超硬核详细的 Comfy UI速成

angelosm

必须赞一个，楼主不要停止你的才华，我作为小白在ai上吃过你所说的亏，想不到在这里遇到这样的干货，还是很感动的，希望大神继续努力。

一二级没处理好

请问大佬我用comfyui跑图每跑一次就会增加硬盘内存过了几天后又会减少这是为什么？

Anooying

话说有模型分享吗，直接打包挂论坛卖就行，或者卖弄好的ui包这样

随便的小号

6g显卡无缘了

emby1478

大佬 最近N卡太坑了 请问A卡有可能么

emby1478

Gaslight001314 发表于 2025-11-10 22:44
这个我没太试过。。我现在在研究二次元涩图为主，暂时还没接触过换脸，但是如果是comfyUI的话，有一个rea ...

有人看有人看！大佬牛批

wq25432

看来又要出血了，N卡内存太小了  跑不了了

小白月

CUDA 错误：设备上无法执行内核镜像，有大哥知道这是怎么回事吗

rainbow57

大家伙都是为了满足自己的那一点嗜好才弄AI的吗？有没有学了Comfy UI当作生产力的，出来分享一下心得？

bluefat

Ninpoi19 发表于 2025-11-10 22:23
哥们，在涩情论坛教comfyui，你是真正滴英雄。话说换脸本地模型有什么好推荐的吗？现在还是insawp fp16（忘 ...

本地换脸试试facefusion，效果差点，胜在简单

我看你3cm不虚

我就说老王是学习网站吧

rgbvcd3

我最近学SD呢，听说SD比COMFYUI简单些，所以从简单的开始

松垮垮asd

大佬我为什么用cup.bat打开后，控制台加载一会就提示按任意键继续然后不跳网站，用CPU.bat打开后一下子就进网站了，这是什么原因啊大佬

松垮垮asd

大佬为什么我双击gup控制台加载完之后就提示我按任意键继续，按完之后也没有跳网站，而我用cup跑就一下进网站了

123456789okOK

大佬，可以分享一下工作流吗，我刚买回5070ti想试试

123456789okOK

连接主机...
连接主机成功
Welcome to Ubuntu 22.04.5 LTS (GNU/Linux 6.6.87.2-microsoft-standard-WSL2 x86_64)

• Documentation:  https://help.ubuntu.com

• Management:     https://landscape.canonical.com

• Support:        https://ubuntu.com/pro


This system has been minimized by removing packages and content that are
not required on a system that users do not log into.

To restore this content, you can run the 'unminimize' command.
Last login: Fri Dec 12 02:25:44 2025 from 172.18.0.1
(diffusion-pipe) root@1d615dc8ebd0:~# deepspeed --num_gpus=1 diffusion-pipe/train.py --deepspeed --config diffusion-pipe/examples/main_FLUX.1-dev.toml
[2025-12-12 03:17:03,522] [INFO] [real_accelerator.py:254:get_accelerator] Setting ds_accelerator to cuda (auto detect)
[2025-12-12 03:17:07,168] [WARNING] [runner.py:215:fetch_hostfile] Unable to find hostfile, will proceed with training with local resources only.
[2025-12-12 03:17:07,169] [INFO] [runner.py:605:main] cmd = /root/miniconda3/envs/diffusion-pipe/bin/python3.12 -u -m deepspeed.launcher.launch --world_info=eyJsb2NhbGhvc3QiOiBbMF19 --master_addr=127.0.0.1 --master_port=29500 --enable_each_rank_log=None diffusion-pipe/train.py --deepspeed --config diffusion-pipe/examples/main_FLUX.1-dev.toml
[2025-12-12 03:17:08,413] [INFO] [real_accelerator.py:254:get_accelerator] Setting ds_accelerator to cuda (auto detect)
[2025-12-12 03:17:10,671] [INFO] [launch.py:146:main] WORLD INFO DICT: {'localhost': [0]}
[2025-12-12 03:17:10,671] [INFO] [launch.py:152:main] nnodes=1, num_local_procs=1, node_rank=0
[2025-12-12 03:17:10,671] [INFO] [launch.py:163:main] global_rank_mapping=defaultdict(<class 'list'>, {'localhost': [0]})
[2025-12-12 03:17:10,671] [INFO] [launch.py:164:main] dist_world_size=1
[2025-12-12 03:17:10,671] [INFO] [launch.py:168:main] Setting CUDA_VISIBLE_DEVICES=0
[2025-12-12 03:17:10,672] [INFO] [launch.py:256:main] process 35732 spawned with command: ['/root/miniconda3/envs/diffusion-pipe/bin/python3.12', '-u', 'diffusion-pipe/train.py', '--local_rank=0', '--deepspeed', '--config', 'diffusion-pipe/examples/main_FLUX.1-dev.toml']
[2025-12-12 03:17:13,140] [INFO] [real_accelerator.py:254:get_accelerator] Setting ds_accelerator to cuda (auto detect)
[2025-12-12 03:17:16,324] [INFO] [comm.py:675:init_distributed] cdb=None
[2025-12-12 03:17:16,324] [INFO] [comm.py:706:init_distributed] Initializing TorchBackend in DeepSpeed with backend nccl
[rank0]: Traceback (most recent call last):
[rank0]:   File "/root/diffusion-pipe/train.py", line 303, in <module>
[rank0]:     model = flux.FluxPipeline(config)
[rank0]:             ^^^^^^^^^^^^^^^^^^^^^^^^^
[rank0]:   File "/root/diffusion-pipe/models/flux.py", line 183, in __init__
[rank0]:     transformer = diffusers.FluxTransformer2DModel.from_pretrained(
[rank0]:                   ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
[rank0]:   File "/root/miniconda3/envs/diffusion-pipe/lib/python3.12/site-packages/huggingface_hub/utils/_validators.py", line 114, in _inner_fn
[rank0]:     return fn(*args, **kwargs)
[rank0]:            ^^^^^^^^^^^^^^^^^^^
[rank0]:   File "/root/miniconda3/envs/diffusion-pipe/lib/python3.12/site-packages/diffusers/models/modeling_utils.py", line 1179, in from_pretrained
[rank0]:     resolved_model_file, sharded_metadata = _get_checkpoint_shard_files(
[rank0]:                                             ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
[rank0]:   File "/root/miniconda3/envs/diffusion-pipe/lib/python3.12/site-packages/diffusers/utils/hub_utils.py", line 393, in _get_checkpoint_shard_files
[rank0]:     raise FileNotFoundError(
[rank0]: FileNotFoundError: /root/diffusion/models/FLUX.1-dev/transformer/ does not appear to have a file named /root/diffusion/models/FLUX.1-dev/transformer/diffusion_pytorch_model-00001-of-00003.safetensors which is required according to the checkpoint index.
[rank0]:[W1212 03:17:18.789442383 ProcessGroupNCCL.cpp:1524] Warning: WARNING: destroy_process_group() was not called before program exit, which can leak resources. For more info, please see https://pytorch.org/docs/stable/distributed.html#shutdown (function operator())
[2025-12-12 03:17:20,674] [INFO] [launch.py:319:sigkill_handler] Killing subprocess 35732
[2025-12-12 03:17:20,674] [ERROR] [launch.py:325:sigkill_handler] ['/root/miniconda3/envs/diffusion-pipe/bin/python3.12', '-u', 'diffusion-pipe/train.py', '--local_rank=0', '--deepspeed', '--config', 'diffusion-pipe/examples/main_FLUX.1-dev.toml'] exits with return code = 1

liuxingyus1

a卡和i卡跑图大概也没问题，我用b580搞过本地跑图和本地翻译漫画都还好（b站都有教的，comf和sd都有整合包。sakura有i卡支持
就是没有方便的n卡一键安装包，一些插件支持的也不好

yjh941209

楼主下次发帖的时候加个收费附件就可以收到积分了，附件里面随便放个东西，愿意给的自己买就行了

mjbvxxd

瞬间上手，可能因为我有开发基础，大佬牛逼

雪上无霜

大佬，可以+WX或者QQ聊一聊这方面的技术吗？