Update app.py

app.py

@@ -1,153 +1,169 @@
-import gradio as gr
-import numpy as np
+# app.py - MangaMorph (Gradio) CPU-friendly template
+import os
 import random
-
-# import spaces #[uncomment to use ZeroGPU]
-from diffusers import DiffusionPipeline
+import numpy as np
+from PIL import Image
+import gradio as gr
 import torch
+from diffusers import DiffusionPipeline
+from diffusers import EulerDiscreteScheduler  # scheduler choice
 
-device = "cuda" if torch.cuda.is_available() else "cpu"
-model_repo_id = "stabilityai/sdxl-turbo"  # Replace to the model you would like to use
+# ---- CONFIG ----
+# Change this model id if you prefer another (see note about license/access above)
+MODEL_ID = os.getenv("MODEL_ID", "hakurei/waifu-diffusion")  
 
-if torch.cuda.is_available():
-    torch_dtype = torch.float16
-else:
-    torch_dtype = torch.float32
+# If your model requires a token, set HUGGINGFACE_HUB_TOKEN in Space secrets
+HF_TOKEN = os.getenv("HUGGINGFACE_HUB_TOKEN", None)
 
-pipe = DiffusionPipeline.from_pretrained(model_repo_id, torch_dtype=torch_dtype)
-pipe = pipe.to(device)
+device = "cuda" if torch.cuda.is_available() else "cpu"
+torch_dtype = torch.float32 if device == "cpu" else torch.float16
 
+# Limits / defaults for CPU-friendly runs
+DEFAULT_WIDTH = 512
+DEFAULT_HEIGHT = 512
+DEFAULT_STEPS = 20
+DEFAULT_GUIDANCE = 7.5
 MAX_SEED = np.iinfo(np.int32).max
-MAX_IMAGE_SIZE = 1024
 
+# Load pipeline (wrapped in try/except so error messages are shown in app log)
+def load_pipeline():
+    try:
+        scheduler = EulerDiscreteScheduler.from_pretrained(MODEL_ID, subfolder="scheduler") if os.getenv("USE_EULER", "1") == "1" else None
+        pipe = DiffusionPipeline.from_pretrained(
+            MODEL_ID,
+            torch_dtype=torch_dtype,
+            use_auth_token=HF_TOKEN,
+        )
+        # attach scheduler only if available and desired
+        if isinstance(pipe.scheduler, type(None)) and scheduler is not None:
+            pipe.scheduler = scheduler
+        pipe = pipe.to(device)
+        # For CPU: disable safety checker to avoid long CPU runs (optional)
+        try:
+            pipe.safety_checker = None
+        except Exception:
+            pass
+        return pipe
+    except Exception as e:
+        raise RuntimeError(f"Failed to load model '{MODEL_ID}': {e}")
+
+# lazy load
+PIPE = None
+def get_pipe():
+    global PIPE
+    if PIPE is None:
+        PIPE = load_pipeline()
+    return PIPE
+
+# Default negative prompt tuned to reduce common artifacts
+DEFAULT_NEGATIVE_PROMPT = (
+    "low quality, bad anatomy, blurry, deformed, extra limbs, mutated hands, "
+    "poorly drawn face, watermark, text, signature, lowres, oversaturated"
+)
 
-# @spaces.GPU #[uncomment to use ZeroGPU]
 def infer(
-    prompt,
-    negative_prompt,
-    seed,
-    randomize_seed,
-    width,
-    height,
-    guidance_scale,
-    num_inference_steps,
-    progress=gr.Progress(track_tqdm=True),
+    prompt: str,
+    negative_prompt: str,
+    seed: int,
+    randomize_seed: bool,
+    width: int,
+    height: int,
+    guidance_scale: float,
+    num_inference_steps: int,
 ):
+    if not prompt:
+        return None, "Please enter a prompt."
+
     if randomize_seed:
         seed = random.randint(0, MAX_SEED)
 
-    generator = torch.Generator().manual_seed(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]
-
-    return image, seed
-
+    gen = torch.Generator(device=device)
+    gen = gen.manual_seed(seed)
+
+    pipe = get_pipe()
+
+    # Cap size to avoid OOM on CPU
+    width = min(width, 768)
+    height = min(height, 768)
+
+    try:
+        output = pipe(
+            prompt=prompt,
+            negative_prompt=(negative_prompt or DEFAULT_NEGATIVE_PROMPT),
+            width=width,
+            height=height,
+            guidance_scale=float(guidance_scale),
+            num_inference_steps=int(num_inference_steps),
+            generator=gen,
+        )
+        image = output.images[0]
+        # simple postprocessing: convert to RGB and return
+        if isinstance(image, Image.Image):
+            image = image.convert("RGB")
+        return image, f"Seed: {seed}"
+    except Exception as e:
+        # retry logic: try again with smaller steps/guidance if CPU fails
+        try:
+            output = pipe(
+                prompt=prompt,
+                negative_prompt=(negative_prompt or DEFAULT_NEGATIVE_PROMPT),
+                width=width,
+                height=height,
+                guidance_scale=max(3.0, float(guidance_scale) - 1.0),
+                num_inference_steps=max(5, int(num_inference_steps) - 5),
+                generator=gen,
+            )
+            image = output.images[0]
+            if isinstance(image, Image.Image):
+                image = image.convert("RGB")
+            return image, f"Recovered (retry) — Seed: {seed}"
+        except Exception as e2:
+            return None, f"Generation failed: {e2}"
+
+# ---- UI ----
+css = """
+#main { max-width: 880px; margin: auto; }
+.header { text-align: center; }
+.small { font-size: 0.9rem; color: #666; }
+"""
 
 examples = [
-    "Astronaut in a jungle, cold color palette, muted colors, detailed, 8k",
-    "An astronaut riding a green horse",
-    "A delicious ceviche cheesecake slice",
+    "A young anime girl standing in a rain-soaked neon street, detailed lighting, cinematic",
+    "A samurai in traditional armor on a cliff at sunset, dramatic lighting, anime style",
+    "Cozy room with anime character by window reading, soft warm light"
 ]
 
-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")
+with gr.Blocks(css=css, theme=gr.themes.Default()) as demo:
+    with gr.Column(elem_id="main"):
+        gr.Markdown("<div class='header'><h2>MangaMorph — Anime Scene Generator</h2>"
+                    "<div class='small'>Text → Anime image | CPU-optimized | Use Model ID or set HF token in Secrets</div></div>")
 
         with gr.Row():
-            prompt = gr.Text(
-                label="Prompt",
-                show_label=False,
-                max_lines=1,
-                placeholder="Enter your prompt",
-                container=False,
-            )
-
-            run_button = gr.Button("Run", scale=0, variant="primary")
+            prompt = gr.Textbox(label="Prompt", placeholder="Describe the anime scene you want...", lines=2)
+            run_btn = gr.Button("Generate", variant="primary")
 
-        result = gr.Image(label="Result", show_label=False)
-
-        with gr.Accordion("Advanced Settings", open=False):
-            negative_prompt = gr.Text(
-                label="Negative prompt",
-                max_lines=1,
-                placeholder="Enter a negative prompt",
-                visible=False,
-            )
-
-            seed = gr.Slider(
-                label="Seed",
-                minimum=0,
-                maximum=MAX_SEED,
-                step=1,
-                value=0,
-            )
-
-            randomize_seed = gr.Checkbox(label="Randomize seed", value=True)
+        with gr.Row():
+            gallery = gr.Image(label="Result")
 
+        with gr.Accordion("Advanced settings", open=False):
+            negative = gr.Textbox(label="Negative prompt (optional)", placeholder=DEFAULT_NEGATIVE_PROMPT, lines=2, value=DEFAULT_NEGATIVE_PROMPT)
+            seed = gr.Number(label="Seed (0 = randomize)", value=0)
+            randomize = gr.Checkbox(label="Randomize seed", value=True)
             with gr.Row():
-                width = gr.Slider(
-                    label="Width",
-                    minimum=256,
-                    maximum=MAX_IMAGE_SIZE,
-                    step=32,
-                    value=1024,  # Replace with defaults that work for your model
-                )
-
-                height = gr.Slider(
-                    label="Height",
-                    minimum=256,
-                    maximum=MAX_IMAGE_SIZE,
-                    step=32,
-                    value=1024,  # Replace with defaults that work for your model
-                )
-
+                width = gr.Slider(label="Width", minimum=256, maximum=768, step=64, value=DEFAULT_WIDTH)
+                height = gr.Slider(label="Height", minimum=256, maximum=768, step=64, value=DEFAULT_HEIGHT)
             with gr.Row():
-                guidance_scale = gr.Slider(
-                    label="Guidance scale",
-                    minimum=0.0,
-                    maximum=10.0,
-                    step=0.1,
-                    value=0.0,  # Replace with defaults that work for your model
-                )
-
-                num_inference_steps = gr.Slider(
-                    label="Number of inference steps",
-                    minimum=1,
-                    maximum=50,
-                    step=1,
-                    value=2,  # Replace with defaults that work for your model
-                )
+                guidance = gr.Slider(label="Guidance scale", minimum=1.0, maximum=15.0, step=0.1, value=DEFAULT_GUIDANCE)
+                steps = gr.Slider(label="Steps", minimum=5, maximum=50, step=1, value=DEFAULT_STEPS)
 
         gr.Examples(examples=examples, inputs=[prompt])
-    gr.on(
-        triggers=[run_button.click, prompt.submit],
+
+        status = gr.Textbox(label="Status / Seed", interactive=False)
+
+    run_btn.click(
         fn=infer,
-        inputs=[
-            prompt,
-            negative_prompt,
-            seed,
-            randomize_seed,
-            width,
-            height,
-            guidance_scale,
-            num_inference_steps,
-        ],
-        outputs=[result, seed],
+        inputs=[prompt, negative, seed, randomize, width, height, guidance, steps],
+        outputs=[gallery, status],
     )
 
 if __name__ == "__main__":