ComfyUI Framepack on RTX 4060 - Black Output & ‘clip missing: [‘text_projection.weight’]'

[mentsurfari]

Hi all, having trouble getting Framepack to work in ComfyUI on my RTX 4060 (8GB VRAM) with kijai’s ComfyUI-FramePackWrapper. I keep getting black output and the error: clip missing: [‘text_projection.weight’].

I’ve checked and the llava_llama3_fp8_scaled.safetensors and clip_l.safetensors files are in my ComfyUI/models/text_encoders folder. The Framepack model (…FramePackI2V_HY_fp8_e4m3fn.safetensors) is in ComfyUI/models/diffusion_models, sigclip (…sigclip_vision_patch14_384.safetensors) in ComfyUI/models/clip_vision, and the VAE (…hunyuan_video_vae_bf16.safetensors) in ComfyUI/models/vae.

Anyone else run into this ‘clip missing’ issue with Framepack and similar hardware, or have any thoughts on what might be the cause? Really looking forward to trying this out. Thanks!

[ltdrdata]

You can ignore

clip missing: ['text_projection.weight']

It simply occurs because a specific key required when loading the text encoder model in ComfyUI is missing from the model, but it generally does not affect functionality.

Black image output and the issue are irrelevant issues.

[mentsurfari]

Thank you so much for that! Any idea what could be causing the black output? I've tried cfg 0-7, steps 15-50, denoise 0.5-1, sampler unipc_bh1, different tile sizes...but nothing seems to work

[aliabougazia]

Did you find out the reason behind the black video? It was working fine for me and out of nowhere I got the black output.

[ltdrdata]

There are several possible causes for black image outputs. Sometimes the issue lies with the model itself, and other times it's due to a problem with the GPU driver. In many cases, switching to a different model or changing the PyTorch version can resolve the issue.

That's why it's important to first test using the standard usage method provided by the developer.