Update app.py

app.py

@@ -1,100 +1,104 @@
 import gradio as gr
 import torch
-import time
 import random
+import time
 import re
 from diffusers import DiffusionPipeline, LCMScheduler
 
-# -------------------------------------------------
-# MODEL SETUP (CPU REALITY MODE)
-# -------------------------------------------------
-model_id = "runwayml/stable-diffusion-v1-5"
-adapter_id = "latent-consistency/lcm-lora-sdv1-5"
+# -------------------------------
+# MODEL SETUP (CPU SAFE)
+# -------------------------------
+MODEL_ID = "runwayml/stable-diffusion-v1-5"
+ADAPTER_ID = "latent-consistency/lcm-lora-sdv1-5"
 
 pipe = DiffusionPipeline.from_pretrained(
-    model_id,
+    MODEL_ID,
     torch_dtype=torch.float32,
     safety_checker=None
 )
 pipe.scheduler = LCMScheduler.from_config(pipe.scheduler.config)
-pipe.load_lora_weights(adapter_id)
+pipe.load_lora_weights(ADAPTER_ID)
 pipe.to("cpu")
 
 pipe.enable_attention_slicing()
 pipe.enable_vae_slicing()
 pipe.set_progress_bar_config(disable=True)
 
-# -------------------------------------------------
-# STRONG PROMPT UNDERSTANDING (ANIMAL-SAFE)
-# -------------------------------------------------
+# -------------------------------
+# PROMPT UNDERSTANDING ENGINE
+# -------------------------------
 def refine_prompt(user_prompt: str):
-    p = user_prompt.lower()
-
-    is_cute = any(w in p for w in ["cute", "adorable", "kawaii"])
-    is_realistic = any(w in p for w in ["realistic", "photo", "photograph"])
-
-    animal = re.search(
-        r"(snake|cat|dog|fox|rabbit|dragon|bird|frog|hamster)", p
-    )
-    subject = animal.group(1) if animal else user_prompt
-
-    # CORE STRUCTURE (this is crucial)
-    prompt = (
-        f"a single small {subject}, full body visible, "
-        f"centered composition, facing camera"
-    )
+    """
+    Deterministically converts user input to a structured prompt and negative prompt.
+    Ensures SD generates exactly the object the user wants.
+    """
+    # Lowercase and strip
+    p = user_prompt.lower().strip()
+
+    # Attempt to extract a single object from known list
+    known_objects = ["apple","banana","snake","cat","dog","fox","rabbit","dragon","bird","frog","hamster"]
+    obj_match = next((w for w in known_objects if w in p), None)
+    subject = obj_match if obj_match else p
+
+    # Style detection
+    is_cute = any(w in p for w in ["cute","adorable","kawaii"])
+    is_realistic = any(w in p for w in ["realistic","photo","photograph"])
+    is_cartoon = any(w in p for w in ["cartoon","anime","illustration"])
+
+    # Base prompt template
+    prompt = f"a single {subject}, centered, isolated"
 
     if is_cute:
-        prompt += (
-            ", cute, friendly, rounded body, "
-            "big expressive eyes, soft lighting, "
-            "smooth cartoon style, pastel colors"
-        )
+        prompt += ", cute, friendly, rounded body, big expressive eyes, soft lighting, smooth cartoon style, pastel colors"
+    elif is_cartoon:
+        prompt += ", cartoon style, clean lines, vibrant colors, simple background"
     elif is_realistic:
-        prompt += (
-            ", ultra realistic, natural anatomy, "
-            "professional wildlife photography"
-        )
+        prompt += ", ultra realistic, natural anatomy, professional photography"
     else:
-        prompt += ", clean illustration style, detailed"
-
-    prompt += ", simple plain background"
+        prompt += ", high quality, detailed, clean background"
 
+    # Strong negative prompt to prevent hallucinations
     negative = (
-        "multiple animals, duplicate, scary, horror, grotesque, "
-        "realistic snake scales, fangs, aggression, "
-        "blurry, low quality, cropped, out of frame"
+        "multiple objects, duplicate, blurry, low quality, cropped, out of frame, "
+        "horror, grotesque, aggressive, scary, weird colors, artifacts"
     )
 
     return prompt, negative
 
-# -------------------------------------------------
-# REALISTIC ETA (CPU HONEST)
-# -------------------------------------------------
+# -------------------------------
+# ETA CALCULATION
+# -------------------------------
 def estimate_time(steps, resolution):
-    # Measured HF Free CPU averages
-    base = 20  # model overhead
-    step_cost_512 = 22  # seconds per step @512
-
-    scale = (int(resolution) / 512) ** 2
-    est = base + (steps * step_cost_512 * scale)
-
-    return f"⏱️ Estimated time: ~{int(est)} seconds"
-
-# -------------------------------------------------
-# GENERATION WITH LIVE STATUS
-# -------------------------------------------------
+    # CPU empirical timing (seconds per step)
+    per_step = {
+        256: 6,
+        512: 12,
+        768: 25,
+        1024: 45
+    }[int(resolution)]
+
+    overhead = 10  # initial model load / conditioning
+    est = overhead + steps * per_step
+    minutes = est // 60
+    seconds = est % 60
+    return f"⏱️ Estimated time: ~{int(minutes)}m {int(seconds)}s"
+
+# -------------------------------
+# GENERATION FUNCTION (with live status)
+# -------------------------------
 def generate(prompt, resolution, steps):
-    start = time.time()
+    start_time = time.time()
     yield None, "🧠 Understanding your prompt..."
-
+    
     refined_prompt, neg_prompt = refine_prompt(prompt)
-
-    yield None, "🎨 Generating image (CPU, this takes time)..."
-
+    
+    yield None, "🎨 Generating image (CPU, please wait)..."
+    
     seed = random.randint(0, 10**9)
     gen = torch.Generator("cpu").manual_seed(seed)
 
+    pipe.scheduler.set_timesteps(int(steps))  # ensure LCM fast path
+    
     img = pipe(
         prompt=refined_prompt,
         negative_prompt=neg_prompt,
@@ -104,45 +108,36 @@ def generate(prompt, resolution, steps):
         height=int(resolution),
         generator=gen
     ).images[0]
-
-    duration = int(time.time() - start)
+    
+    duration = int(time.time() - start_time)
     yield [img], f"✅ Finished in {duration}s | Seed: {seed}"
 
-# -------------------------------------------------
-# UI
-# -------------------------------------------------
+# -------------------------------
+# GRADIO UI
+# -------------------------------
 with gr.Blocks(theme=gr.themes.Soft()) as demo:
-    gr.Markdown("# 👾 CREEPER AI — CPU HONEST MODE")
-
+    gr.Markdown("# 👾 CREEPER AI — CPU SMART IMAGE GENERATION")
+    
     with gr.Row():
         with gr.Column():
-            prompt_in = gr.Textbox(
-                label="Prompt",
-                placeholder="cute snake",
-                lines=2
-            )
-
-            resolution = gr.Radio(
-                [256, 512, 768, 1024],
-                value=512,
-                label="Resolution"
-            )
-
-            steps = gr.Slider(
-                2, 10, value=4, step=1, label="Steps"
-            )
-
-            eta = gr.Markdown("⏱️ Estimated time: ~90 seconds")
-            btn = gr.Button("Generate")
-
+            prompt_in = gr.Textbox(label="Prompt", placeholder="cute snake", lines=2)
+            
+            resolution = gr.Radio([256, 512, 768, 1024], value=512, label="Resolution")
+            
+            steps = gr.Slider(2, 8, value=4, step=1, label="Steps")
+            
+            eta = gr.Markdown("⏱️ Estimated time: ~1m 0s")
+            gen_btn = gr.Button("Generate")
+        
         with gr.Column():
             status = gr.Markdown("🟢 Ready")
             gallery = gr.Gallery(columns=1)
-
+    
+    # Update ETA dynamically
     for ctrl in [steps, resolution]:
         ctrl.change(estimate_time, [steps, resolution], eta)
-
-    btn.click(
+    
+    gen_btn.click(
         generate,
         inputs=[prompt_in, resolution, steps],
         outputs=[gallery, status]