initial commit with app files

app.py

@@ -0,0 +1,224 @@
+import gradio as gr
+import numpy as np
+import random
+import spaces
+import torch
+import os
+from diffusers import StableDiffusionXLControlNetPipeline, ControlNetModel, AutoencoderKL
+from diffusers.utils import load_image
+from peft import PeftModel
+from PIL import Image
+from huggingface_hub import hf_hub_download
+
+# Paths (update as needed)
+base_model = "stabilityai/stable-diffusion-xl-base-1.0"
+LORA_REPO_ID = "azad-uddin/blocky-character-uv"
+LORA_FILENAME = "blocky-character.safetensors" # Or whatever your LoRA file is named
+controlnet_model = "lllyasviel/sd-controlnet-canny"  # or your own
+INPUT_IMAGE_PATH = "uv_outline.png"
+prompt = "blockychar, futuristic knight, UV Texture"
+
+if torch.cuda.is_available():
+    dtype = torch.bfloat16
+else:
+    dtype = torch.float32
+device = "cuda" if torch.cuda.is_available() else "cpu"
+
+print(f"Loading model: {base_model} on device {device}...")
+# Load ControlNet and pipeline
+controlnet = ControlNetModel.from_pretrained(controlnet_model, torch_dtype=dtype)
+pipe = StableDiffusionXLControlNetPipeline.from_pretrained(
+    base_model,
+    controlnet=controlnet,
+    torch_dtype=dtype
+)
+lora_path = hf_hub_download(
+        repo_id=LORA_REPO_ID,
+        filename=LORA_FILENAME,
+        use_auth_token=os.getenv("hf_token") # Use HF_TOKEN for private LoRA repo
+    )
+pipe.load_lora_weights(os.path.dirname(lora_path), weight_name=LORA_FILENAME)
+pipe.to(device)
+
+input_image_path = hf_hub_download(
+        repo_id=LORA_REPO_ID,
+        filename=INPUT_IMAGE_PATH,
+        use_auth_token=os.getenv("hf_token")
+    )
+# Load UV outline image
+control_image = Image.open(input_image_path).convert("RGB")
+
+MAX_SEED = np.iinfo(np.int32).max
+MAX_IMAGE_SIZE = 1024 # Or 512 if using SD 1.5 base
+@spaces.GPU(duration=65) # Adjust duration as needed
+def generate_token(
+    prompt: str,
+    negative_prompt: str = "",
+    lora_scale: float = 1.0, # Control how much the LoRA influences the output
+    seed: int = 42,
+    randomize_seed: bool = False,
+    width: int = 1024,
+    height: int = 1024,
+    guidance_scale: float = 7.0,
+    num_inference_steps: int = 30,
+    progress=gr.Progress(track_tqdm=True),
+):
+    if randomize_seed:
+        seed = random.randint(0, MAX_SEED)
+
+    generator = torch.Generator(device=device).manual_seed(seed)
+
+    # Generate the image
+    image = pipe(
+        prompt=prompt,
+        negative_prompt=negative_prompt,
+        controlnet_conditioning_image=control_image.resize((1024, 1024)),
+        guidance_scale=guidance_scale,
+        num_inference_steps=num_inference_steps,
+        width=width,
+        height=height,
+        generator=generator,
+        # This is how you apply LoRA scale during inference without fusing/unfusing
+        cross_attention_kwargs={"scale": lora_scale},
+        # good_vae=vae,
+    ).images[0]
+
+    # Save as PNG
+    output_dir = "outputs"
+    os.makedirs(output_dir, exist_ok=True)
+    output_path = f"{output_dir}/generated_uvTexture_{seed}.png"
+    image.save(output_path, "PNG")
+    
+    return output_path, seed
+
+examples = [
+    "blockychar, Ironman with damaged suit, UV Texture",
+    "blockychar, Batman character with dark cape and cowl, UV Texture",
+    "blockychar, Donald Duck in dress of Trump, UV Texture",
+]
+
+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("# Custom Crypto Token Generator")
+        gr.Markdown(
+            "Generate unique crypto token images based on your text prompts. "
+            "This Space uses a fine-tuned LoRA model to understand 'token' concepts."
+            "\n\n**Instructions:** Describe your desired token, including its theme, materials, and style. "
+            "For best results, include your LoRA's trigger word `tokenart style`"
+            "and descriptive terms like `metallic`, `circular`, `glowing`."
+        )
+
+        with gr.Row():
+            prompt = gr.Text(
+                label="Prompt",
+                show_label=False,
+                max_lines=1,
+                placeholder="e.g., 'ironman tokenart, metallic, red and gold, arc reactor, futuristic, 3d render'",
+                container=False,
+            )
+            run_button = gr.Button("Generate Token", scale=0, variant="primary")
+
+        result_image = gr.Image(label="Generated Crypto Token", show_label=True, type="filepath")
+
+        with gr.Accordion("Advanced Settings", open=False):
+            negative_prompt = gr.Text(
+                label="Negative prompt",
+                max_lines=1,
+                placeholder="e.g., blurry, low quality, text, watermark",
+                value="blurry, low quality, text, watermark, deformed, bad anatomy",
+            )
+            lora_scale = gr.Slider(
+                label="LoRA Scale",
+                minimum=0.0,
+                maximum=1.0,
+                step=0.05,
+                value=0.8, # Recommended to start slightly below 1.0
+                info="Controls how much the LoRA influences the generation. 0.0 for no LoRA, 1.0 for full effect."
+            )
+            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():
+                width = gr.Slider(
+                    label="Width",
+                    minimum=512,
+                    maximum=MAX_IMAGE_SIZE,
+                    step=64,
+                    value=1024,
+                )
+                height = gr.Slider(
+                    label="Height",
+                    minimum=512,
+                    maximum=MAX_IMAGE_SIZE,
+                    step=64,
+                    value=1024,
+                )
+            with gr.Row():
+                guidance_scale = gr.Slider(
+                    label="Guidance scale (CFG)",
+                    minimum=1.0,
+                    maximum=15.0,
+                    step=0.5,
+                    value=7.0,
+                )
+                num_inference_steps = gr.Slider(
+                    label="Inference Steps",
+                    minimum=10,
+                    maximum=100,
+                    step=5,
+                    value=30,
+                )
+
+        gr.Examples(
+            examples=examples,
+            inputs=[prompt],
+            outputs=[result_image, seed],
+            fn=generate_token,
+            cache_examples=True,
+            cache_mode="lazy"
+        )
+    gr.on(
+        triggers=[run_button.click, prompt.submit],
+        fn=generate_token,
+        inputs=[
+            prompt,
+            negative_prompt,
+            lora_scale,
+            seed,
+            randomize_seed,
+            width,
+            height,
+            guidance_scale,
+            num_inference_steps,
+        ],
+        outputs=[result_image, seed],
+    )
+
+if __name__ == "__main__":
+    demo.launch()
+# Generate image
+# result = pipe(
+#     prompt=prompt,
+#     negative_prompt="",
+#     num_inference_steps=28,
+#     guidance_scale=4,
+#     controlnet_conditioning_image=control_image,
+#     height=1024,
+#     width=1024
+# ).images[0]
+
+# result.save("generated_uv_texture.png")
+# print("Saved: generated_uv_texture.png")

requirements.txt

@@ -0,0 +1,14 @@
+git+https://github.com/huggingface/diffusers.git
+git+https://github.com/huggingface/peft.git
+gradio==4.29.0 # Explicitly pin Gradio to a known stable version
+accelerate
+diffusers
+torch
+numpy
+transformers
+xformers
+sentencepiece
+scipy
+pydantic==2.10.6 # Pinning pydantic to avoid potential conflicts with Gradio
+Pillow
+spaces