First commit

.gitignore

@@ -0,0 +1,4 @@
+__pycache__
+.cog
+cache
+vae-cache

README.md

@@ -0,0 +1,23 @@
+# realistic-vision-v5-inpainting Cog model
+
+This is an implementation of inpainting using the model [SG161222/Realistic_Vision_V5.0_noVAE](https://huggingface.co/SG161222/Realistic_Vision_V5.0_noVAE) as a Cog model. [Cog packages machine learning models as standard containers.](https://github.com/replicate/cog)
+
+First, download the pre-trained weights:
+
+    cog run script/download-weights
+
+Then, you can run predictions:
+
+    cog predict -i image=@demo.png -i mask=@mask.png
+
+## Example:
+
+Input - "a tabby cat, high resolution, sitting on a park bench"
+
+![alt text](demo.png)
+
+![alt text](mask.png)
+
+Output:
+
+![alt text](output.png)

cog.yaml

@@ -0,0 +1,18 @@
+# Configuration for Cog ⚙️
+build:
+  gpu: true
+  python_version: "3.10"
+  system_packages:
+    - "libgl1-mesa-glx"
+    - "libglib2.0-0"
+  python_packages:
+    - "torch==2.0.1"
+    - "torchvision"
+    - "safetensors==0.3.1"
+    - "diffusers==0.19.0"
+    - "transformers==4.30.2"
+    - "accelerate==0.20.3"
+    - "omegaconf"
+
+# predict.py defines how predictions are run on your model
+predict: "predict.py:Predictor"

predict.py

@@ -0,0 +1,89 @@
+# Prediction interface for Cog ⚙️
+from cog import BasePredictor, Input, Path
+import os
+import math
+import torch
+from PIL import Image
+from diffusers import AutoencoderKL, StableDiffusionInpaintPipeline
+
+MODEL_NAME = "SG161222/Realistic_Vision_V5.0_noVAE"
+MODEL_CACHE = "cache"
+VAE_CACHE = "vae-cache"
+
+class Predictor(BasePredictor):
+    def setup(self):
+        """Load the model into memory to make running multiple predictions efficient"""
+        vae = AutoencoderKL.from_single_file(
+            "https://huggingface.co/stabilityai/sd-vae-ft-mse-original/blob/main/vae-ft-mse-840000-ema-pruned.safetensors",
+            cache_dir=VAE_CACHE
+        )
+        pipe = StableDiffusionInpaintPipeline.from_pretrained(
+            MODEL_CACHE,
+            vae=vae,
+        )
+        self.pipe = pipe.to("cuda")
+
+    def scale_down_image(self, image_path, max_size):
+        #Open the Image
+        image = Image.open(image_path)
+        #Get the Original width and height
+        width, height = image.size
+        # Calculate the scaling factor to fit the image within the max_size
+        scaling_factor = min(max_size/width, max_size/height)
+        # Calaculate the new width and height
+        new_width = int(width * scaling_factor)
+        new_height = int(height * scaling_factor)
+        #resize the image
+        resized_image = image.resize((new_width, new_height))
+        cropped_image = self.crop_center(resized_image)
+        return cropped_image
+
+    def crop_center(self, pil_img):
+        img_width, img_height = pil_img.size
+        crop_width = self.base(img_width)
+        crop_height = self.base(img_height)
+        return pil_img.crop(
+                (
+                    (img_width - crop_width) // 2,
+                    (img_height - crop_height) // 2,
+                    (img_width + crop_width) // 2,
+                    (img_height + crop_height) // 2)
+                )
+
+    def base(self, x):
+        return int(8 * math.floor(int(x)/8))
+
+    def predict(
+        self,
+        image: Path = Input(description="Input image"),
+        prompt: str = "a tabby cat, high resolution, sitting on a park bench",
+        mask: Path = Input(description="Mask image"),
+        negative_prompt: str = "(deformed iris, deformed pupils, semi-realistic, cgi, 3d, render, sketch, cartoon, drawing, anime:1.4), text, close up, cropped, out of frame, worst quality, low quality, jpeg artifacts, ugly, duplicate, morbid, mutilated, extra fingers, mutated hands, poorly drawn hands, poorly drawn face, mutation, deformed, blurry, dehydrated, bad anatomy, bad proportions, extra limbs, cloned face, disfigured, gross proportions, malformed limbs, missing arms, missing legs, extra arms, extra legs, fused fingers, too many fingers, long neck",
+        strength: float = Input(description="strength/weight", ge=0, le=1, default=0.8),
+        steps: int = Input(description=" num_inference_steps", ge=0, le=100, default=20),
+        seed: int = Input(description="Leave blank to randomize",  default=None),
+    ) -> Path:
+        """Run a single prediction on the model"""
+        if (seed == 0) or (seed == None):
+            seed = int.from_bytes(os.urandom(2), byteorder='big')
+        generator = torch.Generator('cuda').manual_seed(seed)
+        print("Using seed:", seed)
+
+        r_image = self.scale_down_image(image,1280)
+        r_mask = self.scale_down_image(mask, 1280)
+        width, height = r_image.size
+        image = self.pipe(
+            prompt=prompt,
+            image=r_image,
+            mask_image=r_mask,
+            strength=strength,
+            width=width,
+            height=height,
+            negative_prompt=negative_prompt,
+            num_inference_steps=steps,
+            generator=generator,
+        ).images[0]
+
+        out_path = Path(f"/tmp/output.png")
+        image.save(out_path)
+        return  out_path

script/download-weights

@@ -0,0 +1,30 @@
+#!/usr/bin/env python
+
+# Run this before you deploy it on replicate
+import os
+import sys
+import torch
+from diffusers import AutoencoderKL, StableDiffusionImg2ImgPipeline
+
+# append project directory to path so predict.py can be imported
+sys.path.append('.')
+from predict import MODEL_NAME, MODEL_CACHE, VAE_CACHE
+
+# Make cache folders
+if not os.path.exists(MODEL_CACHE):
+    os.makedirs(MODEL_CACHE)
+
+if not os.path.exists(VAE_CACHE):
+    os.makedirs(VAE_CACHE)
+
+url = "https://huggingface.co/stabilityai/sd-vae-ft-mse-original/blob/main/vae-ft-mse-840000-ema-pruned.safetensors"
+vae = AutoencoderKL.from_single_file(
+    url,
+    cache_dir=VAE_CACHE
+)
+
+pipe = StableDiffusionImg2ImgPipeline.from_pretrained(
+    MODEL_NAME,
+    torch_dtype=torch.float16,
+)
+pipe.save_pretrained(MODEL_CACHE, safe_serialization=True)