Create app.py

app.py

@@ -0,0 +1,439 @@
+import subprocess
+subprocess.run(['sh', './spaces.sh'])
+
+import os
+
+os.putenv('PYTORCH_NVML_BASED_CUDA_CHECK','1')
+os.putenv('TORCH_LINALG_PREFER_CUSOLVER','1')
+alloc_conf_parts = [
+    'expandable_segments:True',
+    'pinned_use_background_threads:True'  # Specific to pinned memory.
+]
+os.environ['PYTORCH_CUDA_ALLOC_CONF'] = ','.join(alloc_conf_parts)
+os.environ["SAFETENSORS_FAST_GPU"] = "1"
+os.putenv('HF_HUB_ENABLE_HF_TRANSFER','1')
+
+import spaces
+
+import gradio as gr
+import numpy as np
+import random
+
+
+
+import torch
+torch.backends.cuda.matmul.allow_tf32 = False
+torch.backends.cuda.matmul.allow_bf16_reduced_precision_reduction = False
+torch.backends.cuda.matmul.allow_fp16_reduced_precision_reduction = False
+torch.backends.cudnn.allow_tf32 = False
+torch.backends.cudnn.deterministic = False
+torch.backends.cudnn.benchmark = False
+torch.backends.cuda.preferred_blas_library="cublas"
+torch.backends.cuda.preferred_linalg_library="cusolver"
+torch.set_float32_matmul_precision("highest")
+
+from diffusers import StableDiffusion3Pipeline, SD3Transformer2DModel, AutoencoderKL
+from transformers import CLIPTextModelWithProjection, T5EncoderModel
+from transformers import CLIPTokenizer, T5TokenizerFast
+
+import re
+import paramiko
+import urllib
+import time
+from image_gen_aux import UpscaleWithModel
+from huggingface_hub import hf_hub_download
+import datetime
+import cyper
+from PIL import Image
+
+#from accelerate import Accelerator
+
+#accelerator = Accelerator(mixed_precision="bf16")
+
+hftoken = os.getenv("HF_AUTH_TOKEN") 
+
+code = r'''
+import torch
+import paramiko
+import os
+import socket
+import threading # NEW IMPORT
+import queue     # NEW IMPORT
+FTP_HOST = 'noahcohn.com'
+FTP_USER = 'ford442'
+FTP_PASS = os.getenv("FTP_PASS")
+FTP_DIR = 'img.noahcohn.com/stablediff/'
+FTP_HOST_FALLBACK = '1ink.us'
+FTP_DIR_FALLBACK = 'img.1ink.us/stablediff/'
+
+# --- WORKER FUNCTION FOR THREADING ---
+# This function contains the logic to connect to a single host.
+# It will be executed by each of our threads.
+def connect_worker(host, result_queue):
+    """Tries to connect to a single host and puts the successful transport object into the queue."""
+    transport = None
+    try:
+        transport = paramiko.Transport((host, 22))
+        # We still use the 5-second timeout for the handshake
+        transport.start_client(timeout=5)
+        transport.auth_password(username=FTP_USER, password=FTP_PASS)
+        
+        # If we reach here, the connection was successful.
+        # Put the result in the queue for the main thread to use.
+        print(f"✅ Connection to {host} succeeded first.")
+        result_queue.put(transport)
+    except (paramiko.SSHException, socket.timeout, EOFError) as e:
+        # This is an expected failure, just print a note.
+        print(f"ℹ️ Connection to {host} failed or was too slow: {e}")
+        if transport:
+            transport.close()
+    except Exception as e:
+        # Handle any other unexpected errors.
+        print(f"❌ Unexpected error connecting to {host}: {e}")
+        if transport:
+            transport.close()
+            
+def upload_to_ftp(filename):
+    """
+    Attempts to connect to two FTP hosts simultaneously and uses the first one that responds.
+    It now uses a corresponding directory for the primary and fallback hosts.
+    """
+    hosts = [FTP_HOST]
+    if FTP_HOST_FALLBACK:
+        hosts.append(FTP_HOST_FALLBACK)
+    
+    result_queue = queue.Queue()
+    threads = []
+    
+    print(f"--> Racing connections to {hosts} for uploading {filename}...")
+    for host in hosts:
+        thread = threading.Thread(target=connect_worker, args=(host, result_queue))
+        thread.daemon = True
+        thread.start()
+        threads.append(thread)
+    try:
+        winning_transport = result_queue.get(timeout=7)
+        
+        # --- THIS IS THE NEW LOGIC ---
+        # 1. Determine which host won the race.
+        winning_host = winning_transport.getpeername()[0]
+        
+        # 2. Select the correct destination directory based on the winning host.
+        #    If the fallback directory isn't specified, it safely defaults to the primary directory.
+        if winning_host == FTP_HOST:
+            destination_directory = FTP_DIR
+        else:
+            destination_directory = FTP_DIR_FALLBACK if FTP_DIR_FALLBACK else FTP_DIR
+        print(f"--> Proceeding with upload to {winning_host} in directory {destination_directory}...")
+        # 3. Construct the full destination path using the selected directory.
+        sftp = paramiko.SFTPClient.from_transport(winning_transport)
+        destination_path = os.path.join(destination_directory, os.path.basename(filename))
+        sftp.put(filename, destination_path)
+        
+        print(f"✅ Successfully uploaded {filename}.")
+        
+        sftp.close()
+        winning_transport.close()
+    except queue.Empty:
+        print("❌ Critical Error: Neither FTP host responded in time.")
+    except Exception as e:
+        print(f"❌ An unexpected error occurred during SFTP operation: {e}")
+'''
+
+pyx = cyper.inline(code, fast_indexing=True, directives=dict(boundscheck=False, wraparound=False, language_level=3))
+
+device = torch.device("cuda:0" if torch.cuda.is_available() else "cpu")
+
+#vae=AutoencoderKL.from_pretrained("ford442/sdxl-vae-bf16", use_safetensors=True, subfolder='vae',token=True)
+vaeX=AutoencoderKL.from_pretrained("ford442/stable-diffusion-3.5-large-fp32", safety_checker=None, use_safetensors=True, subfolder='vae', low_cpu_mem_usage=False, torch_dtype=torch.float32, token=True)
+pipe = StableDiffusion3Pipeline.from_pretrained(
+    #"stabilityai  #  stable-diffusion-3.5-large",
+    "ford442/stable-diffusion-3.5-large-bf16",
+    trust_remote_code=True,
+    #vae=None,
+    #vae=AutoencoderKL.from_pretrained("ford442/stable-diffusion-3.5-large-fp32", use_safetensors=True, subfolder='vae',token=True),
+     #scheduler = FlowMatchHeunDiscreteScheduler.from_pretrained('ford442/stable-diffusion-3.5-large-bf16', subfolder='scheduler',token=True),
+    #text_encoder=None, #CLIPTextModelWithProjection.from_pretrained("ford442/stable-diffusion-3.5-large-bf16", subfolder='text_encoder', token=True),
+    # text_encoder=CLIPTextModelWithProjection.from_pretrained("ford442/stable-diffusion-3.5-large-bf16", subfolder='text_encoder', token=True),
+    #text_encoder_2=None, #CLIPTextModelWithProjection.from_pretrained("ford442/stable-diffusion-3.5-large-bf16", subfolder='text_encoder_2',token=True),
+  #  text_encoder_2=CLIPTextModelWithProjection.from_pretrained("ford442/stable-diffusion-3.5-large-bf16", subfolder='text_encoder_2',token=True),
+    #text_encoder_3=None, #T5EncoderModel.from_pretrained("ford442/stable-diffusion-3.5-large-bf16", subfolder='text_encoder_3',token=True),
+  #  text_encoder_3=T5EncoderModel.from_pretrained("ford442/stable-diffusion-3.5-large-bf16", subfolder='text_encoder_3',token=True),
+    #tokenizer=CLIPTokenizer.from_pretrained("ford442/stable-diffusion-3.5-large-bf16", add_prefix_space=True, subfolder="tokenizer", token=True),
+    #tokenizer_2=CLIPTokenizer.from_pretrained("ford442/stable-diffusion-3.5-large-bf16", add_prefix_space=True, subfolder="tokenizer_2", token=True),
+    transformer=None,
+    #tokenizer_3=T5TokenizerFast.from_pretrained("ford442/stable-diffusion-3.5-large-bf16", add_prefix_space=False, use_fast=True, subfolder="tokenizer_3", token=True),
+    #torch_dtype=torch.bfloat16,
+    use_safetensors=True,
+)
+#text_encoder=CLIPTextModelWithProjection.from_pretrained("ford442/stable-diffusion-3.5-large-bf16", subfolder='text_encoder', token=True).to(torch.device("cuda:0"), dtype=torch.bfloat16)
+#text_encoder_2=CLIPTextModelWithProjection.from_pretrained("ford442/stable-diffusion-3.5-large-bf16", subfolder='text_encoder_2',token=True).to(torch.device("cuda:0"), dtype=torch.bfloat16)
+#text_encoder_3=T5EncoderModel.from_pretrained("ford442/stable-diffusion-3.5-large-bf16", subfolder='text_encoder_3',token=True).to(torch.device("cuda:0"), dtype=torch.bfloat16)
+ll_transformer=SD3Transformer2DModel.from_pretrained("ford442/stable-diffusion-3.5-large-bf16", subfolder='transformer',token=True).to(torch.device("cuda:0"), dtype=torch.bfloat16)
+pipe.transformer=ll_transformer
+pipe.load_lora_weights("ford442/sdxl-vae-bf16", weight_name="LoRA/UltraReal.safetensors")
+
+#pipe.to(accelerator.device)
+pipe.to(device=device, dtype=torch.bfloat16)
+
+upscaler_2 = UpscaleWithModel.from_pretrained("Kim2091/ClearRealityV1").to(torch.device('cuda'))
+
+MAX_SEED = np.iinfo(np.int32).max
+
+MAX_IMAGE_SIZE = 4096
+
+@spaces.GPU(duration=70)
+def infer_60(
+    prompt,
+    negative_prompt_1,
+    negative_prompt_2,
+    negative_prompt_3,
+    width,
+    height,
+    guidance_scale,
+    num_inference_steps,
+    progress=gr.Progress(track_tqdm=True),
+):
+    seed = random.randint(0, MAX_SEED)
+    generator = torch.Generator(device='cuda').manual_seed(seed)
+    print('-- generating image --')
+    sd_image = pipe(
+            prompt=prompt,
+            prompt_2=prompt,
+            prompt_3=prompt,
+            negative_prompt=negative_prompt_1,
+            negative_prompt_2=negative_prompt_2,
+            negative_prompt_3=negative_prompt_3,
+            guidance_scale=guidance_scale,
+            num_inference_steps=num_inference_steps,
+            width=width,
+            height=height,
+            generator=generator,
+            max_sequence_length=512
+    ).images[0]
+    print('-- got image --')
+    timestamp = datetime.datetime.now().strftime("%Y%m%d_%H%M%S")
+    sd35_path = f"sd35ll_{timestamp}.png"
+    sd_image.save(sd35_path,optimize=False,compress_level=0)
+    pyx.upload_to_ftp(sd35_path)
+    with torch.no_grad():
+        upscale = upscaler_2(sd_image, tiling=True, tile_width=256, tile_height=256)
+        upscale2 = upscaler_2(upscale, tiling=True, tile_width=256, tile_height=256)
+    print('-- got upscaled image --')
+    downscale = upscale2.resize((upscale2.width // 4, upscale2.height // 4),Image.LANCZOS)
+    upscale_path = f"sd35ll_upscale_{timestamp}.png"
+    downscale.save(upscale_path,optimize=False,compress_level=0)
+    pyx.upload_to_ftp(upscale_path)
+    return sd_image, prompt
+
+@spaces.GPU(duration=100)
+def infer_90(
+    prompt,
+    negative_prompt_1,
+    negative_prompt_2,
+    negative_prompt_3,
+    width,
+    height,
+    guidance_scale,
+    num_inference_steps,
+    progress=gr.Progress(track_tqdm=True),
+):
+    seed = random.randint(0, MAX_SEED)
+    generator = torch.Generator(device='cuda').manual_seed(seed)
+    print('-- generating image --')
+    sd_image = pipe(
+            prompt=prompt,
+            prompt_2=prompt,
+            prompt_3=prompt,
+            negative_prompt=negative_prompt_1,
+            negative_prompt_2=negative_prompt_2,
+            negative_prompt_3=negative_prompt_3,
+            guidance_scale=guidance_scale,
+            num_inference_steps=num_inference_steps,
+            width=width,
+            height=height,
+            generator=generator,
+            max_sequence_length=512
+    ).images[0]
+    print('-- got image --')
+    timestamp = datetime.datetime.now().strftime("%Y%m%d_%H%M%S")
+    sd35_path = f"sd35ll_{timestamp}.png"
+    sd_image.save(sd35_path,optimize=False,compress_level=0)
+    pyx.upload_to_ftp(sd35_path)
+    with torch.no_grad():
+        upscale = upscaler_2(sd_image, tiling=True, tile_width=256, tile_height=256)
+        upscale2 = upscaler_2(upscale, tiling=True, tile_width=256, tile_height=256)
+    print('-- got upscaled image --')
+    downscale = upscale2.resize((upscale2.width // 4, upscale2.height // 4),Image.LANCZOS)
+    upscale_path = f"sd35ll_upscale_{timestamp}.png"
+    downscale.save(upscale_path,optimize=False,compress_level=0)
+    pyx.upload_to_ftp(upscale_path)
+    return sd_image, prompt
+
+@spaces.GPU(duration=120)
+def infer_110(
+    prompt,
+    negative_prompt_1,
+    negative_prompt_2,
+    negative_prompt_3,
+    width,
+    height,
+    guidance_scale,
+    num_inference_steps,
+    progress=gr.Progress(track_tqdm=True),
+):
+    seed = random.randint(0, MAX_SEED)
+    generator = torch.Generator(device='cuda').manual_seed(seed)
+    print('-- generating image --')
+    sd_image = pipe(
+            prompt=prompt,
+            prompt_2=prompt,
+            prompt_3=prompt,
+            negative_prompt=negative_prompt_1,
+            negative_prompt_2=negative_prompt_2,
+            negative_prompt_3=negative_prompt_3,
+            guidance_scale=guidance_scale,
+            num_inference_steps=num_inference_steps,
+            width=width,
+            height=height,
+            generator=generator,
+            max_sequence_length=512
+    ).images[0]
+    print('-- got image --')
+    timestamp = datetime.datetime.now().strftime("%Y%m%d_%H%M%S")
+    sd35_path = f"sd35ll_{timestamp}.png"
+    sd_image.save(sd35_path,optimize=False,compress_level=0)
+    pyx.upload_to_ftp(sd35_path)
+    with torch.no_grad():
+        upscale = upscaler_2(sd_image, tiling=True, tile_width=256, tile_height=256)
+        upscale2 = upscaler_2(upscale, tiling=True, tile_width=256, tile_height=256)
+    print('-- got upscaled image --')
+    downscale = upscale2.resize((upscale2.width // 4, upscale2.height // 4),Image.LANCZOS)
+    upscale_path = f"sd35ll_upscale_{timestamp}.png"
+    downscale.save(upscale_path,optimize=False,compress_level=0)
+    pyx.upload_to_ftp(upscale_path)
+    return sd_image, prompt
+    
+css = """
+#col-container {margin: 0 auto;max-width: 640px;}
+body{background-color: blue;}
+"""
+
+with gr.Blocks(theme=gr.themes.Origin(),css=css) as demo:
+    with gr.Column(elem_id="col-container"):
+        gr.Markdown(" # StableDiffusion 3.5 Large with UltraReal lora test")
+        expanded_prompt_output = gr.Textbox(label="Prompt", lines=1)  # Add this line
+        with gr.Row():
+            prompt = gr.Text(
+                label="Prompt",
+                show_label=False,
+                max_lines=1,
+                placeholder="Enter your prompt",
+                container=False,
+            )
+            run_button_60 = gr.Button("Run 60", scale=0, variant="primary")
+            run_button_90 = gr.Button("Run 90", scale=0, variant="primary")
+            run_button_110 = gr.Button("Run 110", scale=0, variant="primary")
+        result = gr.Image(label="Result", show_label=False)
+        with gr.Accordion("Advanced Settings", open=True):
+            negative_prompt_1 = gr.Text(
+                label="Negative prompt 1",
+                max_lines=1,
+                placeholder="Enter a negative prompt",
+                visible=True,
+                value="bad anatomy, poorly drawn hands, distorted face, blurry, out of frame, low resolution, grainy, pixelated, disfigured, mutated, extra limbs, bad composition"
+            )
+            negative_prompt_2 = gr.Text(
+                label="Negative prompt 2",
+                max_lines=1,
+                placeholder="Enter a second negative prompt",
+                visible=True,
+                value="unrealistic, cartoon, anime, sketch, painting, drawing, illustration, graphic, digital art, render, 3d, blurry, deformed, disfigured, poorly drawn, bad anatomy, mutated, extra limbs, ugly, out of frame, bad composition, low resolution, grainy, pixelated, noisy, oversaturated, undersaturated, (worst quality, low quality:1.3), (bad hands, missing fingers:1.2)"
+            )
+            negative_prompt_3 = gr.Text(
+                label="Negative prompt 3",
+                max_lines=1,
+                placeholder="Enter a third negative prompt",
+                visible=True,
+                value="(worst quality, low quality:1.3), (bad anatomy, bad hands, missing fingers, extra digit, fewer digits:1.2), (blurry:1.1), cropped, watermark, text, signature, logo, jpeg artifacts, (ugly, deformed, disfigured:1.2), (poorly drawn:1.2), mutated, extra limbs, (bad proportions, gross proportions:1.2), (malformed limbs, missing arms, missing legs, extra arms, extra legs:1.2), (fused fingers, too many fingers, long neck:1.2), (unnatural body, unnatural pose:1.1), out of frame, (bad composition, poorly composed:1.1), (oversaturated, undersaturated:1.1), (grainy, pixelated:1.1), (low resolution, noisy:1.1), (unrealistic, distorted:1.1), (extra fingers, mutated hands, poorly drawn hands, bad hands:1.3), (missing fingers:1.3)"
+            )
+            num_iterations = gr.Number(
+                value=1000, 
+                label="Number of Iterations")
+            with gr.Row():
+                width = gr.Slider(
+                    label="Width",
+                    minimum=256,
+                    maximum=MAX_IMAGE_SIZE,
+                    step=32,
+                    value=768,
+                )
+                height = gr.Slider(
+                    label="Height",
+                    minimum=256,
+                    maximum=MAX_IMAGE_SIZE,
+                    step=32,
+                    value=768,
+                )
+                guidance_scale = gr.Slider(
+                    label="Guidance scale",
+                    minimum=0.0,
+                    maximum=30.0,
+                    step=0.1,
+                    value=4.2,
+                )
+                num_inference_steps = gr.Slider(
+                    label="Number of inference steps",
+                    minimum=1,
+                    maximum=500,
+                    step=1,
+                    value=100,
+                )
+        gr.on(
+        triggers=[run_button_60.click, prompt.submit],
+        fn=infer_60,
+        inputs=[
+            prompt,
+            negative_prompt_1,
+            negative_prompt_2,
+            negative_prompt_3,
+            width,
+            height,
+            guidance_scale,
+            num_inference_steps,
+        ],
+        outputs=[result, expanded_prompt_output],
+        )
+        gr.on(
+        triggers=[run_button_90.click, prompt.submit],
+        fn=infer_90,
+        inputs=[
+            prompt,
+            negative_prompt_1,
+            negative_prompt_2,
+            negative_prompt_3,
+            width,
+            height,
+            guidance_scale,
+            num_inference_steps,
+        ],
+        outputs=[result, expanded_prompt_output],
+        )
+        gr.on(
+        triggers=[run_button_110.click, prompt.submit],
+        fn=infer_110,
+        inputs=[
+            prompt,
+            negative_prompt_1,
+            negative_prompt_2,
+            negative_prompt_3,
+            width,
+            height,
+            guidance_scale,
+            num_inference_steps,
+        ],
+        outputs=[result, expanded_prompt_output],
+        )
+
+if __name__ == "__main__":
+    demo.launch()