AttributeError: module 'ldm.modules.encoders.modules' has no attribute 'FrozenCLIPEmbedder'

[sike2017]

Apologize if this is a noob question. When I run

python scripts/img2img.py --prompt "A fantasy landscape, trending on artstation" --init-img d:\code\Painter\astronaut_rides_horse.png --strength 0.8

I got output:

Global seed set to 42
Loading model from models/ldm/stable-diffusion-v1/model.ckpt
Global Step: 470000
LatentDiffusion: Running in eps-prediction mode
DiffusionWrapper has 859.52 M params.
making attention of type 'vanilla' with 512 in_channels
Working with z of shape (1, 4, 32, 32) = 4096 dimensions.
making attention of type 'vanilla' with 512 in_channels
Traceback (most recent call last):
  File "d:\code\stable-diffusion\scripts\img2img.py", line 293, in <module>
    main()
  File "d:\code\stable-diffusion\scripts\img2img.py", line 200, in main
    model = load_model_from_config(config, f"{opt.ckpt}")
  File "d:\code\stable-diffusion\scripts\img2img.py", line 34, in load_model_from_config
    model = instantiate_from_config(config.model)
  File "D:\ProgramData\Anaconda3\envs\dmodel2\lib\site-packages\ldm\util.py", line 78, in instantiate_from_config
    return get_obj_from_str(config["target"])(**config.get("params", dict()))
  File "D:\ProgramData\Anaconda3\envs\dmodel2\lib\site-packages\ldm\models\diffusion\ddpm.py", line 461, in __init__
    self.instantiate_cond_stage(cond_stage_config)
  File "D:\ProgramData\Anaconda3\envs\dmodel2\lib\site-packages\ldm\models\diffusion\ddpm.py", line 519, in instantiate_cond_stage
    model = instantiate_from_config(config)
  File "D:\ProgramData\Anaconda3\envs\dmodel2\lib\site-packages\ldm\util.py", line 78, in instantiate_from_config
    return get_obj_from_str(config["target"])(**config.get("params", dict()))
  File "D:\ProgramData\Anaconda3\envs\dmodel2\lib\site-packages\ldm\util.py", line 86, in get_obj_from_str
    return getattr(importlib.import_module(module, package=None), cls)
AttributeError: module 'ldm.modules.encoders.modules' has no attribute 'FrozenCLIPEmbedder'

[KleinXin]

Any one knows how to solve this issue？

[KleinXin]

It is said this issue is solved in this post

https://github.com/CompVis/stable-diffusion/issues/59

I tried the method they mentioned, but the problem is still unsolved.

[TheLastBen]

Copy the folder ldm from the main repo to the folder : scripts

[KleinXin]

Copy the folder ldm from the main repo to the folder : scripts

thx

After deleting /opt/conda/lib/python3.8/site-packages/ldm/ , copying ldm folder in the main repo to /opt/conda/lib/python3.8/site-packages/ldm/ and comment the 4th line of /opt/conda/lib/python3.8/site-packages/ldm/modules/encoders/modules.py where clip is imported, the demo can run successfully.

By the way, pytorch 1.11.0 with cuda 10.2 is needed.

@sike2017 You can also try this method.

[sike2017]

I’ve already solve this problem successfully. Thank you very much.

发件人: @.> 发送时间: 2022年9月23日 00:22:40 收件人: @.> 抄送: @.>; @.> 主题: Re: [CompVis/stable-diffusion] AttributeError: module 'ldm.modules.encoders.modules' has no attribute 'FrozenCLIPEmbedder' (Issue #190)

Copy the folder ldm from the main repo to the folder : scripts

thx

After deleting /opt/conda/lib/python3.8/site-packages/ldm/ , copying ldm folder in the main repo to /opt/conda/lib/python3.8/site-packages/ldm/ and comment the 4th line of /opt/conda/lib/python3.8/site-packages/ldm/modules/encoders/modules.py where clip is imported, the demo can run successfully.

By the way, pytorch 1.11.0 with cuda 10.2 is needed.

@sike2017https://github.com/sike2017 You can also try this method.

― Reply to this email directly, view it on GitHubhttps://github.com/CompVis/stable-diffusion/issues/190#issuecomment-1255263072, or unsubscribehttps://github.com/notifications/unsubscribe-auth/AIDG7CW6AKFTQI5QIK4H3RLV7SBU5ANCNFSM6AAAAAAQEVINFQ. You are receiving this because you were mentioned.Message ID: @.***>

[vfyang]

I found the problem: Upon upgrade (update the code), I ended up with two encoders folders: ./ldm/modules/encoders ./src/latent-diffusion/ldm/modules/encoders Bothe have modules.py, one is 8K the other 4K. Not consistent.

I solved the problem by deleting ./src/latent-diffusion/ldm/modules/encoders. (Actually copy the ./ldm/modules/encoders over ./src/latent-diffusion/ldm/modules/encoders also fix the problem.

It would be nice to fix the code, and make them consistent.

[vfyang]

With out the fix, the webui.py failed to load and quit.

With fixing it, the webui.py filed to load

[CyrusWang57]

我发现了问题： 升级（更新代码）后，我得到了两个编码器文件夹： ./ldm/modules/encoders ./src/latent-diffusion/ldm/modules/encoders两个 都有modules.py，一个是8K另一个4K。不一致。

我通过删除 ./src/latent-diffusion/ldm/modules/encoders 解决了这个问题。（实际上将 ./ldm/modules/encoders 复制到 ./src/latent-diffusion/ldm/modules/encoders 上也可以解决问题。

修复代码并使其保持一致会很好。

it works!!! thanks very much!!

[zsh2000]

For me, it is because some inconsistency of several code versions. For some versions, the class is called FrozenCLIPEmbedder, with "models" having default value "ViT-L/14". For others, it is called FrozenClipEmbedder, with no default value for "model".