:

你的工作流值得活下去。工作流操作菜单（点工作流名字 / Graph 按钮）里什么都有：

• Loading is drag-and-drop: drop a workflow JSON anywhere on the canvas and the whole factory rebuilds itself.
• Hidden superpower: images that ComfyUI saves carry their workflow inside the PNG. Drag a ComfyUI-made picture onto the canvas → the exact workflow that made it appears. Every output is its own recipe card.
• Sites like comfyworkflows.com and the built-in Templates browser are full of workflows to study — load one, find the KSampler, trace backwards.

• 加载就是拖放：把工作流 JSON 拖到画布上任何位置，整座工厂自己重建。
• 隐藏超能力：ComfyUI 保存的图片把工作流藏在 PNG 内部。把一张 ComfyUI 做的图拖上画布 → 做出它的那个工作流原样出现。每件作品都是自己的配方卡。
• comfyworkflows.com 这类网站和内置的 Templates 浏览器里全是可以研究的工作流——加载一个，找到 KSampler，往回追。

{
  "1": {"class_type":"UNETLoader","inputs":{"unet_name":"wan2.2_ti2v_5B_fp16.safetensors","weight_dtype":"default"}},
  "2": {"class_type":"ModelSamplingSD3","inputs":{"model":["1",0],"shift":8.0}},
  "3": {"class_type":"CLIPLoader","inputs":{"clip_name":"umt5_xxl_fp8_e4m3fn_scaled.safetensors","type":"wan"}},
  "4": {"class_type":"VAELoader","inputs":{"vae_name":"wan2.2_vae.safetensors"}},
  "5": {"class_type":"CLIPTextEncode","inputs":{"clip":["3",0],"text":"the cartoon cat stands up and walks forward on all four legs, hind legs clearly lifting off the ground and stepping one at a time, a complete four-legged walking cycle, paws visibly rising and falling, full body always in frame, fluid coherent leg movement, flat 2D cartoon style, clean line art, masterpiece, best quality, stable consistent anatomy"}},
  "6": {"class_type":"CLIPTextEncode","inputs":{"clip":["3",0],"text":"legs glued to the ground, motionless back legs, static hind legs, sliding without stepping, extra leg, third leg, multiple legs, duplicate legs, extra limbs, duplicated limbs, merging legs, fused legs, limb ghosting, deformed legs, bad anatomy, overexposed, static, blurry, low quality, worst quality, jpeg artifacts, watermark, text, flickering, motionless"}},
  "7": {"class_type":"LoadImage","inputs":{"image":"your_character.png"}},
  "8": {"class_type":"Wan22ImageToVideoLatent","inputs":{"vae":["4",0],"width":1280,"height":704,"length":49,"batch_size":1,"start_image":["7",0]}},
  "9": {"class_type":"KSampler","inputs":{"model":["2",0],"positive":["5",0],"negative":["6",0],"latent_image":["8",0],"seed":22222,"steps":30,"cfg":5.0,"sampler_name":"uni_pc","scheduler":"simple","denoise":1.0}},
  "10": {"class_type":"VAEDecode","inputs":{"samples":["9",0],"vae":["4",0]}},
  "11": {"class_type":"VHS_VideoCombine","inputs":{"images":["10",0],"frame_rate":24,"loop_count":0,"filename_prefix":"happy_cat","format":"video/h264-mp4","pix_fmt":"yuv420p","crf":19,"save_metadata":false,"pingpong":false,"save_output":true}}
}

10 When things go wrong出问题的时候

Everyone hits these. Here's the most common one, caught in the wild:

人人都会遇到这些。最常见的一个，现场抓拍：