comflyUI AI gen image on local PC

 

序

龜喵早在2023年12月教海豚如何用 comflyUI 在電腦 gen AI圖

無需靠VPN 飛去外國用 Gemini / GPT Image ，又買 tokens，賺了

安裝流程

ComflyUI > Download > Install

New Instance > Local > 

它自動detect到海豚的 Nvidia GPU

那些進階設定，沒事不要亂動

海豚都不清潔有什麼用

設定Workflow

Image > text to image

可以先 Skip & install AI models

安裝好workflow 介面後，

Template > Image > Text to Image

Z-Image-Turbo-FP8 是官網，推薦的  open-source AI model

成品

更多參數設定可看YT 教學

https://www.youtube.com/watch?v=hl7kwMwkLJU

改用其它 AI models

Place the Model File in the Correct Folder

下載放到 comflyUI 安裝路徑內的適當sub-folder

For all-in-one Checkpoints (Stable Diffusion 1.5, SDXL, etc. containing UNet, CLIP, and VAE combined):Put the file in: ComfyUI/models/checkpoints/

For standalone Diffusion Models / UNets (like FLUX or newer split architectures):Put the file in: ComfyUI/models/diffusion_models/ (or unet/ depending on version)

Refresh ComfyUI

Go back to your ComfyUI browser window.Click Refresh on the main floating menu panel so ComfyUI scans the new file.

Add the Loader Node to Your Workflow

Right-click anywhere on the empty canvas space and select Add Node.

Choose your loading node based on the model type:

For standard models: Select loader -> Load Checkpoint (connects MODEL, CLIP, and VAE outputs directly to your sampler and prompt nodes).

For split/modern models: Select model / advanced -> Load Diffusion Model (Note: You must also add separate text encoder/CLIP loaders like DualCLIPLoader and a Load VAE node to feed your sampler).

Replace or link the outputs of your new loader node into your existing KSampler or workflow chain.

load checkpoint

load Diffusion Model

Flux.1

另一個龜喵推薦的AI generation model，安裝文件

comflyUI 官方已有教學

ComfyUI Flux.1 Text-to-Image Workflow Example - ComfyUI

https://docs.comfy.org/tutorials/flux/flux-1-text-to-image#2-manual-model-installation

civitai

另外還要介紹那個網站

https://civitai.com/models

先看有沒有喜歡的作品，

有很多不同分類

再查一下別人用什麼 AI models

例如 REED_FYRRY_MiX_SDXL

就是向 Furry male character 功能特化了

它底層是用 NoobAI

REED_FYRRY_MiX_SDXL is my first attempt at training a Furry male character model using the NoobAI system. The overall style leans towards anime, but it also carries a certain level of realism in fur texture and body structure.

When I first used it, I felt that the images were very clean, with stable character proportions. The head-to-body ratio, limbs, and facial structure all look natural. The eyes are expressive with a calm yet powerful vibe, and the transition between fur and ears is smooth, making the character look both attractive and believable.

huggingface

內有很多AI models，算是AI models專用的github吧

查一下便找到

NoobAI XL 1.1

https://huggingface.co/Laxhar/noobai-XL-1.1?not-for-all-audiences=true

更多 Text-to-Image AI models

https://huggingface.co/models?pipeline_tag=text-to-image&sort=trending

Reference

Download Comfy Desktop - Run AI on Your Hardware

https://comfy.org/download

How to Use ComfyUI (Step-by-Step Tutorial)

https://www.youtube.com/watch?v=hl7kwMwkLJU

Hugging Face – The AI community building the future.

https://huggingface.co/