import os
os.system(
'pip install --upgrade --pre --extra-index-url https://download.pytorch.org/whl/nightly/cu126 '
'"torch==2.9.0" "torchvision==0.24.0" "torchaudio==2.9.0" '
'"transformers>=4.44" "huggingface-hub>=1.0.0rc6" spaces -q'
)
import spaces
import gradio as gr
import torch
import math
from PIL import Image
from diffusers import QwenImageEditPlusPipeline, FlowMatchEulerDiscreteScheduler
import requests
import logging
import numpy as np
import random
from fastapi import FastAPI, HTTPException
logging.basicConfig(
level=logging.INFO,
filename="qwen_image_editor.log",
filemode="a",
format="%(asctime)s - %(levelname)s - %(message)s",
)
logger = logging.getLogger(__name__)
@spaces.GPU
def translate_albanian_to_english(text: str, language: str = "en"):
if not text.strip():
raise gr.Error("Please enter a description.")
for attempt in range(2):
try:
response = requests.post(
"https://hal1993-mdftranslation1234567890abcdef1234567890-fc073a6.hf.space/v1/translate",
json={"from_language": "sq", "to_language": "en", "input_text": text},
headers={"accept": "application/json", "Content-Type": "application/json"},
timeout=5,
)
response.raise_for_status()
translated = response.json().get("translate", "")
logger.info(f"Translation response: {translated}")
return translated
except Exception as e:
logger.error(f"Translation error (attempt {attempt + 1}): {e}")
if attempt == 1:
raise gr.Error("Translation failed. Please try again.")
raise gr.Error("Translation failed. Please try again.")
scheduler_config = {
"base_image_seq_len": 256,
"base_shift": math.log(3),
"invert_sigmas": False,
"max_image_seq_len": 8192,
"max_shift": math.log(3),
"num_train_timesteps": 1000,
"shift": 1.0,
"shift_terminal": None,
"stochastic_sampling": False,
"time_shift_type": "exponential",
"use_beta_sigmas": False,
"use_dynamic_shifting": True,
"use_exponential_sigmas": False,
"use_karras_sigmas": False,
}
scheduler = FlowMatchEulerDiscreteScheduler.from_config(scheduler_config)
pipeline = QwenImageEditPlusPipeline.from_pretrained(
"Qwen/Qwen-Image-Edit-2509",
scheduler=scheduler,
torch_dtype=torch.bfloat16,
)
pipeline.to("cuda")
pipeline.set_progress_bar_config(disable=None)
pipeline.load_lora_weights(
"lightx2v/Qwen-Image-Lightning",
weight_name="Qwen-Image-Lightning-8steps-V2.0-bf16.safetensors",
)
pipeline.fuse_lora()
MAX_SEED = np.iinfo(np.int32).max
QUALITY_PROMPT = ", high quality, detailed, vibrant, professional lighting"
@spaces.GPU(duration=60)
def edit_images(image1, image2, prompt):
if image1 is None or image2 is None:
raise gr.Error("Please upload both images")
prompt_en = translate_albanian_to_english(prompt.strip(), language="en")
prompt_final = prompt_en + QUALITY_PROMPT
if not isinstance(image1, Image.Image):
image1 = Image.fromarray(image1)
if not isinstance(image2, Image.Image):
image2 = Image.fromarray(image2)
seed = random.randint(0, MAX_SEED)
true_cfg_scale = 1.0
negative_prompt = ""
num_steps = 8
guidance_scale = 1.0
inputs = {
"image": [image1, image2],
"prompt": prompt_final,
"generator": torch.manual_seed(seed),
"true_cfg_scale": true_cfg_scale,
"negative_prompt": negative_prompt,
"num_inference_steps": num_steps,
"guidance_scale": guidance_scale,
"num_images_per_prompt": 1,
}
logger.info(f"Calling pipeline – Prompt: {prompt_final}")
logger.info(f"Seed: {seed} | Steps: {num_steps}")
with torch.inference_mode():
output = pipeline(**inputs)
return output.images[0]
def create_demo():
with gr.Blocks(css="", title="Qwen Image Editor") as demo:
gr.HTML(
"""
"""
)
with gr.Row(elem_id="general_items"):
gr.Markdown("# ")
gr.Markdown("Blend images together guided by a prompt description.", elem_id="subtitle")
with gr.Column(elem_id="input_column"):
image1_input = gr.Image(
label="First Image",
type="pil",
sources=["upload"],
interactive=True,
elem_classes=["gradio-component", "image-container"],
)
image2_input = gr.Image(
label="Second Image",
type="pil",
sources=["upload"],
interactive=True,
elem_classes=["gradio-component", "image-container"],
)
prompt_input = gr.Textbox(
label="Prompt",
placeholder="Describe how you want the images combined or edited...",
lines=3,
elem_classes=["gradio-component"],
)
run_button = gr.Button(
"Edit!",
variant="primary",
elem_classes=["gradio-component", "gr-button-primary"],
)
output_image = gr.Image(
label="Result Image",
type="pil",
interactive=False,
elem_classes=["gradio-component", "image-container"],
)
gr.on(
triggers=[run_button.click, prompt_input.submit],
fn=edit_images,
inputs=[image1_input, image2_input, prompt_input],
outputs=[output_image],
show_progress="full",
)
return demo
app = FastAPI()
demo = create_demo()
app.mount("/q3w4e5r6t7y8u9i0o1p2l3k4j5h6g7f8d9s0a1q2w3e4r5t6y7u8i9o0p1l2k3j4", demo.app)
@app.get("/{path:path}")
async def catch_all(path: str):
if not path.startswith("spaceishere"):
raise HTTPException(status_code=500, detail="Internal Server Error")
return demo
if __name__ == "__main__":
logger.info(f"Gradio version: {gr.__version__}")
demo.queue().launch(share=True)