{
 "cells": [
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# Unconditional image generation"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Unconditional image generation is a relatively straightforward task. The model only generates images - without any additional context like text or an image - resembling the training data it was trained on.\n",
    "\n",
    "The [DiffusionPipeline](https://huggingface.co/docs/diffusers/main/en/api/diffusion_pipeline#diffusers.DiffusionPipeline) is the easiest way to use a pre-trained diffusion system for inference.\n",
    "\n",
    "Start by creating an instance of [DiffusionPipeline](https://huggingface.co/docs/diffusers/main/en/api/diffusion_pipeline#diffusers.DiffusionPipeline) and specify which pipeline checkpoint you would like to download.\n",
    "You can use any of the 🧨 Diffusers [checkpoints](https://huggingface.co/models?library=diffusers&sort=downloads) from the Hub (the checkpoint you'll use generates images of butterflies).\n",
    "\n",
    "<Tip>\n",
    "\n",
    "💡 Want to train your own unconditional image generation model? Take a look at the training [guide](https://huggingface.co/docs/diffusers/main/en/using-diffusers/training/unconditional_training) to learn how to generate your own images.\n",
    "\n",
    "</Tip>\n",
    "\n",
    "In this guide, you'll use [DiffusionPipeline](https://huggingface.co/docs/diffusers/main/en/api/diffusion_pipeline#diffusers.DiffusionPipeline) for unconditional image generation with [DDPM](https://arxiv.org/abs/2006.11239):"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "from diffusers import DiffusionPipeline\n",
    "\n",
    "generator = DiffusionPipeline.from_pretrained(\"anton-l/ddpm-butterflies-128\")"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "The [DiffusionPipeline](https://huggingface.co/docs/diffusers/main/en/api/diffusion_pipeline#diffusers.DiffusionPipeline) downloads and caches all modeling, tokenization, and scheduling components. \n",
    "Because the model consists of roughly 1.4 billion parameters, we strongly recommend running it on a GPU.\n",
    "You can move the generator object to a GPU, just like you would in PyTorch:"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "generator.to(\"cuda\")"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Now you can use the `generator` to generate an image:"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "image = generator().images[0]"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "The output is by default wrapped into a [`PIL.Image`](https://pillow.readthedocs.io/en/stable/reference/Image.html?highlight=image#the-image-class) object.\n",
    "\n",
    "You can save the image by calling:"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "image.save(\"generated_image.png\")"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Try out the Spaces below, and feel free to play around with the inference steps parameter to see how it affects the image quality!\n",
    "\n",
    "<iframe\n",
    "\tsrc=\"https://stevhliu-ddpm-butterflies-128.hf.space\"\n",
    "\tframeborder=\"0\"\n",
    "\twidth=\"850\"\n",
    "\theight=\"500\"\n",
    "></iframe>"
   ]
  }
 ],
 "metadata": {},
 "nbformat": 4,
 "nbformat_minor": 4
}