stable-diffusion.ipynb

stable-diffusion.ipynb

{
  "nbformat": 4,
  "nbformat_minor": 0,
  "metadata": {
    "colab": {
      "name": "stable-diffusion.ipynb",
      "private_outputs": true,
      "provenance": [],
      "include_colab_link": true
    },
    "kernelspec": {
      "name": "python3",
      "display_name": "Python 3"
    },
    "language_info": {
      "name": "python"
    },
    "accelerator": "GPU",
    "gpuClass": "standard"
  },
  "cells": [
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "view-in-github",
        "colab_type": "text"
      },
      "source": [
        "<a href=\"https://colab.research.google.com/gist/tibaes/588ccdba75b29c86408b39e41aceb755/stable-diffusion.ipynb\" target=\"_parent\"><img src=\"https://colab.research.google.com/assets/colab-badge.svg\" alt=\"Open In Colab\"/></a>"
      ]
    },
    {
      "cell_type": "code",
      "execution_count": null,
      "metadata": {
        "id": "QuFz5uGi-h6G"
      },
      "outputs": [],
      "source": [
        "%pip install --quiet --upgrade diffusers transformers scipy mediapy accelerate ftfy spacy"
      ]
    },
    {
      "cell_type": "code",
      "source": [
        "import subprocess\n",
        "\n",
        "# The xformers package is mandatory to be able to create several 768x768 images.\n",
        "github_url = \"https://github.com/TheLastBen/fast-stable-diffusion\"\n",
        "xformers_wheels = \"xformers-0.0.13.dev0-py3-none-any.whl\"\n",
        "\n",
        "# Obtain GPU info\n",
        "\n",
        "nvidia_output = subprocess.run(['nvidia-smi', '-q'], capture_output=True).stdout\n",
        "\n",
        "gpu_info = [\n",
        "    str(line) for line in str(nvidia_output).split('\\\\n')\n",
        "    if \"Product Name\" in line\n",
        "    ]\n",
        "\n",
        "print(gpu_info)\n",
        "\n",
        "# Identify your GPU\n",
        "\n",
        "gpu_name = None\n",
        "\n",
        "for gpu_test in ['A100', 'K80', 'P100', 'T4', 'V100']:\n",
        "  if any(gpu_test in line for line in gpu_info):\n",
        "    gpu_name = gpu_test\n",
        "    break\n",
        "\n",
        "# Install xformers using pre-compiled Python wheels\n",
        "%pip install -q {github_url}/raw/main/precompiled/{gpu_name}/{xformers_wheels}"
      ],
      "metadata": {
        "id": "oP_dBQpSCIkY"
      },
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "source": [
        "# model_id = \"stabilityai/stable-diffusion-2-base\"\n",
        "model_id = \"stabilityai/stable-diffusion-2\""
      ],
      "metadata": {
        "id": "GR4vF2bw-sHR"
      },
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "source": [
        "from diffusers import PNDMScheduler, DDIMScheduler, LMSDiscreteScheduler, EulerDiscreteScheduler\n",
        "\n",
        "# scheduler = PNDMScheduler.from_pretrained(model_id, subfolder=\"scheduler\")\n",
        "# scheduler = DDIMScheduler.from_pretrained(model_id, subfolder=\"scheduler\")\n",
        "# scheduler = LMSDiscreteScheduler.from_pretrained(model_id, subfolder=\"scheduler\")\n",
        "scheduler = EulerDiscreteScheduler.from_pretrained(model_id, subfolder=\"scheduler\")"
      ],
      "metadata": {
        "id": "vF9Q0xKX8gLR"
      },
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "source": [
        "import mediapy as media\n",
        "import torch\n",
        "from diffusers import StableDiffusionPipeline\n",
        "\n",
        "device = \"cuda\"\n",
        "\n",
        "pipe = StableDiffusionPipeline.from_pretrained(\n",
        "    model_id,\n",
        "    scheduler=scheduler,\n",
        "    torch_dtype=torch.float16,\n",
        "    revision=\"fp16\",\n",
        "    )\n",
        "pipe = pipe.to(device)\n",
        "pipe.enable_xformers_memory_efficient_attention()\n",
        "\n",
        "if model_id.endswith('-base'):\n",
        "  image_length = 512\n",
        "else:\n",
        "  image_length = 768\n",
        "\n"
      ],
      "metadata": {
        "id": "bG2hkmSEvByV"
      },
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "source": [
        "prompt = \"A pikachu fine dining with a view to the Eiffel Tower\"\n",
        "num_images = 4\n",
        "\n",
        "images = pipe(\n",
        "    prompt,\n",
        "    num_images_per_prompt=num_images,\n",
        "    guidance_scale=9,\n",
        "    num_inference_steps=25,\n",
        "    height=image_length,\n",
        "    width=image_length,\n",
        "    ).images\n",
        "    \n",
        "media.show_images(images)\n",
        "images[0].save(\"output.jpg\")"
      ],
      "metadata": {
        "id": "AUc4QJfE-uR9"
      },
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "source": [],
      "metadata": {
        "id": "AxmBhekRzzUt"
      },
      "execution_count": null,
      "outputs": []
    }
  ]
}