Hey there, I encountered an error when attempting to run the SD3 model on my Windows machine (p3.2xlarge EC2 instance that has 8 NVIDIA® V100 Tensor Core GPUs). Any ideas on what the issue might be? My machine already has the Nvidia driver installed, and this is the first time I’ve used ComfyUI, as I previously used WebUI.
Error occurred when executing CLIPTextEncode:
'NoneType' object has no attribute 'tokenize'
File "C:\Users\Administrator\Downloads\new_ComfyUI_windows_portable_nvidia_cu121_or_cpu\ComfyUI_windows_portable\ComfyUI\execution.py", line 151, in recursive_execute
output_data, output_ui = get_output_data(obj, input_data_all)
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "C:\Users\Administrator\Downloads\new_ComfyUI_windows_portable_nvidia_cu121_or_cpu\ComfyUI_windows_portable\ComfyUI\execution.py", line 81, in get_output_data
return_values = map_node_over_list(obj, input_data_all, obj.FUNCTION, allow_interrupt=True)
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "C:\Users\Administrator\Downloads\new_ComfyUI_windows_portable_nvidia_cu121_or_cpu\ComfyUI_windows_portable\ComfyUI\execution.py", line 74, in map_node_over_list
results.append(getattr(obj, func)(**slice_dict(input_data_all, i)))
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "C:\Users\Administrator\Downloads\new_ComfyUI_windows_portable_nvidia_cu121_or_cpu\ComfyUI_windows_portable\ComfyUI\nodes.py", line 57, in encode
tokens = clip.tokenize(text)
^^^^^^^^^^^^^
1 Like
I also didn’t use ComfyUI before, i notice it’s works only with sd3_medium_incl_clips.safetensors checkpoint from official repository https://huggingface.co/stabilityai/stable-diffusion-3-medium/tree/main, with any others i have same errors, even with official from civitai.
Update. I think I am find out. Checkpoint above doesn’t need a text encoder, for other it have to be. I donwloaded and put the text encoder in ‘clip’ folder. Then i returned conections by default in ComfyUI and it’s works for me with other checkpoints.
@DemonAlone could you kindly share how you downloaded and installed the encoder? Sorry I am completely new to SD/ComfyUI
The link in my preveously message. The folder ‘text_encoders’, you need three of those files: ‘clip_g.safetensors’, ‘clip_l.safetensors’ and ‘t5xxl_fp16.safetensor’ and the put them all in ComfyUI\models\clip. As i understood some models already have those text encoders some are not. Something similar to VAE.
1 Like
So I was onlhy about to find clip_l.safetensors but not the other two on huggingface, could you pls kindly share where you found them? huge thanks!
I checked my link right now, all three files are avialable.
Probably it is not actual anymore. You can use WebUI from Automatic1111 ‘sd3’ branch instead. I tried it too and it works, the same text encoders will download automatically.
Oh, I downloaded and put them in the folder as you instructed, but it’s still not working and I’m seeing the same error messages. 
Can you make a screenshot of your workflow?
it’s sd3_medium_example_workflow_basic.json, the only that I made was that I changed checkpoint to sd3_medium.safetensors
Still getting
Error occurred when executing CLIPTextEncode:
CUDA error: operation not supported
CUDA kernel errors might be asynchronously reported at some other API call, so the stacktrace below might be incorrect.
For debugging consider passing CUDA_LAUNCH_BLOCKING=1.
Compile with TORCH_USE_CUDA_DSA to enable device-side assertions.
Yes, but reason is different. If you use not a portable version of ComfyUI, try use from here: https://github.com/comfyanonymous/ComfyUI/releases
If even this won’t works, probably stay only option use ‘sd3’ version of WebUi…
