The difference Results between the CODE and ComfyUI

[Xuan-World]

prompt: 'a pretty girl with long hair is having dinner with the other two people'

comfyui results:

code results:

same mask , same seed , same prompt ,same models .

I didn't modify the model configuration you adopted.

I wonder how this happens.

[Xuan-World]

Wow, I guess I make a step on it. I find the code here
result = pipe( prompt=prompt, height=size[1], width=size[0], control_image=image, control_mask=mask, num_inference_steps=28, generator=generator, controlnet_conditioning_scale=0.9, guidance_scale=3.5, negative_prompt="bad,ugly,deformed", true_guidance_scale=3.5 # default: 3.5 for alpha and 1.0 for beta ).images[0]

The true_guidance_scale is 1.0 in source CODE and 3.5 in ComfyUI. And Since the ComfyUI is better than CODE result, I change the code here to be 3.5. than I got the satisfying result below via code:

To prove my finding, I change the cfg-guider in ComfyUI and the awful result comes out as below.

So , what do you think about this question? @JPlin