ComfyUI User Guide

Chapter 1 ComfyUI is an open-source, free-of-charge, AI figure-out tool that puts the generation process in front of you by node-linking. And you can see it as an I.A. Lego in the drawing world: each function is a piece of wood, and you connect it with a line, and it's a complete line of production. Compared to other AI production tools: Pictures & Video & Audio can be done, with different workflows and models. Chapter 2

In the previous 18 articles, I told you about the machines, the Ventura, the TowerNet, LoRA, and a list of downgrades for you. And as soon as the article came out, the first person in the comment section said, "What's the use of video-generated RAM? I was a little bit surprised because he was right. For the first time in my life, I used ComfyUI to make a video, a five-second little clip, to run half the whole software card. Cut to the task manager and look -- 32G memory is full, the system kills the process..

In front of 17, you've got the ComfyUI, the map, the quality, the LoRA, the control Net. But there's this Cane, where almost every rookie hits, and I don't go around -- the first time I tried to run FLUX, I was so excited to drag into the flow of work, a little Queue Prompt, an interface card, a big red CUDA out of memory. I thought the card was broken, reloading, quenching, all night. And finally it turns out, it's not broken, it's visible that it doesn't..

The first 16 articles come down, and you've been able to do a high-level retrospect from the Zero, the Map, the Quality, the LoRA for style, the ControlNet for positions, the Ipadapter for faces. At this point, let's talk about a thing you'll bump into sooner or later: a self-defined node. The first time I stepped on this pit, I just learned how to figure it out. At the forum, a particularly gruesome work flow map was seen, and JSON was rushing into ComfyUI, and the entire canvas was red. Node is all "no node." I thought the software was broken..

The first 15 articles you basically got the ComfyUI: the machine, the drawing, the style change, the control, the face lock, the zooming up, the video, and the 15th ComfyUI, from installation to mastery, the low-visibility scenario: Boogu, the big model where 8G cards ran, the "breath-saving" Bougu model. But there's a "painting ceiling" that I never touched -- FLUX.1. It's generally recognized by those who play AI drawings that FLUX.1 shows people like skin..

14 articles in front of you basically got the ComfyUI: installed, schematic, style change, control, face lock, zoom in and even video. But I know that a lot of friends are stuck on one of the most realistic issues — lack of visibility. I was the victim of the 8G card. Watching the Flux show, full precision, all kinds of big models, it's like CUDA out of memory, and it's just falling apart. And then I realized, instead of being perfect, it's better to look for a "breath-saving" model and a quantitative play. We're talking about one I've been playing lately..

The first 13 articles come down, and you've got a whole set of static maps — machines, drawings, modulations, magnifications, locks. But there's one thing I've been keeping quiet about, because it's the easyest way to talk to 8G users: video. I gotta be honest, the first time I wanted to touch the video, the mind was, "Forget it, 8G." Queue out of memory. And then I knew, not eight Gs, I couldn't play, I used the wrong model, I didn't quantify it. Now ltX-2..

The first 12 articles have come down, and you've been able to install ComfyUI, comedyUI, modulate, correct errors, draw drawings, hang-up Lora, use ControlNet, take Depth control space, and lock your face with IPAdapter. There's almost nothing to stop you from doing this. But there's one problem, and I guess you hit it a long time ago -- the picture came out, and it went up in detail. Especially when we're users of 4060 8G, want to come up with a big 2048 x 2048? You'll see it in a minute

Title 11 ComfyUI, from installation to mastery, Depth dig deep + high-resolution magnification (supplied) you control space with Decth and know the ControlNet tube position. But there's a more annoying thing: a set of different scenes of the same character, the first round face, the second square face, the third straight hair change -- I've been staring at three pictures for half a day, not three separated twins. Then it became clear: ControlNet tube chart and pose, IPSASpte..

Book 10 ComfyUI, from installation to mastery, ControlNet, three scenes (in your direction) you learned the ControlNet Three Brothers, and you know howstrength is screwed. But then I left a pit unfilled: which are the three? I've stepped on my own real experience: I've thought about changing a real photo to Sabbunk style, using Canny. Turns out the wrinkles of the new clothes..

Book 09 ComfyUI, from installation to mastery, LoRA uses the field (selection/down/display/string weight) you have placed the LoRA skill pack on the bottom model, and the style, role and hand are steady. But there's one question that you're probably going to hit sooner or later: I want the picture to follow a reference, or I want two people standing one by the other and not running, and I don't know what the hint is. I used to make a "two handshake" with three hundred words on the tip, either back-to-back or hand-in-hand, and I had to redraw a dozen. Then I met..

The first eight articles come down, and you've changed style from a zero-packed ComfyUI, a map, a modulation, a self-rescue. By this, let's fill out the last piece of the puzzle -- Lora. Why the last piece? That's all you've got in front of you. To paint a watercolor, to fix the face of an animated character, to keep your hands off the "Six Fingers," it's hard to do it in a stable way with a background and a hint. Hanging a Lora, a few hundred gigabytes of paper, right now "a skill pack." My first..

In the first seven, you've been able to report mistakes from zero, graphics, quality, self-help. But there's a problem: every time you get a hard "squeeze" from a text phrase, you want a picture, you want a position, you can't describe it simply by words. It's time for meg2img. I found out that this function opened up the new world -- taking an off-the-shelf picture as "the bottom" and making AI change it 10 times more efficient than a hint from zero. This hand-and-hand piece tells you: what's going on with the drawings, what's going on with the noise intensity, three real-world scenes..

Previous (Page 06) ComfyUI, from installation entry to mastery, produced maps for quality improvement (resolution/step/sampler/CFG/high-resolution magnification) Parameters are adjusted and the paints are up. But playing AI drawings, who hasn't seen all the red words? When I first got wrong, I went straight down -- a lot of English was brushed down, it looked like the program went down and almost took the software off the reload. Then it turns out that ComfyUI's error is actually a "symptomatic book" and that most of the problems are solved by reading the last line of the red word. Some of the red ones that hit a rookie the most..

Last page, ComfyUI, from installation to mastery, Workstream Guide (read the node chain + build + save re-use), you put up the work stream, and the picture comes out. But does it feel like someone else's drawings are beautiful and clear, and your hair is gray, and your fingers are two more? I just got out of the graph, and I just pulled the parameters -- step to 100, CFG to 15, and I thought, "The bigger the better," and it went up to plastic. It was later known that the paints were not stacked with stacks of parameters, and that each was moved to the appropriate compartment. This one's the worst..

Last page, ComfyUI, from installation to mastery, complete guide to model selection. This is the most crucial step in the solution: how to connect a pile of diamonds into a visible picture. When I was using the ComfyUI, I saw people sharing a work stream like a book -- a line full of blocks and colours, from where I saw it. Until I had a heart method: to use the work stream as a water line for the plant, the raw materials go in from the left, the finished product comes out from the right, the line is the conveyor belt. This one taught you three things: to read the default workstream, to make one of your own, to save..

The last (Page 03) ComfyUI taught you where to put the model file, where to go, how to use Manager as a key plug-in. This is the one where we solve the problem of a new player: which is the bottom model? I was stuck in this. Once it's open, there's thousands of models in the model station, names more than a bluff, SDXL, Flux, Pony, and all kinds of XL, Turbo... I'm hot in my head, looking at which parameters are bigger, and..

Last page (No. 02) ComfyUI, from installation to mastery, knows everything about the interface, and from scratch, you finally see the nodal chain and you can figure it out. But to be honest, you had a bottom-up, zero-plug-in, and there was very little you could do. This is the one with the two "models" and "plug-ins." It's a real watershed for the ComfyUi from being able to run. When I was playing, I put down a bunch of models, all piled up in Mode..

There's one detail that many people don't know: double the resolution and double the presence. Many of my friends have recently asked me about ComfyUI's computer configuration, and this article is being read out today. Honestly, I've been asked that question dozens of times. Those who asked 3060 if they could run Flux, those who wanted to spend 20,000 with 4090 for SDXL, and those who struggled for half a day with AMD or NVIDIA. I spent most of the month running the nine models that are currently in the mainstream over the ComfyUI, documenting each model's visible occupancy, velocity, quantitative performance. Today..

open: how much is it going to cost to take a picture? let's say you're gonna shoot a set of scripts. the "french retro" package was selected, and the studio was set up for three hours, with a make-up all morning, and the final selection -- 20 fine-tuning and $2,800. and then you're gonna spit, "it's not like you're done with your face, but you're saying it's not good enough, and it's better than my own. " this money is spent, how to put it, it's not wrong, but it's not worth it. where is it? not the photo itself, the whole set..

Last page (Page 01) ComfyUI was installed from the first door of installation to the first of the five ways that ComfyUI was installed. But to be honest, at that point, you probably still had to "do the step, don't know what you're doing." This is the one that's going to work -- you get a good look at the interface, you know what every square is, and you can put it out of a blank canvas. Let's start with my embarrassment: I'm staring at the black box for half a day, afraid it's stuck, but it's loading the model. Co..

The question that was asked most of the time in the backstage was never "What's a good model?" Where's the download address? "Can I run this fucking computer?" How do you get the first picture after you've finished filling up those nodes? " Seven or eight such messages were received a day. I used to write ComfyUI, straight from q..

IPAdapter is a very powerful plugin with three main functions: First, face replacement. IPAdapter can accurately recognize the facial features of the characters in the reference map and introduce the features to the result map to realize the face replacement effect. The second is material migration. In addition to recognizing the facial features of the characters, IPAdapter can also recognize the material of the reference image and apply it to the result image. The last is style migration. In addition to the above two functions, IPAdapter is also able to recognize the style of the reference diagram and migrate the result to the result diagram...

Why is it that other people's prompts produce such beautiful, layered images that look great? And I can only type "1gril" even if I rack my brains. ...... In fact, the development of AI to the present, with DeepSeek, GPT so many language models, who still write their own cue words ah, to write the cue word this work to the AI model, not only write fast, and write much better than their own. We can fully embed DeepSeek into comfyUI, as long as the input keywords, DeepSeek will automatically help us...