import gradio as gr import torch import random import time from datetime import datetime from diffusers import StableDiffusionPipeline, EulerAncestralDiscreteScheduler import json # --- Configuration & Model Loading --- # Replicating the logic of your ComfyUI JSON: # - CheckpointLoaderSimple -> runwayml/stable-diffusion-v1-5 # - KSampler -> EulerAncestralDiscreteScheduler # - EmptyLatentImage -> 512x512 MODEL_ID = "runwayml/stable-diffusion-v1-5" DEVICE = "cuda" if torch.cuda.is_available() else "cpu" # Load the pipeline (simulating the backend compute) print(f"Loading model {MODEL_ID} on {DEVICE}...") try: pipe = StableDiffusionPipeline.from_pretrained( MODEL_ID, torch_dtype=torch.float16 if DEVICE == "cuda" else torch.float32, safety_checker=None # Disable for speed/demo purposes ) pipe.scheduler = EulerAncestralDiscreteScheduler.from_config(pipe.scheduler.config) if DEVICE == "cuda": pipe.to(DEVICE) print("Model loaded successfully.") except Exception as e: print(f"Error loading model: {e}") pipe = None # --- Core Logic Functions --- def generate_image_with_memory( prompt, negative_prompt, steps, cfg, seed, memory_db ): """ Generates an image and updates the 'Long Term Memory' database. """ if pipe is None: return None, memory_db, "Error: Model not loaded. Check GPU availability." start_time = time.time() # Handle Seed (0 = random) if seed == 0: seed = random.randint(0, 2147483647) generator = torch.Generator(device=DEVICE).manual_seed(seed) # Generate Image (Replicating KSampler + VAE Decode) try: result = pipe( prompt=prompt, negative_prompt=negative_prompt, num_inference_steps=int(steps), guidance_scale=float(cfg), height=512, width=512, generator=generator ) image = result.images[0] except Exception as e: return None, memory_db, f"Generation Failed: {str(e)}" # Update Memory Database timestamp = datetime.now().strftime("%Y-%m-%d %H:%M:%S") new_entry = { "timestamp": timestamp, "prompt": prompt, "negative": negative_prompt, "seed": seed, "steps": steps, "cfg": cfg } # Prepend new entry to memory updated_memory = [new_entry] + memory_db processing_time = f"{time.time() - start_time:.2f}s" return image, updated_memory, f"Generated in {processing_time} | Seed: {seed}" def recall_prompt(memory_db, evt: gr.SelectData): """ Recalls a specific prompt from history based on user selection. """ if memory_db is None or len(memory_db) == 0: return "", "", 0, 20, 7.0 index = evt.index[0] selected = memory_db[index] return ( selected["prompt"], selected["negative"], int(selected["seed"]), int(selected["steps"]), float(selected["cfg"]) ) def export_memory(memory_db): """ Exports the memory database to a downloadable JSON file. """ if not memory_db: return None json_str = json.dumps(memory_db, indent=4) return json_str # --- Gradio 6 UI Construction --- with gr.Blocks() as demo: # Header gr.HTML("""

🧠 AI-Native Memory Core

Automated Long Term Super Intelligent Human Recall | Persistent Database

Built with anycoder
""") with gr.Row(): # --- Sidebar: Controls & Memory Database --- with gr.Sidebar(width=350): gr.Markdown("### ⚙️ Generation Parameters") with gr.Accordion("Advanced Settings", open=True): steps = gr.Slider(minimum=1, maximum=150, value=20, step=1, label="Sampling Steps", info="Higher = more detail") cfg = gr.Slider(minimum=1, maximum=30, value=7.0, step=0.5, label="CFG Scale", info="Prompt adherence") seed = gr.Number(value=0, label="Seed (0 = Random)", precision=0) gr.Markdown("---") gr.Markdown("### 💾 Persistent Memory Database") gr.Markdown("Click a row to recall prompt settings.") memory_display = gr.Dataframe( label="Prompt History", headers=["Time", "Prompt", "Seed"], datatype=["str", "str", "number"], interactive=False, wrap=True, height=300 ) with gr.Row(): export_btn = gr.Button("💾 Export Memory (JSON)", size="sm") download_file = gr.File(visible=False, label="Download Database") gr.Markdown("---") gr.Markdown("### 📱 App Access") # Simulating the APK request via PWA/Install instructions gr.Markdown(""" To install on Android:
1. Open this in Chrome
2. Tap Menu (⋮)
3. Tap "Add to Home Screen" """) app_link_btn = gr.Button("📲 Open App View", link="/?view=standalone", variant="primary") # --- Main Area: Chat & Generation --- with gr.Column(scale=2): with gr.Group(): prompt_input = gr.Textbox( label="Positive Prompt", placeholder="Describe your vision...", lines=3, show_copy_button=True ) negative_input = gr.Textbox( label="Negative Prompt", value="low quality, blurry, distorted, bad anatomy", lines=2, visible=False # Collapsed for clean UI initially ) with gr.Row(): toggle_neg = gr.Button("Show/Hide Negative", size="sm") generate_btn = gr.Button("✨ Generate & Remember", variant="primary", scale=2) # Output Area with gr.Row(): with gr.Column(): result_image = gr.Image(label="Generated Artifact", type="pil", height=512) status_text = gr.Textbox(label="System Status", interactive=False) # --- State Management --- # We use Gradio State to hold the "Memory Database" in the session memory_state = gr.State(value=[]) # --- Event Listeners --- # 1. Toggle Negative Prompt toggle_neg.click(lambda: gr.Textbox(visible=True), None, negative_input) # 2. Generate Image & Update Memory generate_btn.click( fn=generate_image_with_memory, inputs=[prompt_input, negative_input, steps, cfg, seed, memory_state], outputs=[result_image, memory_state, status_text], api_visibility="public" ) # 3. Update Memory Table Display whenever state changes # We format the list of dicts into a dataframe-friendly format def format_memory_for_display(db): if not db: return [] return [[row["timestamp"], row["prompt"][:50] + "...", row["seed"]] for row in db] # We use a dummy trigger or just chain it. # In Gradio 6, we can use the `change` event on the state itself or chain outputs. # Here we chain the output of generate to update the view directly. # Note: To keep it simple, we modify the generate function to return the formatted dataframe if needed, # or we can use a separate update trigger. Let's do it inline for efficiency. # Redefining generate to return formatted table for immediate display def generate_with_ui_update(prompt, neg, steps, cfg, seed, db): img, new_db, status = generate_image_with_memory(prompt, neg, steps, cfg, seed, db) formatted_table = format_memory_for_display(new_db) return img, new_db, status, formatted_table generate_btn.click( fn=generate_with_ui_update, inputs=[prompt_input, negative_input, steps, cfg, seed, memory_state], outputs=[result_image, memory_state, status_text, memory_display], api_visibility="public" ) # 4. Recall from Memory (Click on Table) memory_display.select( fn=recall_prompt, inputs=[memory_state], outputs=[prompt_input, negative_input, seed, steps, cfg], api_visibility="private" ) # 5. Export Memory export_btn.click( fn=export_memory, inputs=[memory_state], outputs=[download_file], api_visibility="private" ) # --- Launch Configuration --- # Gradio 6 Syntax: Theme and parameters go in launch(), not Blocks() demo.launch( theme=gr.themes.Soft( primary_hue="indigo", secondary_hue="blue", neutral_hue="slate", font=gr.themes.GoogleFont("Inter"), text_size="lg", spacing_size="lg", radius_size="md" ).set( button_primary_background_fill="*primary_600", button_primary_background_fill_hover="*primary_700", block_title_text_weight="600", block_border_width="1px", block_border_color="*neutral_200" ), footer_links=[ {"label": "Built with anycoder", "url": "https://huggingface.co/spaces/akhaliq/anycoder"}, {"label": "Hugging Face", "url": "https://huggingface.co"} ], share=False # Set to True if you want a public link immediately )