the result is a all black image ?
here is my infer code , is there has any wrong?
import os
from PIL import Image
import torch
from diffusers import QwenImageEditPipeline
pipeline = QwenImageEditPipeline.from_pretrained("ovedrive/qwen-image-edit-4bit")
print("pipeline loaded")
pipeline.to(torch.bfloat16)
pipeline.to("cuda")
pipeline.set_progress_bar_config(disable=None)
image = Image.open("./input/0.png").convert("RGB")
prompt = "Add a small wooden sign in the foreground in front of thepenguins with the text 'Welcome to Penquin Beach'."
inputs = {
"image": image,
"prompt": prompt,
"generator": torch.manual_seed(0),
"true_cfg_scale": 4.0,
"negative_prompt": " ",
"num_inference_steps": 50,
}
with torch.inference_mode():
output = pipeline(**inputs)
output_image = output.images[0]
output_image.save("output_image_edit.png")
print("image saved at", os.path.abspath("output_image_edit.png"))
You shouldnt use the vae from my repo. you need to use the original VAE. Only transformer and text_encoder are important. the rest you can use from the original Qwen-Image-Edit repo.
If that is something you can try and let me know that'd be great. I am not on my GPU Dev machine so I will post a working example in a day when i am back on my desk.
Btw check if you are using vae tiling, that causes issues with qwen . Don’t use it.
check updated sample code in the model card.
Thanks, I'll try it today！