Magic Prompt Enhancer Lora for Qwen 2.5 3B

[Quantum]

Magic Prompt Enhancer Lora

This lora has been trained to enhance image generation prompt and add more details to it.

Example:

Input Prompt: A white girl in a snow background

Enhanced Prompt: A white girl with long, curly hair is standing in a snow-covered field. She is wearing a red coat, a white scarf, and a pair of black boots. The background is filled with snow-covered trees and a few buildings. The sky is overcast, with a few clouds. The girl is looking up at the sky, with a peaceful expression on her face. The image has a soft, dreamy quality to it.


Base Model: Qwen 2.5 3B

How to use:

1. Install pip dependencies

Code: Copy

pip install -U transformers peft

2. Download & unzip LORA file

Hidden Content

You have not unlocked this post's content yet. Please reply to this thread to unlock the content.

3. Change base prompt and lora path and run python code

Code: Copy

from transformers import AutoModelForCausalLM, AutoTokenizer, AutoModel, set_seed
from peft import PeftModel
import torch
set_seed(12345) #Optional

prompt = "A white girl in a snow background"

model_name = "Qwen/Qwen2.5-3B"
lora_path = "/home/user/Downloads/prompt_enhancer_qwen2.5_3b"

model = AutoModelForCausalLM.from_pretrained(model_name, device_map="auto")
tokenizer = AutoTokenizer.from_pretrained(model_name, device_map="auto")

model = PeftModel.from_pretrained(model, lora_path)

text = """### Instruction:
Enhance this stable diffusion prompt. Expand on the given text and add more details to it.

### Input:
{}

### Response:
""".format(prompt)

print(text)

model_inputs = tokenizer([text], return_tensors="pt").to(model.device)

generated_ids = model.generate(
    **model_inputs,
    max_new_tokens=300
)
generated_ids = [
    output_ids[len(input_ids):] for input_ids, output_ids in zip(model_inputs.input_ids, generated_ids)
]

response = tokenizer.batch_decode(generated_ids, skip_special_tokens=True)[0]
print("Enhanced prompt:", response)