Update app.py

small fix from init app.py

app.py

@@ -1,161 +1,131 @@
-# app.py
 import gradio as gr
 import numpy as np
 import random
+
+# import spaces #[uncomment to use ZeroGPU]
 from diffusers import DiffusionPipeline
-from diffusers import (
-    DDIMScheduler,
-    PNDMScheduler,
-    LMSDiscreteScheduler,
-    EulerDiscreteScheduler,
-    DPMSolverMultistepScheduler,
-)
 import torch
+from typing import Optional
 
-device = "cuda" if torch.cuda.is_available() else "cpu"
-MAX_SEED = np.iinfo(np.int32).max
-MAX_IMAGE_SIZE = 1024
-
+# кэш для пайплайнов (чтобы не перезагружать модель при каждом запросе)
+PIPE_CACHE: dict[str, DiffusionPipeline] = {}
 DEFAULT_MODEL = "CompVis/stable-diffusion-v1-4"
 MODEL_OPTIONS = [
     "CompVis/stable-diffusion-v1-4",
+    "stabilityai/stable-diffusion-2-1",
     "stabilityai/sdxl-turbo",
-    # add other model ids you want to expose here
 ]
 
-SCHEDULER_MAP = {
-    "default": None,
-    "DDIM": DDIMScheduler,
-    "PNDM": PNDMScheduler,
-    "LMS": LMSDiscreteScheduler,
-    "Euler": EulerDiscreteScheduler,
-    "DPMSolver": DPMSolverMultistepScheduler,
-}
-
-
-def get_torch_dtype():
-    return torch.float16 if torch.cuda.is_available() else torch.float32
 
+device = "cuda" if torch.cuda.is_available() else "cpu"
+model_repo_id = "stabilityai/sdxl-turbo"  # Replace to the model you would like to use
 
-def load_pipeline(model_id: str, scheduler_name: str = "default"):
-    """Load pipeline from pretrained model_id and optionally replace scheduler."""
-    torch_dtype = get_torch_dtype()
-    pipe = DiffusionPipeline.from_pretrained(model_id, torch_dtype=torch_dtype)
-    # try to replace scheduler if requested
-    sched_cls = SCHEDULER_MAP.get(scheduler_name)
-    if sched_cls is not None:
-        try:
-            pipe.scheduler = sched_cls.from_config(pipe.scheduler.config)
-        except Exception:
-            # fallback to default if replacement failed
-            pass
-    pipe = pipe.to(device)
-    return pipe
-
-
-# preload default pipeline (may take time on startup)
-print(f"Loading default model {DEFAULT_MODEL} ...")
-try:
-    default_pipe = load_pipeline(DEFAULT_MODEL, "default")
-    print("Loaded default model.")
-except Exception as e:
-    default_pipe = None
-    print("Failed to preload default model:", e)
-
-css = """
-#col-container {
-    margin: 0 auto;
-    max-width: 880px;
-}
-"""
+if torch.cuda.is_available():
+    torch_dtype = torch.float16
+else:
+    torch_dtype = torch.float32
 
-examples = [
-    "Astronaut in a jungle, cold color palette, muted colors, detailed, 8k",
-    "An astronaut riding a green horse",
-    "A delicious ceviche cheesecake slice",
-]
+def get_pipe(model_id: str):
+    if model_id in PIPE_CACHE:
+        return PIPE_CACHE[model_id]
+    # загружаем и кэшируем
+    p = DiffusionPipeline.from_pretrained(model_id, torch_dtype=torch_dtype)
+    p = p.to(device)
+    PIPE_CACHE[model_id] = p
+    return p
 
-def load_model_and_update(model_id, scheduler_name):
-    """Called when user selects a model or scheduler: load and return new pipeline + status."""
-    try:
-        pipe = load_pipeline(model_id, scheduler_name)
-        return pipe, f"Loaded `{model_id}` (scheduler: {scheduler_name})"
-    except Exception as e:
-        return None, f"Error loading `{model_id}`: {e}"
 
+# pipe = DiffusionPipeline.from_pretrained(model_repo_id, torch_dtype=torch_dtype)
+# pipe = pipe.to(device)
 
-def size_to_dims(size_str):
-    try:
-        w, h = map(int, size_str.split("x"))
-        # clamp to limits
-        w = min(max(256, w), MAX_IMAGE_SIZE)
-        h = min(max(256, h), MAX_IMAGE_SIZE)
-        return gr.Slider.update(value=w), gr.Slider.update(value=h)
-    except Exception:
-        return gr.Slider.update(value=512), gr.Slider.update(value=512)
+MAX_SEED = np.iinfo(np.int32).max
+MAX_IMAGE_SIZE = 1024
 
 
+# @spaces.GPU #[uncomment to use ZeroGPU]
 def infer(
-    prompt,
-    negative_prompt,
-    seed,
-    randomize_seed,
-    width,
-    height,
-    guidance_scale,
-    num_inference_steps,
-    pipe_state,  # gr.State containing pipeline
+    model_id: Optional[str] = DEFAULT_MODEL,
+    prompt: str = "",
+    negative_prompt: str = "",
+    seed: int = 42,
+    randomize_seed: bool = False,
+    width: int = 512,
+    height: int = 512,
+    guidance_scale: float = 7.0,
+    num_inference_steps: int = 20,
+    scheduler_name: Optional[str] = None,
     progress=gr.Progress(track_tqdm=True),
 ):
-    if pipe_state is None:
-        return None, seed, "Model not loaded."
+    # получаем/загружаем нужный pipe
+    pipe = get_pipe(model_id)
+
+    # при желании можно подменить scheduler по имени (опционально)
+    if scheduler_name:
+        # примерная схема: словарь name->класс scheduler
+        # при необходимости добавить другие scheduler'ы — импортируйте их сверху и добавьте сюда
+        try:
+            from diffusers import DDIMScheduler, EulerAncestralDiscreteScheduler, PNDMScheduler
+            sched_map = {
+                "DDIM": DDIMScheduler,
+                "EulerAncestral": EulerAncestralDiscreteScheduler,
+                "PNDM": PNDMScheduler,
+            }
+            if scheduler_name in sched_map:
+                pipe.scheduler = sched_map[scheduler_name].from_config(pipe.scheduler.config)
+        except Exception:
+            # если что-то пошло не так — просто используем дефолтный scheduler
+            pass
 
     if randomize_seed:
         seed = random.randint(0, MAX_SEED)
 
-    # create generator on proper device
-    if device.startswith("cuda"):
-        generator = torch.Generator(device=device).manual_seed(seed)
-    else:
-        generator = torch.Generator().manual_seed(seed)
-
-    try:
-        out = pipe_state(
-            prompt=prompt,
-            negative_prompt=negative_prompt if negative_prompt else None,
-            guidance_scale=float(guidance_scale),
-            num_inference_steps=int(num_inference_steps),
-            width=int(width),
-            height=int(height),
-            generator=generator,
-        )
-        image = out.images[0]
-        return image, seed, "OK"
-    except Exception as e:
-        return None, seed, f"Inference error: {e}"
+    generator = torch.Generator().manual_seed(int(seed))
 
+    image = pipe(
+        prompt=prompt,
+        negative_prompt=negative_prompt,
+        guidance_scale=guidance_scale,
+        num_inference_steps=num_inference_steps,
+        width=width,
+        height=height,
+        generator=generator,
+    ).images[0]
 
-with gr.Blocks(css=css, title="Text-to-Image") as demo:
-    pipe_state = gr.State(value=default_pipe)
+    return image, seed
 
-    with gr.Column(elem_id="col-container"):
-        gr.Markdown("# Text-to-Image — demo")
 
-        with gr.Row():
-            model_selector = gr.Dropdown(
-                label="Model ID",
-                choices=MODEL_OPTIONS,
-                value=DEFAULT_MODEL,
-                interactive=True,
-            )
-            scheduler_selector = gr.Dropdown(
-                label="Scheduler",
-                choices=list(SCHEDULER_MAP.keys()),
-                value="default",
-                interactive=True,
-            )
 
-        status = gr.Markdown("Model status: ready" if default_pipe else "Model status: not loaded")
+examples = [
+    "Astronaut in a jungle, cold color palette, muted colors, detailed, 8k",
+    "An astronaut riding a green horse",
+    "A delicious ceviche cheesecake slice",
+]
+
+css = """
+#col-container {
+    margin: 0 auto;
+    max-width: 640px;
+}
+"""
+
+with gr.Blocks(css=css) as demo:
+    with gr.Column(elem_id="col-container"):
+        gr.Markdown(" # Text-to-Image Gradio Template")
+
+        # Model selector (выпадающий список)
+        model_select = gr.Dropdown(
+            label="Model",
+            choices=MODEL_OPTIONS,
+            value=DEFAULT_MODEL,
+            interactive=True,
+        )
+    
+        # опциональный селектор scheduler
+        scheduler_select = gr.Dropdown(
+            label="Scheduler (optional)",
+            choices=["", "DDIM", "EulerAncestral", "PNDM"],
+            value="",
+        )
 
         with gr.Row():
             prompt = gr.Text(
@@ -165,7 +135,8 @@ with gr.Blocks(css=css, title="Text-to-Image") as demo:
                 placeholder="Enter your prompt",
                 container=False,
             )
-            run_button = gr.Button("Run", variant="primary")
+
+            run_button = gr.Button("Run", scale=0, variant="primary")
 
         result = gr.Image(label="Result", show_label=False)
 
@@ -185,21 +156,15 @@ with gr.Blocks(css=css, title="Text-to-Image") as demo:
                 value=42,
             )
 
-            randomize_seed = gr.Checkbox(label="Randomize seed", value=False)
+            randomize_seed = gr.Checkbox(label="Randomize seed", value=True)
 
             with gr.Row():
-                size_preset = gr.Dropdown(
-                    label="Size preset",
-                    choices=["512x512", "768x512", "1024x1024"],
-                    value="512x512",
-                )
-
                 width = gr.Slider(
                     label="Width",
                     minimum=256,
                     maximum=MAX_IMAGE_SIZE,
                     step=32,
-                    value=512,
+                    value=1024,  # Replace with defaults that work for your model
                 )
 
                 height = gr.Slider(
@@ -207,47 +172,32 @@ with gr.Blocks(css=css, title="Text-to-Image") as demo:
                     minimum=256,
                     maximum=MAX_IMAGE_SIZE,
                     step=32,
-                    value=512,
+                    value=1024,  # Replace with defaults that work for your model
                 )
 
             with gr.Row():
                 guidance_scale = gr.Slider(
                     label="Guidance scale",
                     minimum=0.0,
-                    maximum=20.0,
+                    maximum=10.0,
                     step=0.1,
-                    value=7.0,
+                    value=7.0,  # Replace with defaults that work for your model
                 )
 
                 num_inference_steps = gr.Slider(
                     label="Number of inference steps",
                     minimum=1,
-                    maximum=150,
+                    maximum=50,
                     step=1,
-                    value=20,
+                    value=20,  # Replace with defaults that work for your model
                 )
 
         gr.Examples(examples=examples, inputs=[prompt])
-
-    # Events
-    model_selector.change(
-        fn=load_model_and_update,
-        inputs=[model_selector, scheduler_selector],
-        outputs=[pipe_state, status],
-        queue=True,
-    )
-    scheduler_selector.change(
-        fn=load_model_and_update,
-        inputs=[model_selector, scheduler_selector],
-        outputs=[pipe_state, status],
-        queue=True,
-    )
-
-    size_preset.change(fn=size_to_dims, inputs=size_preset, outputs=[width, height])
-
-    run_button.click(
+    gr.on(
+        triggers=[run_button.click, prompt.submit],
         fn=infer,
         inputs=[
+            model_select,
             prompt,
             negative_prompt,
             seed,
@@ -256,11 +206,10 @@ with gr.Blocks(css=css, title="Text-to-Image") as demo:
             height,
             guidance_scale,
             num_inference_steps,
-            pipe_state,
+            scheduler_select
         ],
-        outputs=[result, seed, status],
-        queue=True,
+        outputs=[result, seed],
     )
 
 if __name__ == "__main__":
-    demo.launch()
+    demo.launch()