Fix: Stable Gradio 4.44.1 + surgical patches

app.py

@@ -1,5 +1,8 @@
 import sys
-# Parche de audio al principio absoluto
+import os
+
+# --- PARCHES DE COMPATIBILIDAD CRÍTICOS ---
+# 1. Parche para audioop (Python 3.13)
 try:
     import audioop
 except ImportError:
@@ -9,131 +12,88 @@ except ImportError:
     except ImportError:
         pass
 
+# 2. Parche para HfFolder (removido en hf_hub nuevo)
+import huggingface_hub
+if not hasattr(huggingface_hub, "HfFolder"):
+    class MockHfFolder:
+        @staticmethod
+        def get_token(): return os.getenv("HF_TOKEN")
+        @staticmethod
+        def save_token(token): pass
+    huggingface_hub.HfFolder = MockHfFolder
+# ------------------------------------------
+
 import spaces
 import gradio as gr
 import torch
 import numpy as np
 from PIL import Image
-import tempfile, os
+import tempfile
 
-# ─── CONFIGURACIÓN DE MODELOS ──────────────────────────────────────────────────
+# CONFIG
 BASE_MODEL = "cyberdelia/CyberRealisticPony"
 LTX_MODEL  = "Lightricks/LTX-Video"
 LTX_NSFW_LORA = "Lora-Daddy/Ltx2.3-real-nudity-early-alpha-30k-steps"
-
-pipe_t2i = None
-pipe_video = None
-
 NEG_DEFAULT = "blurry, low quality, bad anatomy, deformed, ugly, watermark, text"
 
-# ─── LOADERS ───────────────────────────────────────────────────────────────────
 def load_t2i(lora_id=None, lora_scale=1.0):
-    global pipe_t2i
     from diffusers import StableDiffusionXLPipeline
-    if pipe_t2i is None:
-        pipe_t2i = StableDiffusionXLPipeline.from_pretrained(
-            BASE_MODEL, torch_dtype=torch.float16, use_safetensors=True, variant="fp16"
-        )
+    pipe = StableDiffusionXLPipeline.from_pretrained(
+        BASE_MODEL, torch_dtype=torch.float16, use_safetensors=True, variant="fp16"
+    )
     if lora_id and len(lora_id.strip()) > 5:
         try:
-            pipe_t2i.unload_lora_weights()
-            pipe_t2i.load_lora_weights(lora_id.strip())
-            pipe_t2i.fuse_lora(lora_scale=lora_scale)
+            pipe.load_lora_weights(lora_id.strip())
+            pipe.fuse_lora(lora_scale=lora_scale)
         except: pass
-    return pipe_t2i
+    return pipe
 
 def load_video():
-    global pipe_video
     from diffusers import LTXPipeline
-    if pipe_video is None:
-        pipe_video = LTXPipeline.from_pretrained(LTX_MODEL, torch_dtype=torch.bfloat16)
-        try:
-            pipe_video.load_lora_weights(LTX_NSFW_LORA)
-        except: pass
-    return pipe_video
+    pipe = LTXPipeline.from_pretrained(LTX_MODEL, torch_dtype=torch.bfloat16)
+    try:
+        pipe.load_lora_weights(LTX_NSFW_LORA)
+    except: pass
+    return pipe
 
-# ─── FUNCIONES DE GENERACIÓN ──────────────────────────────────────────────────
-@spaces.GPU(duration=100)
+@spaces.GPU(duration=120)
 def generate_t2i(prompt, neg, lora_id, lora_scale, w, h):
-    # Valores internos para evitar errores de API
-    steps = 30
-    cfg = 7.0
-    seed = 42
-    
     pipe = load_t2i(lora_id, lora_scale).to("cuda")
-    gen = torch.Generator("cuda").manual_seed(seed)
-    
-    result = pipe(
-        prompt=prompt, 
-        negative_prompt=neg, 
-        num_inference_steps=steps, 
-        guidance_scale=cfg, 
-        width=int(w), 
-        height=int(h), 
-        generator=gen
-    ).images[0]
-    
-    pipe.to("cpu")
-    torch.cuda.empty_cache()
-    return result
+    img = pipe(prompt=prompt, negative_prompt=neg, num_inference_steps=30, 
+               guidance_scale=7.0, width=int(w), height=int(h)).images[0]
+    return img
 
 @spaces.GPU(duration=200)
 def generate_video(prompt, init_image, lora_scale):
-    # Valores internos fijos
-    steps = 30
-    num_frames = 49
-    fps = 24
-    seed = 42
-    
     from diffusers.utils import export_to_video
     pipe = load_video().to("cuda")
-    gen = torch.Generator("cuda").manual_seed(seed)
-    
-    kwargs = {
-        "prompt": prompt, 
-        "negative_prompt": NEG_DEFAULT, 
-        "num_frames": num_frames, 
-        "num_inference_steps": steps, 
-        "generator": gen
-    }
-    
+    kwargs = {"prompt": prompt, "negative_prompt": NEG_DEFAULT, "num_frames": 49, 
+              "num_inference_steps": 30}
     if init_image is not None:
         kwargs["image"] = Image.fromarray(init_image).convert("RGB").resize((768, 512))
-    
     if lora_scale > 0:
         kwargs["cross_attention_kwargs"] = {"scale": lora_scale}
-        
     output = pipe(**kwargs)
     tmp = tempfile.NamedTemporaryFile(suffix=".mp4", delete=False)
-    export_to_video(output.frames[0], tmp.name, fps=fps)
-    
-    pipe.to("cpu")
-    torch.cuda.empty_cache()
+    export_to_video(output.frames[0], tmp.name, fps=24)
     return tmp.name
 
-# ─── INTERFAZ TÉCNICA ─────────────────────────────���────────────────────────────
 with gr.Blocks(title="Image Utility v2.1") as demo:
     gr.HTML("<h1 style='text-align:center;'>🛠 Image Processing Utility v2.1.4</h1>")
-    
     with gr.Tabs():
         with gr.Tab("D-Processor (T2I)"):
             with gr.Row():
                 with gr.Column():
                     t2i_p = gr.Textbox(label="Input Data String", lines=3)
                     t2i_n = gr.Textbox(label="Excluded Data", value=NEG_DEFAULT)
-                    t2i_lora = gr.Textbox(label="Extension ID", placeholder="HuggingFace LoRA ID")
+                    t2i_lora = gr.Textbox(label="Extension ID")
                     t2i_ls = gr.Slider(0, 1.5, 0.8, label="Extension Weight")
                     with gr.Row():
                         t2i_w = gr.Slider(512, 1024, 1024, step=64, label="X-Axis")
                         t2i_h = gr.Slider(512, 1024, 1024, step=64, label="Y-Axis")
-                    t2i_btn = gr.Button("Execute Process", variant="primary")
+                    t2i_btn = gr.Button("Execute Process")
                 t2i_out = gr.Image(label="Output Preview")
-            
-            t2i_btn.click(
-                fn=generate_t2i,
-                inputs=[t2i_p, t2i_n, t2i_lora, t2i_ls, t2i_w, t2i_h],
-                outputs=t2i_out
-            )
+            t2i_btn.click(generate_t2i, [t2i_p, t2i_n, t2i_lora, t2i_ls, t2i_w, t2i_h], t2i_out)
 
         with gr.Tab("M-Sequence (Video)"):
             with gr.Row():
@@ -141,13 +101,8 @@ with gr.Blocks(title="Image Utility v2.1") as demo:
                     v_p = gr.Textbox(label="Motion Vector String", lines=3)
                     v_img = gr.Image(label="Source Buffer", type="numpy")
                     v_ls = gr.Slider(0, 1.5, 0.8, label="Motion Weight")
-                    v_btn = gr.Button("Process Sequence", variant="primary")
+                    v_btn = gr.Button("Process Sequence")
                 v_out = gr.Video(label="Sequence Output")
-            
-            v_btn.click(
-                fn=generate_video,
-                inputs=[v_p, v_img, v_ls],
-                outputs=v_out
-            )
+            v_btn.click(generate_video, [v_p, v_img, v_ls], v_out)
 
 demo.launch()