Flux is generating squishy, warped images

[avst65t]

This is the real life quality I'm aiming to achieve:

But....

I don't know why in my case flux is generating squishy, warped images like this:

prompt: "A highly detailed headshot portrait from the chest up of ashley standing against a plain, neutral background. She has long wavy brown hair. She is wearing a black cashmere sweater. Her arms are crossed in front of her chest. She is standing at some distance. She is gazing directly at the camera with a engaging smile. The lighting should be soft and even, highlighting her facial features and adding subtle texture to the sweater. The atmosphere should convey calm confidence and introspection. Use a Nikon 80mm f/1.4G lens while ensuring soft and natural lighting to enhance the subject's features."

I used Replicate platform to train and run model via API. Here are my training parameters:
zip file
trigger word: ashley
model name
steps: 2000

Inference parameters:
lora_scale: 1
num_outputs: 1
aspect_ratio: "4:5" (also tried 1:1)
guidance scale: 2.5 (also tried 2, 3, 3.5)
output_quality: 100
extra_lora_scale: 1
num_inference_steps: 30 (tried 45)

I even removed lens information from prompt but still no luck. Why is this happening in my case? Community please help