{"id":55753,"date":"2026-08-13T09:25:25","date_gmt":"2026-08-13T01:25:25","guid":{"rendered":"https:\/\/www.1ai.net\/?p=55753"},"modified":"2026-08-10T17:58:50","modified_gmt":"2026-08-10T09:58:50","slug":"comfyui-%e4%bb%8e%e5%ae%89%e8%a3%85%e5%85%a5%e9%97%a8%e5%88%b0%e7%b2%be%e9%80%9a%ef%bc%8c%e6%a8%a1%e5%9e%8b%e9%80%89%e6%8b%a9%e5%ae%8c%e5%85%a8%e6%8c%87%e5%8d%97","status":"publish","type":"post","link":"https:\/\/www.1ai.net\/en\/55753.html","title":{"rendered":"ComfyUI from installation entrance to mastery, complete guide for model selection"},"content":{"rendered":"

\"b1ffc8e6j00tjjt1g00mbd000or00dtp\"<\/p>\n

Previous (Page 03)ComfyUI from installation entrance to mastery, model and plugin management (where\/where to go\/one key)<\/a> I taught you where to put the model file, where to go, how to use Manager as a key plug. This is the one where we solve the problem of a new player:Which one is it<\/strong><\/p>\n

I was stuck in this. Load up ComfyUI<\/a> When I opened up, there were thousands of models in the model station, names more than a bluff, what SDXL, Flux, Pony, and all kinds of XL, Turbo... I got a hot head, looked at which parameters were bigger, and the first one was a big blood-filled model, 4060 directly visible, and red-worded. And this is the way of the pit that I stepped upon myself, and then touched it clearly\u3002<\/p>\n

First, we'll start with the word \"model.\"<\/strong><\/p>\n

Many newcomers think that \"model\" is a file, but at least four things in ComfyUI are called \"model\" and it works differently. I used a metaphor for cooking:<\/p>\n\n\n\n\n\n\n\n
\n
typology<\/section>\n<\/td>\n
\n
What is it<\/section>\n<\/td>\n
\n
Size<\/section>\n<\/td>\n<\/tr>\n
Checkpoint Bottom Model<\/strong><\/td>\n\n
The whole menu, the whole thing<\/section>\n<\/td>\n
\n
TWO TO SEVEN<\/section>\n<\/td>\n<\/tr>\n
LoRA superheave<\/strong><\/td>\n\n
Add a spoonful of pepper noodles, fine-tune style\/role<\/section>\n<\/td>\n
\n
10 TO 300 MB<\/section>\n<\/td>\n<\/tr>\n
VAE DECODER<\/strong><\/td>\n\n
Translating subspace data into colour maps<\/section>\n<\/td>\n
\n
300MB<\/section>\n<\/td>\n<\/tr>\n
Embedding<\/strong><\/td>\n\n
Minitip patches, cold door concepts<\/section>\n<\/td>\n
\n
KBS<\/section>\n<\/td>\n<\/tr>\n<\/tbody>\n<\/table>\n

We're talking about this one\u00a0Checkpoint<\/strong>How, because the bottom model was wrong, it was all white. Lora and VAE are all about it\u3002<\/p>\n

Two, four major mainstream models. One watch<\/strong><\/p>\n\n\n\n\n\n\n\n
\n
Bottom Model<\/section>\n<\/td>\n
\n
Documentation<\/section>\n<\/td>\n
\n
Resolution<\/section>\n<\/td>\n
\n
display memory<\/section>\n<\/td>\n
\n
Good at it<\/section>\n<\/td>\n<\/tr>\n
SD1.5<\/strong><\/td>\n\n
TWO TO FOUR<\/section>\n<\/td>\n
\n
5122<\/section>\n<\/td>\n
\n
4GB<\/section>\n<\/td>\n
\n
Lightweight\/old LoRA\/Animation<\/section>\n<\/td>\n<\/tr>\n
SDXL<\/strong><\/td>\n\n
\u22486.9GB<\/section>\n<\/td>\n
\n
10242<\/section>\n<\/td>\n
\n
8GB<\/section>\n<\/td>\n
\n
General Paint Major<\/section>\n<\/td>\n<\/tr>\n
Ponny<\/strong><\/td>\n\n
6.5-6.9GB<\/section>\n<\/td>\n
\n
10242<\/section>\n<\/td>\n
\n
8GB<\/section>\n<\/td>\n
\n
Binary\/Role<\/section>\n<\/td>\n<\/tr>\n
Flux<\/strong><\/td>\n\n
IT'S FULL OF BLOOD<\/section>\n<\/td>\n
\n
1024-20482<\/section>\n<\/td>\n
\n
12GB+<\/section>\n<\/td>\n
\n
Best written\/renderable<\/section>\n<\/td>\n<\/tr>\n<\/tbody>\n<\/table>\n

Remember the two relationships, don't get stunned by names:<\/strong>Pony is not a new structure, it is fine-tuned on the basis of SDXL, which is basically the same as SDXL, but is more dichotomy. 2 Flux is another technical route (DiT), with a high paint-mass ceiling, but it is particularly visible, with new hands not to be covered in blood\u3002<\/p>\n

FOCUS: HOW TO SELECT THE 8GB DEPOSIT (RTX 4060)<\/strong><\/p>\n

THAT'S WHAT OUR READERS CARE ABOUT. MY OWN MACHINE IS 4060. HERE'S WHAT I CAN TELL YOU:<\/p>\n

1. SDXL IS THE DESSERT AREA\u3002<\/strong>A fp16 version of about 6.9GB, plus VAE and the temporary occupation of reasoning, allows 8GB to run 1024 x 1024, with a figure of about one or two minutes. The quality, speed, and visibility are best balanced and new hands close their eyes to SDXL\u3002<\/p>\n

Flux can run but slow, carefully chosen\u3002<\/strong>Blood full of Flux unopened 23GB, 8G direct blast. But it works\u00a0QUANTIFIED VERSION OF GGF<\/strong>: Q4 About 6.5-6.8GB, Q5 About 8GB. I've tried Q4 on 4060, but I have to drive-lowvram to throw some of the calculations to memory, and a picture takes more than a minute, and it's easy. Recommendations:Play SDXL and come back and try the Flux Quantification<\/strong>.<\/p>\n

Poney is suitable for a binary\u3002<\/strong>The 8GB is capable of 1024 x 1024, and many LoRAs are pressed. Mainly trying to draw animated characters and keep them in line, Ponny is more right than SDXL. Small pit: Pony has to create clip_skip=2 in CLIP Text Encode, otherwise it's easy to come up with a \"smuggle.\"\u3002<\/p>\n

4. SD1.5 FITS FOR FAST-TEST ERROR AND ANIMATION\u3002<\/strong>It's only 2-4GB, 4GB, and it's gonna run. The painting is not as good as SDXL\/Flux, but it is easy to quickly measure work flow and to make animated prototypes. I still have an SD1.5 as a test machine\u3002<\/p>\n

IV. Where best to go (domestic priority)<\/strong><\/p>\n\n\n\n\n\n\n
\n
Priority<\/section>\n<\/td>\n
\n
Site<\/section>\n<\/td>\n
\n
clarification<\/section>\n<\/td>\n<\/tr>\n
\n
National priorities<\/section>\n<\/td>\n
LiblibAI(mile)\/ GitCode AI<\/strong><\/td>\n\n
No magic, no registration, fast<\/section>\n<\/td>\n<\/tr>\n
\n
International stations<\/section>\n<\/td>\n
Civitai \/ HuggingFace<\/strong><\/td>\n\n
Best resource, network tools required<\/section>\n<\/td>\n<\/tr>\n
\n
Mirror<\/section>\n<\/td>\n
cf-miror.com<\/strong><\/td>\n\n
i'll replace it with a hugglingface.co<\/section>\n<\/td>\n<\/tr>\n<\/tbody>\n<\/table>\n

Next, the suffix is .safetensors or .ckpt, placed in omfyUI\/models\/checkpoints\/, returned to ComfyUI point Refresh can be seen in Load Checkpoint under Larry\u3002<\/p>\n

V. List of start-up hands<\/strong><\/p>\n

Don't come up and hoard dozens of models, and take the hard drive and pick the eye. I suggest you take these three first and cover most of the scenes:<\/p>\n

1.\u00a0SDXL BASE TEMPLATE<\/strong>(search sd_xl_base_1.0): general introduction, any style\u3002
\n2.\u00a0Pony Diffusion V6 XL<\/strong>(Checking the name): Specially a binary, identical role\u3002
\n3.\u00a0It's a real-life model<\/strong>(Checking Realistic Vision or Juggernaut XL such as SDXL is a written model): Use when trying to draw a true sense of hostage\u3002<\/p>\n

Just three, run the ComfyUI straight, and then slowly add it in line with the preferences. Models aren't many, they're working\u3002<\/p>\n

Six, I stepped on the pit. Save you two hours<\/strong><\/p>\n\n\n\n\n\n\n\n\n
\n
The pit<\/section>\n<\/td>\n
\n
Correct approach<\/section>\n<\/td>\n<\/tr>\n
\n
Blindly pursued Flux bursting memory<\/section>\n<\/td>\n
\n
8G Quantified version of GGF with SDXL or Flux<\/section>\n<\/td>\n<\/tr>\n
\n
SD1.5 HARD OUT 1024<\/section>\n<\/td>\n
\n
SD1.5 ONLY 512 FOR HIGH-QUALITY SDXL<\/section>\n<\/td>\n<\/tr>\n
\n
VAE MISSING\/MICROCOLOR BIAS<\/section>\n<\/td>\n
\n
VAE for SDXL\/SD1.5 is not universal, Flux uses its own VAE<\/section>\n<\/td>\n<\/tr>\n
\n
Ponny does not set clip_skip=2<\/section>\n<\/td>\n
\n
CLIP Text Encode set 2 and clear it up<\/section>\n<\/td>\n<\/tr>\n
\n
The LoRA set for SD1.5<\/section>\n<\/td>\n
\n
Lora, you need to match the base model. Don't mess with it<\/section>\n<\/td>\n<\/tr>\n<\/tbody>\n<\/table>\n

The model was chosen and done, but how do you connect the nodes into a pictureTitle 05<\/strong>I'll teach you how to read the work stream, build a water line myself, and tell you how to save and reuse the work stream you set\u3002<\/p>","protected":false},"excerpt":{"rendered":"

The last (Page 03) ComfyUI taught you where to put model files, where to go, how to use Manager as a key plugin, from installation to mastery, model and plugin management. This is the one where we solve the problem of a new player: which is the bottom model? I was stuck in this. As soon as the ComfyUI is opened, there are thousands of models in the model station, names more than a bluff, what SDXL, Flux, Pony, and all kinds of XL, Turbo... I'm hot in my head, looking at which parameters are bigger, and the first one is a big blood-filled model, 4060 directly visible, and the red word is painted. It's a hole I stepped on myself, and then I felt it<\/p>","protected":false},"author":1,"featured_media":0,"comment_status":"closed","ping_status":"closed","sticky":false,"template":"","format":"standard","meta":{"footnotes":""},"categories":[149,144],"tags":[1989,6276,4749],"collection":[],"class_list":["post-55753","post","type-post","status-publish","format-standard","hentry","category-jiaocheng","category-baike","tag-comfyui"],"acf":[],"_links":{"self":[{"href":"https:\/\/www.1ai.net\/en\/wp-json\/wp\/v2\/posts\/55753","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/www.1ai.net\/en\/wp-json\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/www.1ai.net\/en\/wp-json\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/www.1ai.net\/en\/wp-json\/wp\/v2\/users\/1"}],"replies":[{"embeddable":true,"href":"https:\/\/www.1ai.net\/en\/wp-json\/wp\/v2\/comments?post=55753"}],"version-history":[{"count":0,"href":"https:\/\/www.1ai.net\/en\/wp-json\/wp\/v2\/posts\/55753\/revisions"}],"wp:attachment":[{"href":"https:\/\/www.1ai.net\/en\/wp-json\/wp\/v2\/media?parent=55753"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/www.1ai.net\/en\/wp-json\/wp\/v2\/categories?post=55753"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/www.1ai.net\/en\/wp-json\/wp\/v2\/tags?post=55753"},{"taxonomy":"collection","embeddable":true,"href":"https:\/\/www.1ai.net\/en\/wp-json\/wp\/v2\/collection?post=55753"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}