Bad quality according to sample_input

[confucianism72]

I tried with your default sample_input and got this: I think the quality is bad. Could you help me check where goes wrong?

I attached my config below:

model_version: sd21
model_base: base
use_lora: false
lora_path: ''
lora_alpha: null
ms_coco_path: ./sample_inputs/normal
prompt_type: blip_caption
batch_size: 4
resolution: 512
guidance_scale: 7.5
num_inference_steps: 50
prompt: ''
n_prompt: ', low quality, noisy, artifact, blurry, watermark'
drop_rate: 0.75
bg_blending_weight: 0.1
GPU_IDX: 7
result_dir: ./sample_outputs
exp_name: nc

[confucianism72]

I set batch size to 16 and only 1 of 16 has 2 buses in it as what described in prompt. Could you fix these?

[confucianism72]