Deforum_Stable_Diffusion.ipynb

{
  "cells": [
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "c442uQJ_gUgy"
      },
      "source": [
        "# **Deforum Stable Diffusion v0.4**\n",
        "[Stable Diffusion](https://github.com/CompVis/stable-diffusion) by Robin Rombach, Andreas Blattmann, Dominik Lorenz, Patrick Esser, Bj\u00f6rn Ommer and the [Stability.ai](https://stability.ai/) Team. [K Diffusion](https://github.com/crowsonkb/k-diffusion) by [Katherine Crowson](https://twitter.com/RiversHaveWings). You need to get the ckpt file and put it on your Google Drive first to use this. It can be downloaded from [HuggingFace](https://huggingface.co/CompVis/stable-diffusion).\n",
        "\n",
        "Notebook by [deforum](https://discord.gg/upmXXsrwZc)"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "T4knibRpAQ06"
      },
      "source": [
        "# Setup"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "2g-f7cQmf2Nt",
        "cellView": "form"
      },
      "source": [
        "#@markdown **NVIDIA GPU**\n",
        "import subprocess\n",
        "sub_p_res = subprocess.run(['nvidia-smi', '--query-gpu=name,memory.total,memory.free', '--format=csv,noheader'], stdout=subprocess.PIPE).stdout.decode('utf-8')\n",
        "print(sub_p_res)"
      ],
      "outputs": [],
      "execution_count": null
    },
    {
      "cell_type": "code",
      "metadata": {
        "cellView": "form",
        "id": "TxIOPT0G5Lx1"
      },
      "source": [
        "#@markdown **Model and Output Paths**\n",
        "# ask for the link\n",
        "print(\"Local Path Variables:\\n\")\n",
        "\n",
        "models_path = \"/content/models\" #@param {type:\"string\"}\n",
        "output_path = \"/content/output\" #@param {type:\"string\"}\n",
        "\n",
        "#@markdown **Google Drive Path Variables (Optional)**\n",
        "mount_google_drive = True #@param {type:\"boolean\"}\n",
        "force_remount = False\n",
        "\n",
        "if mount_google_drive:\n",
        "    from google.colab import drive # type: ignore\n",
        "    try:\n",
        "        drive_path = \"/content/drive\"\n",
        "        drive.mount(drive_path,force_remount=force_remount)\n",
        "        models_path_gdrive = \"/content/drive/MyDrive/AI/models\" #@param {type:\"string\"}\n",
        "        output_path_gdrive = \"/content/drive/MyDrive/AI/StableDiffusion\" #@param {type:\"string\"}\n",
        "        models_path = models_path_gdrive\n",
        "        output_path = output_path_gdrive\n",
        "    except:\n",
        "        print(\"...error mounting drive or with drive path variables\")\n",
        "        print(\"...reverting to default path variables\")\n",
        "\n",
        "import os\n",
        "os.makedirs(models_path, exist_ok=True)\n",
        "os.makedirs(output_path, exist_ok=True)\n",
        "\n",
        "print(f\"models_path: {models_path}\")\n",
        "print(f\"output_path: {output_path}\")"
      ],
      "outputs": [],
      "execution_count": null
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "VRNl2mfepEIe",
        "cellView": "form"
      },
      "source": [
        "#@markdown **Setup Environment**\n",
        "\n",
        "setup_environment = True #@param {type:\"boolean\"}\n",
        "print_subprocess = False #@param {type:\"boolean\"}\n",
        "\n",
        "if setup_environment:\n",
        "    import subprocess, time\n",
        "    print(\"Setting up environment...\")\n",
        "    start_time = time.time()\n",
        "    all_process = [\n",
        "        ['pip', 'install', 'torch==1.12.1+cu113', 'torchvision==0.13.1+cu113', '--extra-index-url', 'https://download.pytorch.org/whl/cu113'],\n",
        "        ['pip', 'install', 'omegaconf==2.2.3', 'einops==0.4.1', 'pytorch-lightning==1.7.4', 'torchmetrics==0.9.3', 'torchtext==0.13.1', 'transformers==4.21.2', 'kornia==0.6.7'],\n",
        "        ['git', 'clone', 'https://github.com/deforum/stable-diffusion'],\n",
        "        ['pip', 'install', '-e', 'git+https://github.com/CompVis/taming-transformers.git@master#egg=taming-transformers'],\n",
        "        ['pip', 'install', '-e', 'git+https://github.com/openai/CLIP.git@main#egg=clip'],\n",
        "        ['pip', 'install', 'accelerate', 'ftfy', 'jsonmerge', 'matplotlib', 'resize-right', 'timm', 'torchdiffeq'],\n",
        "        ['git', 'clone', 'https://github.com/shariqfarooq123/AdaBins.git'],\n",
        "        ['git', 'clone', 'https://github.com/isl-org/MiDaS.git'],\n",
        "        ['git', 'clone', 'https://github.com/MSFTserver/pytorch3d-lite.git'],\n",
        "    ]\n",
        "    for process in all_process:\n",
        "        running = subprocess.run(process,stdout=subprocess.PIPE).stdout.decode('utf-8')\n",
        "        if print_subprocess:\n",
        "            print(running)\n",
        "    \n",
        "    print(subprocess.run(['git', 'clone', 'https://github.com/deforum/k-diffusion/'], stdout=subprocess.PIPE).stdout.decode('utf-8'))\n",
        "    with open('k-diffusion/k_diffusion/__init__.py', 'w') as f:\n",
        "        f.write('')\n",
        "\n",
        "    end_time = time.time()\n",
        "    print(f\"Environment set up in {end_time-start_time:.0f} seconds\")"
      ],
      "outputs": [],
      "execution_count": null
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "81qmVZbrm4uu",
        "cellView": "form"
      },
      "source": [
        "#@markdown **Python Definitions**\n",
        "import json\n",
        "from IPython import display\n",
        "\n",
        "import gc, math, os, pathlib, subprocess, sys, time\n",
        "import cv2\n",
        "import numpy as np\n",
        "import pandas as pd\n",
        "import random\n",
        "import requests\n",
        "import torch\n",
        "import torch.nn as nn\n",
        "import torchvision.transforms as T\n",
        "import torchvision.transforms.functional as TF\n",
        "from contextlib import contextmanager, nullcontext\n",
        "from einops import rearrange, repeat\n",
        "from omegaconf import OmegaConf\n",
        "from PIL import Image\n",
        "from pytorch_lightning import seed_everything\n",
        "from skimage.exposure import match_histograms\n",
        "from torchvision.utils import make_grid\n",
        "from tqdm import tqdm, trange\n",
        "from types import SimpleNamespace\n",
        "from torch import autocast\n",
        "\n",
        "sys.path.extend([\n",
        "    'src/taming-transformers',\n",
        "    'src/clip',\n",
        "    'stable-diffusion/',\n",
        "    'k-diffusion',\n",
        "    'pytorch3d-lite',\n",
        "    'AdaBins',\n",
        "    'MiDaS',\n",
        "])\n",
        "\n",
        "import py3d_tools as p3d\n",
        "\n",
        "from helpers import DepthModel, sampler_fn\n",
        "from k_diffusion.external import CompVisDenoiser\n",
        "from ldm.util import instantiate_from_config\n",
        "from ldm.models.diffusion.ddim import DDIMSampler\n",
        "from ldm.models.diffusion.plms import PLMSSampler\n",
        "\n",
        "def sanitize(prompt):\n",
        "    whitelist = set('abcdefghijklmnopqrstuvwxyz ABCDEFGHIJKLMNOPQRSTUVWXYZ')\n",
        "    tmp = ''.join(filter(whitelist.__contains__, prompt))\n",
        "    return tmp.replace(' ', '_')\n",
        "\n",
        "def anim_frame_warp_2d(prev_img_cv2, args, anim_args, keys, frame_idx):\n",
        "    angle = keys.angle_series[frame_idx]\n",
        "    zoom = keys.zoom_series[frame_idx]\n",
        "    translation_x = keys.translation_x_series[frame_idx]\n",
        "    translation_y = keys.translation_y_series[frame_idx]\n",
        "\n",
        "    center = (args.W // 2, args.H // 2)\n",
        "    trans_mat = np.float32([[1, 0, translation_x], [0, 1, translation_y]])\n",
        "    rot_mat = cv2.getRotationMatrix2D(center, angle, zoom)\n",
        "    trans_mat = np.vstack([trans_mat, [0,0,1]])\n",
        "    rot_mat = np.vstack([rot_mat, [0,0,1]])\n",
        "    xform = np.matmul(rot_mat, trans_mat)\n",
        "\n",
        "    return cv2.warpPerspective(\n",
        "        prev_img_cv2,\n",
        "        xform,\n",
        "        (prev_img_cv2.shape[1], prev_img_cv2.shape[0]),\n",
        "        borderMode=cv2.BORDER_WRAP if anim_args.border == 'wrap' else cv2.BORDER_REPLICATE\n",
        "    )\n",
        "\n",
        "def anim_frame_warp_3d(prev_img_cv2, depth, anim_args, keys, frame_idx):\n",
        "    TRANSLATION_SCALE = 1.0/200.0 # matches Disco\n",
        "    translate_xyz = [\n",
        "        -keys.translation_x_series[frame_idx] * TRANSLATION_SCALE, \n",
        "        keys.translation_y_series[frame_idx] * TRANSLATION_SCALE, \n",
        "        -keys.translation_z_series[frame_idx] * TRANSLATION_SCALE\n",
        "    ]\n",
        "    rotate_xyz = [\n",
        "        math.radians(keys.rotation_3d_x_series[frame_idx]), \n",
        "        math.radians(keys.rotation_3d_y_series[frame_idx]), \n",
        "        math.radians(keys.rotation_3d_z_series[frame_idx])\n",
        "    ]\n",
        "    rot_mat = p3d.euler_angles_to_matrix(torch.tensor(rotate_xyz, device=device), \"XYZ\").unsqueeze(0)\n",
        "    result = transform_image_3d(prev_img_cv2, depth, rot_mat, translate_xyz, anim_args)\n",
        "    torch.cuda.empty_cache()\n",
        "    return result\n",
        "\n",
        "def add_noise(sample: torch.Tensor, noise_amt: float) -> torch.Tensor:\n",
        "    return sample + torch.randn(sample.shape, device=sample.device) * noise_amt\n",
        "\n",
        "def get_output_folder(output_path, batch_folder):\n",
        "    out_path = os.path.join(output_path,time.strftime('%Y-%m'))\n",
        "    if batch_folder != \"\":\n",
        "        out_path = os.path.join(out_path, batch_folder)\n",
        "    os.makedirs(out_path, exist_ok=True)\n",
        "    return out_path\n",
        "\n",
        "def load_img(path, shape, use_alpha_as_mask=False):\n",
        "    # use_alpha_as_mask: Read the alpha channel of the image as the mask image\n",
        "    if path.startswith('http://') or path.startswith('https://'):\n",
        "        image = Image.open(requests.get(path, stream=True).raw)\n",
        "    else:\n",
        "        image = Image.open(path)\n",
        "\n",
        "    if use_alpha_as_mask:\n",
        "        image = image.convert('RGBA')\n",
        "    else:\n",
        "        image = image.convert('RGB')\n",
        "\n",
        "    image = image.resize(shape, resample=Image.LANCZOS)\n",
        "\n",
        "    mask_image = None\n",
        "    if use_alpha_as_mask:\n",
        "      # Split alpha channel into a mask_image\n",
        "      red, green, blue, alpha = Image.Image.split(image)\n",
        "      mask_image = alpha.convert('L')\n",
        "      image = image.convert('RGB')\n",
        "\n",
        "    image = np.array(image).astype(np.float16) / 255.0\n",
        "    image = image[None].transpose(0, 3, 1, 2)\n",
        "    image = torch.from_numpy(image)\n",
        "    image = 2.*image - 1.\n",
        "\n",
        "    return image, mask_image\n",
        "\n",
        "def load_mask_latent(mask_input, shape):\n",
        "    # mask_input (str or PIL Image.Image): Path to the mask image or a PIL Image object\n",
        "    # shape (list-like len(4)): shape of the image to match, usually latent_image.shape\n",
        "    \n",
        "    if isinstance(mask_input, str): # mask input is probably a file name\n",
        "        if mask_input.startswith('http://') or mask_input.startswith('https://'):\n",
        "            mask_image = Image.open(requests.get(mask_input, stream=True).raw).convert('RGBA')\n",
        "        else:\n",
        "            mask_image = Image.open(mask_input).convert('RGBA')\n",
        "    elif isinstance(mask_input, Image.Image):\n",
        "        mask_image = mask_input\n",
        "    else:\n",
        "        raise Exception(\"mask_input must be a PIL image or a file name\")\n",
        "\n",
        "    mask_w_h = (shape[-1], shape[-2])\n",
        "    mask = mask_image.resize(mask_w_h, resample=Image.LANCZOS)\n",
        "    mask = mask.convert(\"L\")\n",
        "    return mask\n",
        "\n",
        "def prepare_mask(mask_input, mask_shape, mask_brightness_adjust=1.0, mask_contrast_adjust=1.0):\n",
        "    # mask_input (str or PIL Image.Image): Path to the mask image or a PIL Image object\n",
        "    # shape (list-like len(4)): shape of the image to match, usually latent_image.shape\n",
        "    # mask_brightness_adjust (non-negative float): amount to adjust brightness of the iamge, \n",
        "    #     0 is black, 1 is no adjustment, >1 is brighter\n",
        "    # mask_contrast_adjust (non-negative float): amount to adjust contrast of the image, \n",
        "    #     0 is a flat grey image, 1 is no adjustment, >1 is more contrast\n",
        "    \n",
        "    mask = load_mask_latent(mask_input, mask_shape)\n",
        "\n",
        "    # Mask brightness/contrast adjustments\n",
        "    if mask_brightness_adjust != 1:\n",
        "        mask = TF.adjust_brightness(mask, mask_brightness_adjust)\n",
        "    if mask_contrast_adjust != 1:\n",
        "        mask = TF.adjust_contrast(mask, mask_contrast_adjust)\n",
        "\n",
        "    # Mask image to array\n",
        "    mask = np.array(mask).astype(np.float32) / 255.0\n",
        "    mask = np.tile(mask,(4,1,1))\n",
        "    mask = np.expand_dims(mask,axis=0)\n",
        "    mask = torch.from_numpy(mask)\n",
        "\n",
        "    if args.invert_mask:\n",
        "        mask = ( (mask - 0.5) * -1) + 0.5\n",
        "    \n",
        "    mask = np.clip(mask,0,1)\n",
        "    return mask\n",
        "\n",
        "def maintain_colors(prev_img, color_match_sample, mode):\n",
        "    if mode == 'Match Frame 0 RGB':\n",
        "        return match_histograms(prev_img, color_match_sample, multichannel=True)\n",
        "    elif mode == 'Match Frame 0 HSV':\n",
        "        prev_img_hsv = cv2.cvtColor(prev_img, cv2.COLOR_RGB2HSV)\n",
        "        color_match_hsv = cv2.cvtColor(color_match_sample, cv2.COLOR_RGB2HSV)\n",
        "        matched_hsv = match_histograms(prev_img_hsv, color_match_hsv, multichannel=True)\n",
        "        return cv2.cvtColor(matched_hsv, cv2.COLOR_HSV2RGB)\n",
        "    else: # Match Frame 0 LAB\n",
        "        prev_img_lab = cv2.cvtColor(prev_img, cv2.COLOR_RGB2LAB)\n",
        "        color_match_lab = cv2.cvtColor(color_match_sample, cv2.COLOR_RGB2LAB)\n",
        "        matched_lab = match_histograms(prev_img_lab, color_match_lab, multichannel=True)\n",
        "        return cv2.cvtColor(matched_lab, cv2.COLOR_LAB2RGB)\n",
        "\n",
        "\n",
        "def make_callback(sampler_name, dynamic_threshold=None, static_threshold=None, mask=None, init_latent=None, sigmas=None, sampler=None, masked_noise_modifier=1.0):  \n",
        "    # Creates the callback function to be passed into the samplers\n",
        "    # The callback function is applied to the image at each step\n",
        "    def dynamic_thresholding_(img, threshold):\n",
        "        # Dynamic thresholding from Imagen paper (May 2022)\n",
        "        s = np.percentile(np.abs(img.cpu()), threshold, axis=tuple(range(1,img.ndim)))\n",
        "        s = np.max(np.append(s,1.0))\n",
        "        torch.clamp_(img, -1*s, s)\n",
        "        torch.FloatTensor.div_(img, s)\n",
        "\n",
        "    # Callback for samplers in the k-diffusion repo, called thus:\n",
        "    #   callback({'x': x, 'i': i, 'sigma': sigmas[i], 'sigma_hat': sigmas[i], 'denoised': denoised})\n",
        "    def k_callback_(args_dict):\n",
        "        if dynamic_threshold is not None:\n",
        "            dynamic_thresholding_(args_dict['x'], dynamic_threshold)\n",
        "        if static_threshold is not None:\n",
        "            torch.clamp_(args_dict['x'], -1*static_threshold, static_threshold)\n",
        "        if mask is not None:\n",
        "            init_noise = init_latent + noise * args_dict['sigma']\n",
        "            is_masked = torch.logical_and(mask >= mask_schedule[args_dict['i']], mask != 0 )\n",
        "            new_img = init_noise * torch.where(is_masked,1,0) + args_dict['x'] * torch.where(is_masked,0,1)\n",
        "            args_dict['x'].copy_(new_img)\n",
        "\n",
        "    # Function that is called on the image (img) and step (i) at each step\n",
        "    def img_callback_(img, i):\n",
        "        # Thresholding functions\n",
        "        if dynamic_threshold is not None:\n",
        "            dynamic_thresholding_(img, dynamic_threshold)\n",
        "        if static_threshold is not None:\n",
        "            torch.clamp_(img, -1*static_threshold, static_threshold)\n",
        "        if mask is not None:\n",
        "            i_inv = len(sigmas) - i - 1\n",
        "            init_noise = sampler.stochastic_encode(init_latent, torch.tensor([i_inv]*batch_size).to(device), noise=noise)\n",
        "            is_masked = torch.logical_and(mask >= mask_schedule[i], mask != 0 )\n",
        "            new_img = init_noise * torch.where(is_masked,1,0) + img * torch.where(is_masked,0,1)\n",
        "            img.copy_(new_img)\n",
        "              \n",
        "    if init_latent is not None:\n",
        "        noise = torch.randn_like(init_latent, device=device) * masked_noise_modifier\n",
        "    if sigmas is not None and len(sigmas) > 0:\n",
        "        mask_schedule, _ = torch.sort(sigmas/torch.max(sigmas))\n",
        "    elif len(sigmas) == 0:\n",
        "        mask = None # no mask needed if no steps (usually happens because strength==1.0)\n",
        "    if sampler_name in [\"plms\",\"ddim\"]: \n",
        "        # Callback function formated for compvis latent diffusion samplers\n",
        "        if mask is not None:\n",
        "            assert sampler is not None, \"Callback function for stable-diffusion samplers requires sampler variable\"\n",
        "            batch_size = init_latent.shape[0]\n",
        "\n",
        "        callback = img_callback_\n",
        "    else: \n",
        "        # Default callback function uses k-diffusion sampler variables\n",
        "        callback = k_callback_\n",
        "\n",
        "    return callback\n",
        "\n",
        "def sample_from_cv2(sample: np.ndarray) -> torch.Tensor:\n",
        "    sample = ((sample.astype(float) / 255.0) * 2) - 1\n",
        "    sample = sample[None].transpose(0, 3, 1, 2).astype(np.float16)\n",
        "    sample = torch.from_numpy(sample)\n",
        "    return sample\n",
        "\n",
        "def sample_to_cv2(sample: torch.Tensor, type=np.uint8) -> np.ndarray:\n",
        "    sample_f32 = rearrange(sample.squeeze().cpu().numpy(), \"c h w -> h w c\").astype(np.float32)\n",
        "    sample_f32 = ((sample_f32 * 0.5) + 0.5).clip(0, 1)\n",
        "    sample_int8 = (sample_f32 * 255)\n",
        "    return sample_int8.astype(type)\n",
        "\n",
        "def transform_image_3d(prev_img_cv2, depth_tensor, rot_mat, translate, anim_args):\n",
        "    # adapted and optimized version of transform_image_3d from Disco Diffusion https://github.com/alembics/disco-diffusion \n",
        "    w, h = prev_img_cv2.shape[1], prev_img_cv2.shape[0]\n",
        "\n",
        "    aspect_ratio = float(w)/float(h)\n",
        "    near, far, fov_deg = anim_args.near_plane, anim_args.far_plane, anim_args.fov\n",
        "    persp_cam_old = p3d.FoVPerspectiveCameras(near, far, aspect_ratio, fov=fov_deg, degrees=True, device=device)\n",
        "    persp_cam_new = p3d.FoVPerspectiveCameras(near, far, aspect_ratio, fov=fov_deg, degrees=True, R=rot_mat, T=torch.tensor([translate]), device=device)\n",
        "\n",
        "    # range of [-1,1] is important to torch grid_sample's padding handling\n",
        "    y,x = torch.meshgrid(torch.linspace(-1.,1.,h,dtype=torch.float32,device=device),torch.linspace(-1.,1.,w,dtype=torch.float32,device=device))\n",
        "    z = torch.as_tensor(depth_tensor, dtype=torch.float32, device=device)\n",
        "    xyz_old_world = torch.stack((x.flatten(), y.flatten(), z.flatten()), dim=1)\n",
        "\n",
        "    xyz_old_cam_xy = persp_cam_old.get_full_projection_transform().transform_points(xyz_old_world)[:,0:2]\n",
        "    xyz_new_cam_xy = persp_cam_new.get_full_projection_transform().transform_points(xyz_old_world)[:,0:2]\n",
        "\n",
        "    offset_xy = xyz_new_cam_xy - xyz_old_cam_xy\n",
        "    # affine_grid theta param expects a batch of 2D mats. Each is 2x3 to do rotation+translation.\n",
        "    identity_2d_batch = torch.tensor([[1.,0.,0.],[0.,1.,0.]], device=device).unsqueeze(0)\n",
        "    # coords_2d will have shape (N,H,W,2).. which is also what grid_sample needs.\n",
        "    coords_2d = torch.nn.functional.affine_grid(identity_2d_batch, [1,1,h,w], align_corners=False)\n",
        "    offset_coords_2d = coords_2d - torch.reshape(offset_xy, (h,w,2)).unsqueeze(0)\n",
        "\n",
        "    image_tensor = rearrange(torch.from_numpy(prev_img_cv2.astype(np.float32)), 'h w c -> c h w').to(device)\n",
        "    new_image = torch.nn.functional.grid_sample(\n",
        "        image_tensor.add(1/512 - 0.0001).unsqueeze(0), \n",
        "        offset_coords_2d, \n",
        "        mode=anim_args.sampling_mode, \n",
        "        padding_mode=anim_args.padding_mode, \n",
        "        align_corners=False\n",
        "    )\n",
        "\n",
        "    # convert back to cv2 style numpy array\n",
        "    result = rearrange(\n",
        "        new_image.squeeze().clamp(0,255), \n",
        "        'c h w -> h w c'\n",
        "    ).cpu().numpy().astype(prev_img_cv2.dtype)\n",
        "    return result\n",
        "\n",
        "def generate(args, return_latent=False, return_sample=False, return_c=False):\n",
        "    seed_everything(args.seed)\n",
        "    os.makedirs(args.outdir, exist_ok=True)\n",
        "\n",
        "    sampler = PLMSSampler(model) if args.sampler == 'plms' else DDIMSampler(model)\n",
        "    model_wrap = CompVisDenoiser(model)\n",
        "    batch_size = args.n_samples\n",
        "    prompt = args.prompt\n",
        "    assert prompt is not None\n",
        "    data = [batch_size * [prompt]]\n",
        "    precision_scope = autocast if args.precision == \"autocast\" else nullcontext\n",
        "\n",
        "    init_latent = None\n",
        "    mask_image = None\n",
        "    init_image = None\n",
        "    if args.init_latent is not None:\n",
        "        init_latent = args.init_latent\n",
        "    elif args.init_sample is not None:\n",
        "        with precision_scope(\"cuda\"):\n",
        "            init_latent = model.get_first_stage_encoding(model.encode_first_stage(args.init_sample))\n",
        "    elif args.use_init and args.init_image != None and args.init_image != '':\n",
        "        init_image, mask_image = load_img(args.init_image, \n",
        "                                          shape=(args.W, args.H),  \n",
        "                                          use_alpha_as_mask=args.use_alpha_as_mask)\n",
        "        init_image = init_image.to(device)\n",
        "        init_image = repeat(init_image, '1 ... -> b ...', b=batch_size)\n",
        "        with precision_scope(\"cuda\"):\n",
        "            init_latent = model.get_first_stage_encoding(model.encode_first_stage(init_image))  # move to latent space        \n",
        "\n",
        "    if not args.use_init and args.strength > 0 and args.strength_0_no_init:\n",
        "        print(\"\\nNo init image, but strength > 0. Strength has been auto set to 0, since use_init is False.\")\n",
        "        print(\"If you want to force strength > 0 with no init, please set strength_0_no_init to False.\\n\")\n",
        "        args.strength = 0\n",
        "\n",
        "    # Mask functions\n",
        "    if args.use_mask:\n",
        "        assert args.mask_file is not None or mask_image is not None, \"use_mask==True: An mask image is required for a mask. Please enter a mask_file or use an init image with an alpha channel\"\n",
        "        assert args.use_init, \"use_mask==True: use_init is required for a mask\"\n",
        "        assert init_latent is not None, \"use_mask==True: An latent init image is required for a mask\"\n",
        "\n",
        "        mask = prepare_mask(args.mask_file if mask_image is None else mask_image, \n",
        "                            init_latent.shape, \n",
        "                            args.mask_contrast_adjust, \n",
        "                            args.mask_brightness_adjust)\n",
        "        \n",
        "        if (torch.all(mask == 0) or torch.all(mask == 1)) and args.use_alpha_as_mask:\n",
        "            raise Warning(\"use_alpha_as_mask==True: Using the alpha channel from the init image as a mask, but the alpha channel is blank.\")\n",
        "        \n",
        "        mask = mask.to(device)\n",
        "        mask = repeat(mask, '1 ... -> b ...', b=batch_size)\n",
        "    else:\n",
        "        mask = None\n",
        "        \n",
        "    t_enc = int((1.0-args.strength) * args.steps)\n",
        "\n",
        "    # Noise schedule for the k-diffusion samplers (used for masking)\n",
        "    k_sigmas = model_wrap.get_sigmas(args.steps)\n",
        "    k_sigmas = k_sigmas[len(k_sigmas)-t_enc-1:]\n",
        "\n",
        "    if args.sampler in ['plms','ddim']:\n",
        "        sampler.make_schedule(ddim_num_steps=args.steps, ddim_eta=args.ddim_eta, ddim_discretize='fill', verbose=False)\n",
        "\n",
        "    callback = make_callback(sampler_name=args.sampler,\n",
        "                            dynamic_threshold=args.dynamic_threshold, \n",
        "                            static_threshold=args.static_threshold,\n",
        "                            mask=mask, \n",
        "                            init_latent=init_latent,\n",
        "                            sigmas=k_sigmas,\n",
        "                            sampler=sampler)    \n",
        "\n",
        "    results = []\n",
        "    with torch.no_grad():\n",
        "        with precision_scope(\"cuda\"):\n",
        "            with model.ema_scope():\n",
        "                for prompts in data:\n",
        "                    uc = None\n",
        "                    if args.scale != 1.0:\n",
        "                        uc = model.get_learned_conditioning(batch_size * [\"\"])\n",
        "                    if isinstance(prompts, tuple):\n",
        "                        prompts = list(prompts)\n",
        "                    c = model.get_learned_conditioning(prompts)\n",
        "\n",
        "                    if args.init_c != None:\n",
        "                        c = args.init_c\n",
        "\n",
        "                    if args.sampler in [\"klms\",\"dpm2\",\"dpm2_ancestral\",\"heun\",\"euler\",\"euler_ancestral\"]:\n",
        "                        samples = sampler_fn(\n",
        "                            c=c, \n",
        "                            uc=uc, \n",
        "                            args=args, \n",
        "                            model_wrap=model_wrap, \n",
        "                            init_latent=init_latent, \n",
        "                            t_enc=t_enc, \n",
        "                            device=device, \n",
        "                            cb=callback)\n",
        "                    else:\n",
        "                        # args.sampler == 'plms' or args.sampler == 'ddim':\n",
        "                        if init_latent is not None and args.strength > 0:\n",
        "                            z_enc = sampler.stochastic_encode(init_latent, torch.tensor([t_enc]*batch_size).to(device))\n",
        "                        else:\n",
        "                            z_enc = torch.randn([args.n_samples, args.C, args.H // args.f, args.W // args.f], device=device)\n",
        "                        if args.sampler == 'ddim':\n",
        "                            samples = sampler.decode(z_enc, \n",
        "                                                     c, \n",
        "                                                     t_enc, \n",
        "                                                     unconditional_guidance_scale=args.scale,\n",
        "                                                     unconditional_conditioning=uc,\n",
        "                                                     img_callback=callback)\n",
        "                        elif args.sampler == 'plms': # no \"decode\" function in plms, so use \"sample\"\n",
        "                            shape = [args.C, args.H // args.f, args.W // args.f]\n",
        "                            samples, _ = sampler.sample(S=args.steps,\n",
        "                                                            conditioning=c,\n",
        "                                                            batch_size=args.n_samples,\n",
        "                                                            shape=shape,\n",
        "                                                            verbose=False,\n",
        "                                                            unconditional_guidance_scale=args.scale,\n",
        "                                                            unconditional_conditioning=uc,\n",
        "                                                            eta=args.ddim_eta,\n",
        "                                                            x_T=z_enc,\n",
        "                                                            img_callback=callback)\n",
        "                        else:\n",
        "                            raise Exception(f\"Sampler {args.sampler} not recognised.\")\n",
        "\n",
        "                    if return_latent:\n",
        "                        results.append(samples.clone())\n",
        "\n",
        "                    x_samples = model.decode_first_stage(samples)\n",
        "                    if return_sample:\n",
        "                        results.append(x_samples.clone())\n",
        "\n",
        "                    x_samples = torch.clamp((x_samples + 1.0) / 2.0, min=0.0, max=1.0)\n",
        "\n",
        "                    if return_c:\n",
        "                        results.append(c.clone())\n",
        "\n",
        "                    for x_sample in x_samples:\n",
        "                        x_sample = 255. * rearrange(x_sample.cpu().numpy(), 'c h w -> h w c')\n",
        "                        image = Image.fromarray(x_sample.astype(np.uint8))\n",
        "                        results.append(image)\n",
        "    return results"
      ],
      "outputs": [],
      "execution_count": null
    },
    {
      "cell_type": "code",
      "metadata": {
        "cellView": "form",
        "id": "CIUJ7lWI4v53"
      },
      "source": [
        "#@markdown **Select and Load Model**\n",
        "\n",
        "model_config = \"v1-inference.yaml\" #@param [\"custom\",\"v1-inference.yaml\"]\n",
        "model_checkpoint =  \"sd-v1-4.ckpt\" #@param [\"custom\",\"sd-v1-4-full-ema.ckpt\",\"sd-v1-4.ckpt\",\"sd-v1-3-full-ema.ckpt\",\"sd-v1-3.ckpt\",\"sd-v1-2-full-ema.ckpt\",\"sd-v1-2.ckpt\",\"sd-v1-1-full-ema.ckpt\",\"sd-v1-1.ckpt\"]\n",
        "custom_config_path = \"\" #@param {type:\"string\"}\n",
        "custom_checkpoint_path = \"\" #@param {type:\"string\"}\n",
        "\n",
        "load_on_run_all = True #@param {type: 'boolean'}\n",
        "half_precision = True # check\n",
        "check_sha256 = True #@param {type:\"boolean\"}\n",
        "\n",
        "model_map = {\n",
        "    \"sd-v1-4-full-ema.ckpt\": {'sha256': '14749efc0ae8ef0329391ad4436feb781b402f4fece4883c7ad8d10556d8a36a'},\n",
        "    \"sd-v1-4.ckpt\": {'sha256': 'fe4efff1e174c627256e44ec2991ba279b3816e364b49f9be2abc0b3ff3f8556'},\n",
        "    \"sd-v1-3-full-ema.ckpt\": {'sha256': '54632c6e8a36eecae65e36cb0595fab314e1a1545a65209f24fde221a8d4b2ca'},\n",
        "    \"sd-v1-3.ckpt\": {'sha256': '2cff93af4dcc07c3e03110205988ff98481e86539c51a8098d4f2236e41f7f2f'},\n",
        "    \"sd-v1-2-full-ema.ckpt\": {'sha256': 'bc5086a904d7b9d13d2a7bccf38f089824755be7261c7399d92e555e1e9ac69a'},\n",
        "    \"sd-v1-2.ckpt\": {'sha256': '3b87d30facd5bafca1cbed71cfb86648aad75d1c264663c0cc78c7aea8daec0d'},\n",
        "    \"sd-v1-1-full-ema.ckpt\": {'sha256': 'efdeb5dc418a025d9a8cc0a8617e106c69044bc2925abecc8a254b2910d69829'},\n",
        "    \"sd-v1-1.ckpt\": {'sha256': '86cd1d3ccb044d7ba8db743d717c9bac603c4043508ad2571383f954390f3cea'}\n",
        "}\n",
        "\n",
        "# config path\n",
        "ckpt_config_path = custom_config_path if model_config == \"custom\" else os.path.join(models_path, model_config)\n",
        "if os.path.exists(ckpt_config_path):\n",
        "    print(f\"{ckpt_config_path} exists\")\n",
        "else:\n",
        "    ckpt_config_path = \"./stable-diffusion/configs/stable-diffusion/v1-inference.yaml\"\n",
        "print(f\"Using config: {ckpt_config_path}\")\n",
        "\n",
        "# checkpoint path or download\n",
        "ckpt_path = custom_checkpoint_path if model_checkpoint == \"custom\" else os.path.join(models_path, model_checkpoint)\n",
        "ckpt_valid = True\n",
        "if os.path.exists(ckpt_path):\n",
        "    print(f\"{ckpt_path} exists\")\n",
        "else:\n",
        "    print(f\"Please download model checkpoint and place in {os.path.join(models_path, model_checkpoint)}\")\n",
        "    ckpt_valid = False\n",
        "\n",
        "if check_sha256 and model_checkpoint != \"custom\" and ckpt_valid:\n",
        "    import hashlib\n",
        "    print(\"\\n...checking sha256\")\n",
        "    with open(ckpt_path, \"rb\") as f:\n",
        "        bytes = f.read() \n",
        "        hash = hashlib.sha256(bytes).hexdigest()\n",
        "        del bytes\n",
        "    if model_map[model_checkpoint][\"sha256\"] == hash:\n",
        "        print(\"hash is correct\\n\")\n",
        "    else:\n",
        "        print(\"hash in not correct\\n\")\n",
        "        ckpt_valid = False\n",
        "\n",
        "if ckpt_valid:\n",
        "    print(f\"Using ckpt: {ckpt_path}\")\n",
        "\n",
        "def load_model_from_config(config, ckpt, verbose=False, device='cuda', half_precision=True):\n",
        "    map_location = \"cuda\" #@param [\"cpu\", \"cuda\"]\n",
        "    print(f\"Loading model from {ckpt}\")\n",
        "    pl_sd = torch.load(ckpt, map_location=map_location)\n",
        "    if \"global_step\" in pl_sd:\n",
        "        print(f\"Global Step: {pl_sd['global_step']}\")\n",
        "    sd = pl_sd[\"state_dict\"]\n",
        "    model = instantiate_from_config(config.model)\n",
        "    m, u = model.load_state_dict(sd, strict=False)\n",
        "    if len(m) > 0 and verbose:\n",
        "        print(\"missing keys:\")\n",
        "        print(m)\n",
        "    if len(u) > 0 and verbose:\n",
        "        print(\"unexpected keys:\")\n",
        "        print(u)\n",
        "\n",
        "    if half_precision:\n",
        "        model = model.half().to(device)\n",
        "    else:\n",
        "        model = model.to(device)\n",
        "    model.eval()\n",
        "    return model\n",
        "\n",
        "if load_on_run_all and ckpt_valid:\n",
        "    local_config = OmegaConf.load(f\"{ckpt_config_path}\")\n",
        "    model = load_model_from_config(local_config, f\"{ckpt_path}\", half_precision=half_precision)\n",
        "    device = torch.device(\"cuda\") if torch.cuda.is_available() else torch.device(\"cpu\")\n",
        "    model = model.to(device)"
      ],
      "outputs": [],
      "execution_count": null
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "ov3r4RD1tzsT"
      },
      "source": [
        "# Settings"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "0j7rgxvLvfay"
      },
      "source": [
        "### Animation Settings"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "cellView": "form",
        "id": "8HJN2TE3vh-J"
      },
      "source": [
        "\n",
        "def DeforumAnimArgs():\n",
        "\n",
        "    #@markdown ####**Animation:**\n",
        "    animation_mode = 'None' #@param ['None', '2D', '3D', 'Video Input', 'Interpolation'] {type:'string'}\n",
        "    max_frames = 1000 #@param {type:\"number\"}\n",
        "    border = 'wrap' #@param ['wrap', 'replicate'] {type:'string'}\n",
        "\n",
        "    #@markdown ####**Motion Parameters:**\n",
        "    angle = \"0:(0)\"#@param {type:\"string\"}\n",
        "    zoom = \"0:(1.04)\"#@param {type:\"string\"}\n",
        "    translation_x = \"0:(0)\"#@param {type:\"string\"}\n",
        "    translation_y = \"0:(0)\"#@param {type:\"string\"}\n",
        "    translation_z = \"0:(10)\"#@param {type:\"string\"}\n",
        "    rotation_3d_x = \"0:(0)\"#@param {type:\"string\"}\n",
        "    rotation_3d_y = \"0:(0)\"#@param {type:\"string\"}\n",
        "    rotation_3d_z = \"0:(0)\"#@param {type:\"string\"}\n",
        "    noise_schedule = \"0: (0.02)\"#@param {type:\"string\"}\n",
        "    strength_schedule = \"0: (0.65)\"#@param {type:\"string\"}\n",
        "    contrast_schedule = \"0: (1.0)\"#@param {type:\"string\"}\n",
        "\n",
        "    #@markdown ####**Coherence:**\n",
        "    color_coherence = 'Match Frame 0 LAB' #@param ['None', 'Match Frame 0 HSV', 'Match Frame 0 LAB', 'Match Frame 0 RGB'] {type:'string'}\n",
        "    diffusion_cadence = '1' #@param ['1','2','3','4','5','6','7','8'] {type:'string'}\n",
        "\n",
        "    #@markdown ####**3D Depth Warping:**\n",
        "    use_depth_warping = True #@param {type:\"boolean\"}\n",
        "    midas_weight = 0.3#@param {type:\"number\"}\n",
        "    near_plane = 200\n",
        "    far_plane = 10000\n",
        "    fov = 40#@param {type:\"number\"}\n",
        "    padding_mode = 'border'#@param ['border', 'reflection', 'zeros'] {type:'string'}\n",
        "    sampling_mode = 'bicubic'#@param ['bicubic', 'bilinear', 'nearest'] {type:'string'}\n",
        "    save_depth_maps = False #@param {type:\"boolean\"}\n",
        "\n",
        "    #@markdown ####**Video Input:**\n",
        "    video_init_path ='/content/video_in.mp4'#@param {type:\"string\"}\n",
        "    extract_nth_frame = 1#@param {type:\"number\"}\n",
        "\n",
        "    #@markdown ####**Interpolation:**\n",
        "    interpolate_key_frames = False #@param {type:\"boolean\"}\n",
        "    interpolate_x_frames = 4 #@param {type:\"number\"}\n",
        "    \n",
        "    #@markdown ####**Resume Animation:**\n",
        "    resume_from_timestring = False #@param {type:\"boolean\"}\n",
        "    resume_timestring = \"20220829210106\" #@param {type:\"string\"}\n",
        "\n",
        "    return locals()\n",
        "\n",
        "class DeformAnimKeys():\n",
        "    def __init__(self, anim_args):\n",
        "        self.angle_series = get_inbetweens(parse_key_frames(anim_args.angle), anim_args.max_frames)\n",
        "        self.zoom_series = get_inbetweens(parse_key_frames(anim_args.zoom), anim_args.max_frames)\n",
        "        self.translation_x_series = get_inbetweens(parse_key_frames(anim_args.translation_x), anim_args.max_frames)\n",
        "        self.translation_y_series = get_inbetweens(parse_key_frames(anim_args.translation_y), anim_args.max_frames)\n",
        "        self.translation_z_series = get_inbetweens(parse_key_frames(anim_args.translation_z), anim_args.max_frames)\n",
        "        self.rotation_3d_x_series = get_inbetweens(parse_key_frames(anim_args.rotation_3d_x), anim_args.max_frames)\n",
        "        self.rotation_3d_y_series = get_inbetweens(parse_key_frames(anim_args.rotation_3d_y), anim_args.max_frames)\n",
        "        self.rotation_3d_z_series = get_inbetweens(parse_key_frames(anim_args.rotation_3d_z), anim_args.max_frames)\n",
        "        self.noise_schedule_series = get_inbetweens(parse_key_frames(anim_args.noise_schedule), anim_args.max_frames)\n",
        "        self.strength_schedule_series = get_inbetweens(parse_key_frames(anim_args.strength_schedule), anim_args.max_frames)\n",
        "        self.contrast_schedule_series = get_inbetweens(parse_key_frames(anim_args.contrast_schedule), anim_args.max_frames)\n",
        "\n",
        "\n",
        "def get_inbetweens(key_frames, max_frames, integer=False, interp_method='Linear'):\n",
        "    key_frame_series = pd.Series([np.nan for a in range(max_frames)])\n",
        "\n",
        "    for i, value in key_frames.items():\n",
        "        key_frame_series[i] = value\n",
        "    key_frame_series = key_frame_series.astype(float)\n",
        "    \n",
        "    if interp_method == 'Cubic' and len(key_frames.items()) <= 3:\n",
        "      interp_method = 'Quadratic'    \n",
        "    if interp_method == 'Quadratic' and len(key_frames.items()) <= 2:\n",
        "      interp_method = 'Linear'\n",
        "          \n",
        "    key_frame_series[0] = key_frame_series[key_frame_series.first_valid_index()]\n",
        "    key_frame_series[max_frames-1] = key_frame_series[key_frame_series.last_valid_index()]\n",
        "    key_frame_series = key_frame_series.interpolate(method=interp_method.lower(), limit_direction='both')\n",
        "    if integer:\n",
        "        return key_frame_series.astype(int)\n",
        "    return key_frame_series\n",
        "\n",
        "def parse_key_frames(string, prompt_parser=None):\n",
        "    import re\n",
        "    pattern = r'((?P<frame>[0-9]+):[\\s]*[\\(](?P<param>[\\S\\s]*?)[\\)])'\n",
        "    frames = dict()\n",
        "    for match_object in re.finditer(pattern, string):\n",
        "        frame = int(match_object.groupdict()['frame'])\n",
        "        param = match_object.groupdict()['param']\n",
        "        if prompt_parser:\n",
        "            frames[frame] = prompt_parser(param)\n",
        "        else:\n",
        "            frames[frame] = param\n",
        "    if frames == {} and len(string) != 0:\n",
        "        raise RuntimeError('Key Frame string not correctly formatted')\n",
        "    return frames"
      ],
      "outputs": [],
      "execution_count": null
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "63UOJvU3xdPS"
      },
      "source": [
        "### Prompts\n",
        "`animation_mode: None` batches on list of *prompts*. `animation_mode: 2D` uses *animation_prompts* key frame sequence"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "2ujwkGZTcGev"
      },
      "source": [
        "\n",
        "prompts = [\n",
        "    \"a beautiful forest by Asher Brown Durand, trending on Artstation\", #the first prompt I want\n",
        "    \"a beautiful portrait of a woman by Artgerm, trending on Artstation\", #the second prompt I want\n",
        "    #\"the third prompt I don't want it I commented it with an\",\n",
        "]\n",
        "\n",
        "animation_prompts = {\n",
        "    0: \"a beautiful apple, trending on Artstation\",\n",
        "    20: \"a beautiful banana, trending on Artstation\",\n",
        "    30: \"a beautiful coconut, trending on Artstation\",\n",
        "    40: \"a beautiful durian, trending on Artstation\",\n",
        "}"
      ],
      "outputs": [],
      "execution_count": null
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "s8RAo2zI-vQm"
      },
      "source": [
        "# Run"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "qH74gBWDd2oq",
        "cellView": "form"
      },
      "source": [
        "def DeforumArgs():\n",
        "\n",
        "    #@markdown **Image Settings**\n",
        "    W = 512 #@param\n",
        "    H = 512 #@param\n",
        "    W, H = map(lambda x: x - x % 64, (W, H))  # resize to integer multiple of 64\n",
        "\n",
        "    #@markdown **Sampling Settings**\n",
        "    seed = -1 #@param\n",
        "    sampler = 'klms' #@param [\"klms\",\"dpm2\",\"dpm2_ancestral\",\"heun\",\"euler\",\"euler_ancestral\",\"plms\", \"ddim\"]\n",
        "    steps = 50 #@param\n",
        "    scale = 7 #@param\n",
        "    ddim_eta = 0.0 #@param\n",
        "    dynamic_threshold = None\n",
        "    static_threshold = None   \n",
        "\n",
        "    #@markdown **Save & Display Settings**\n",
        "    save_samples = True #@param {type:\"boolean\"}\n",
        "    save_settings = True #@param {type:\"boolean\"}\n",
        "    display_samples = True #@param {type:\"boolean\"}\n",
        "\n",
        "    #@markdown **Batch Settings**\n",
        "    n_batch = 1 #@param\n",
        "    batch_name = \"StableFun\" #@param {type:\"string\"}\n",
        "    filename_format = \"{timestring}_{index}_{prompt}.png\" #@param [\"{timestring}_{index}_{seed}.png\",\"{timestring}_{index}_{prompt}.png\"]\n",
        "    seed_behavior = \"iter\" #@param [\"iter\",\"fixed\",\"random\"]\n",
        "    make_grid = False #@param {type:\"boolean\"}\n",
        "    grid_rows = 2 #@param \n",
        "    outdir = get_output_folder(output_path, batch_name)\n",
        "\n",
        "    #@markdown **Init Settings**\n",
        "    use_init = False #@param {type:\"boolean\"}\n",
        "    strength = 0.0 #@param {type:\"number\"}\n",
        "    strength_0_no_init = True # Set the strength to 0 automatically when no init image is used\n",
        "    init_image = \"https://cdn.pixabay.com/photo/2022/07/30/13/10/green-longhorn-beetle-7353749_1280.jpg\" #@param {type:\"string\"}\n",
        "    # Whiter areas of the mask are areas that change more\n",
        "    use_mask = False #@param {type:\"boolean\"}\n",
        "    use_alpha_as_mask = False # use the alpha channel of the init image as the mask\n",
        "    mask_file = \"https://www.filterforge.com/wiki/images/archive/b/b7/20080927223728%21Polygonal_gradient_thumb.jpg\" #@param {type:\"string\"}\n",
        "    invert_mask = False #@param {type:\"boolean\"}\n",
        "    # Adjust mask image, 1.0 is no adjustment. Should be positive numbers.\n",
        "    mask_brightness_adjust = 1.0  #@param {type:\"number\"}\n",
        "    mask_contrast_adjust = 1.0  #@param {type:\"number\"}\n",
        "\n",
        "    n_samples = 1 # doesnt do anything\n",
        "    precision = 'autocast' \n",
        "    C = 4\n",
        "    f = 8\n",
        "\n",
        "    prompt = \"\"\n",
        "    timestring = \"\"\n",
        "    init_latent = None\n",
        "    init_sample = None\n",
        "    init_c = None\n",
        "\n",
        "    return locals()\n",
        "\n",
        "\n",
        "\n",
        "def next_seed(args):\n",
        "    if args.seed_behavior == 'iter':\n",
        "        args.seed += 1\n",
        "    elif args.seed_behavior == 'fixed':\n",
        "        pass # always keep seed the same\n",
        "    else:\n",
        "        args.seed = random.randint(0, 2**32 - 1)\n",
        "    return args.seed\n",
        "\n",
        "def render_image_batch(args):\n",
        "    args.prompts = {k: f\"{v:05d}\" for v, k in enumerate(prompts)}\n",
        "    \n",
        "    # create output folder for the batch\n",
        "    os.makedirs(args.outdir, exist_ok=True)\n",
        "    if args.save_settings or args.save_samples:\n",
        "        print(f\"Saving to {os.path.join(args.outdir, args.timestring)}_*\")\n",
        "\n",
        "    # save settings for the batch\n",
        "    if args.save_settings:\n",
        "        filename = os.path.join(args.outdir, f\"{args.timestring}_settings.txt\")\n",
        "        with open(filename, \"w+\", encoding=\"utf-8\") as f:\n",
        "            json.dump(dict(args.__dict__), f, ensure_ascii=False, indent=4)\n",
        "\n",
        "    index = 0\n",
        "    \n",
        "    # function for init image batching\n",
        "    init_array = []\n",
        "    if args.use_init:\n",
        "        if args.init_image == \"\":\n",
        "            raise FileNotFoundError(\"No path was given for init_image\")\n",
        "        if args.init_image.startswith('http://') or args.init_image.startswith('https://'):\n",
        "            init_array.append(args.init_image)\n",
        "        elif not os.path.isfile(args.init_image):\n",
        "            if args.init_image[-1] != \"/\": # avoids path error by adding / to end if not there\n",
        "                args.init_image += \"/\" \n",
        "            for image in sorted(os.listdir(args.init_image)): # iterates dir and appends images to init_array\n",
        "                if image.split(\".\")[-1] in (\"png\", \"jpg\", \"jpeg\"):\n",
        "                    init_array.append(args.init_image + image)\n",
        "        else:\n",
        "            init_array.append(args.init_image)\n",
        "    else:\n",
        "        init_array = [\"\"]\n",
        "\n",
        "    # when doing large batches don't flood browser with images\n",
        "    clear_between_batches = args.n_batch >= 32\n",
        "\n",
        "    for iprompt, prompt in enumerate(prompts):  \n",
        "        args.prompt = prompt\n",
        "        print(f\"Prompt {iprompt+1} of {len(prompts)}\")\n",
        "        print(f\"{args.prompt}\")\n",
        "\n",
        "        all_images = []\n",
        "\n",
        "        for batch_index in range(args.n_batch):\n",
        "            if clear_between_batches and batch_index % 32 == 0: \n",
        "                display.clear_output(wait=True)            \n",
        "            print(f\"Batch {batch_index+1} of {args.n_batch}\")\n",
        "            \n",
        "            for image in init_array: # iterates the init images\n",
        "                args.init_image = image\n",
        "                results = generate(args)\n",
        "                for image in results:\n",
        "                    if args.make_grid:\n",
        "                        all_images.append(T.functional.pil_to_tensor(image))\n",
        "                    if args.save_samples:\n",
        "                        if args.filename_format == \"{timestring}_{index}_{prompt}.png\":\n",
        "                            filename = f\"{args.timestring}_{index:05}_{sanitize(prompt)[:160]}.png\"\n",
        "                        else:\n",
        "                            filename = f\"{args.timestring}_{index:05}_{args.seed}.png\"\n",
        "                        image.save(os.path.join(args.outdir, filename))\n",
        "                    if args.display_samples:\n",
        "                        display.display(image)\n",
        "                    index += 1\n",
        "                args.seed = next_seed(args)\n",
        "\n",
        "        #print(len(all_images))\n",
        "        if args.make_grid:\n",
        "            grid = make_grid(all_images, nrow=int(len(all_images)/args.grid_rows))\n",
        "            grid = rearrange(grid, 'c h w -> h w c').cpu().numpy()\n",
        "            filename = f\"{args.timestring}_{iprompt:05d}_grid_{args.seed}.png\"\n",
        "            grid_image = Image.fromarray(grid.astype(np.uint8))\n",
        "            grid_image.save(os.path.join(args.outdir, filename))\n",
        "            display.clear_output(wait=True)            \n",
        "            display.display(grid_image)\n",
        "\n",
        "\n",
        "def render_animation(args, anim_args):\n",
        "    # animations use key framed prompts\n",
        "    args.prompts = animation_prompts\n",
        "\n",
        "    # expand key frame strings to values\n",
        "    keys = DeformAnimKeys(anim_args)\n",
        "\n",
        "    # resume animation\n",
        "    start_frame = 0\n",
        "    if anim_args.resume_from_timestring:\n",
        "        for tmp in os.listdir(args.outdir):\n",
        "            if tmp.split(\"_\")[0] == anim_args.resume_timestring:\n",
        "                start_frame += 1\n",
        "        start_frame = start_frame - 1\n",
        "\n",
        "    # create output folder for the batch\n",
        "    os.makedirs(args.outdir, exist_ok=True)\n",
        "    print(f\"Saving animation frames to {args.outdir}\")\n",
        "\n",
        "    # save settings for the batch\n",
        "    settings_filename = os.path.join(args.outdir, f\"{args.timestring}_settings.txt\")\n",
        "    with open(settings_filename, \"w+\", encoding=\"utf-8\") as f:\n",
        "        s = {**dict(args.__dict__), **dict(anim_args.__dict__)}\n",
        "        json.dump(s, f, ensure_ascii=False, indent=4)\n",
        "        \n",
        "    # resume from timestring\n",
        "    if anim_args.resume_from_timestring:\n",
        "        args.timestring = anim_args.resume_timestring\n",
        "\n",
        "    # expand prompts out to per-frame\n",
        "    prompt_series = pd.Series([np.nan for a in range(anim_args.max_frames)])\n",
        "    for i, prompt in animation_prompts.items():\n",
        "        prompt_series[i] = prompt\n",
        "    prompt_series = prompt_series.ffill().bfill()\n",
        "\n",
        "    # check for video inits\n",
        "    using_vid_init = anim_args.animation_mode == 'Video Input'\n",
        "\n",
        "    # load depth model for 3D\n",
        "    predict_depths = (anim_args.animation_mode == '3D' and anim_args.use_depth_warping) or anim_args.save_depth_maps\n",
        "    if predict_depths:\n",
        "        depth_model = DepthModel(device)\n",
        "        depth_model.load_midas(models_path)\n",
        "        if anim_args.midas_weight < 1.0:\n",
        "            depth_model.load_adabins()\n",
        "    else:\n",
        "        depth_model = None\n",
        "        anim_args.save_depth_maps = False\n",
        "\n",
        "    # state for interpolating between diffusion steps\n",
        "    turbo_steps = 1 if using_vid_init else int(anim_args.diffusion_cadence)\n",
        "    turbo_prev_image, turbo_prev_frame_idx = None, 0\n",
        "    turbo_next_image, turbo_next_frame_idx = None, 0\n",
        "\n",
        "    # resume animation\n",
        "    prev_sample = None\n",
        "    color_match_sample = None\n",
        "    if anim_args.resume_from_timestring:\n",
        "        last_frame = start_frame-1\n",
        "        if turbo_steps > 1:\n",
        "            last_frame -= last_frame%turbo_steps\n",
        "        path = os.path.join(args.outdir,f\"{args.timestring}_{last_frame:05}.png\")\n",
        "        img = cv2.imread(path)\n",
        "        img = cv2.cvtColor(img, cv2.COLOR_BGR2RGB)\n",
        "        prev_sample = sample_from_cv2(img)\n",
        "        if anim_args.color_coherence != 'None':\n",
        "            color_match_sample = img\n",
        "        if turbo_steps > 1:\n",
        "            turbo_next_image, turbo_next_frame_idx = sample_to_cv2(prev_sample, type=np.float32), last_frame\n",
        "            turbo_prev_image, turbo_prev_frame_idx = turbo_next_image, turbo_next_frame_idx\n",
        "            start_frame = last_frame+turbo_steps\n",
        "\n",
        "    args.n_samples = 1\n",
        "    frame_idx = start_frame\n",
        "    while frame_idx < anim_args.max_frames:\n",
        "        print(f\"Rendering animation frame {frame_idx} of {anim_args.max_frames}\")\n",
        "        noise = keys.noise_schedule_series[frame_idx]\n",
        "        strength = keys.strength_schedule_series[frame_idx]\n",
        "        contrast = keys.contrast_schedule_series[frame_idx]\n",
        "        depth = None\n",
        "        \n",
        "        # emit in-between frames\n",
        "        if turbo_steps > 1:\n",
        "            tween_frame_start_idx = max(0, frame_idx-turbo_steps)\n",
        "            for tween_frame_idx in range(tween_frame_start_idx, frame_idx):\n",
        "                tween = float(tween_frame_idx - tween_frame_start_idx + 1) / float(frame_idx - tween_frame_start_idx)\n",
        "                print(f\"  creating in between frame {tween_frame_idx} tween:{tween:0.2f}\")\n",
        "\n",
        "                advance_prev = turbo_prev_image is not None and tween_frame_idx > turbo_prev_frame_idx\n",
        "                advance_next = tween_frame_idx > turbo_next_frame_idx\n",
        "\n",
        "                if depth_model is not None:\n",
        "                    assert(turbo_next_image is not None)\n",
        "                    depth = depth_model.predict(turbo_next_image, anim_args)\n",
        "\n",
        "                if anim_args.animation_mode == '2D':\n",
        "                    if advance_prev:\n",
        "                        turbo_prev_image = anim_frame_warp_2d(turbo_prev_image, args, anim_args, keys, tween_frame_idx)\n",
        "                    if advance_next:\n",
        "                        turbo_next_image = anim_frame_warp_2d(turbo_next_image, args, anim_args, keys, tween_frame_idx)\n",
        "                else: # '3D'\n",
        "                    if advance_prev:\n",
        "                        turbo_prev_image = anim_frame_warp_3d(turbo_prev_image, depth, anim_args, keys, tween_frame_idx)\n",
        "                    if advance_next:\n",
        "                        turbo_next_image = anim_frame_warp_3d(turbo_next_image, depth, anim_args, keys, tween_frame_idx)\n",
        "                turbo_prev_frame_idx = turbo_next_frame_idx = tween_frame_idx\n",
        "\n",
        "                if turbo_prev_image is not None and tween < 1.0:\n",
        "                    img = turbo_prev_image*(1.0-tween) + turbo_next_image*tween\n",
        "                else:\n",
        "                    img = turbo_next_image\n",
        "\n",
        "                filename = f\"{args.timestring}_{tween_frame_idx:05}.png\"\n",
        "                cv2.imwrite(os.path.join(args.outdir, filename), cv2.cvtColor(img.astype(np.uint8), cv2.COLOR_RGB2BGR))\n",
        "                if anim_args.save_depth_maps:\n",
        "                    depth_model.save(os.path.join(args.outdir, f\"{args.timestring}_depth_{tween_frame_idx:05}.png\"), depth)\n",
        "            if turbo_next_image is not None:\n",
        "                prev_sample = sample_from_cv2(turbo_next_image)\n",
        "\n",
        "        # apply transforms to previous frame\n",
        "        if prev_sample is not None:\n",
        "            if anim_args.animation_mode == '2D':\n",
        "                prev_img = anim_frame_warp_2d(sample_to_cv2(prev_sample), args, anim_args, keys, frame_idx)\n",
        "            else: # '3D'\n",
        "                prev_img_cv2 = sample_to_cv2(prev_sample)\n",
        "                depth = depth_model.predict(prev_img_cv2, anim_args) if depth_model else None\n",
        "                prev_img = anim_frame_warp_3d(prev_img_cv2, depth, anim_args, keys, frame_idx)\n",
        "\n",
        "            # apply color matching\n",
        "            if anim_args.color_coherence != 'None':\n",
        "                if color_match_sample is None:\n",
        "                    color_match_sample = prev_img.copy()\n",
        "                else:\n",
        "                    prev_img = maintain_colors(prev_img, color_match_sample, anim_args.color_coherence)\n",
        "\n",
        "            # apply scaling\n",
        "            contrast_sample = prev_img * contrast\n",
        "            # apply frame noising\n",
        "            noised_sample = add_noise(sample_from_cv2(contrast_sample), noise)\n",
        "\n",
        "            # use transformed previous frame as init for current\n",
        "            args.use_init = True\n",
        "            if half_precision:\n",
        "                args.init_sample = noised_sample.half().to(device)\n",
        "            else:\n",
        "                args.init_sample = noised_sample.to(device)\n",
        "            args.strength = max(0.0, min(1.0, strength))\n",
        "\n",
        "        # grab prompt for current frame\n",
        "        args.prompt = prompt_series[frame_idx]\n",
        "        print(f\"{args.prompt} {args.seed}\")\n",
        "\n",
        "        # grab init image for current frame\n",
        "        if using_vid_init:\n",
        "            init_frame = os.path.join(args.outdir, 'inputframes', f\"{frame_idx+1:04}.jpg\")            \n",
        "            print(f\"Using video init frame {init_frame}\")\n",
        "            args.init_image = init_frame\n",
        "\n",
        "        # sample the diffusion model\n",
        "        sample, image = generate(args, return_latent=False, return_sample=True)\n",
        "        if not using_vid_init:\n",
        "            prev_sample = sample\n",
        "\n",
        "        if turbo_steps > 1:\n",
        "            turbo_prev_image, turbo_prev_frame_idx = turbo_next_image, turbo_next_frame_idx\n",
        "            turbo_next_image, turbo_next_frame_idx = sample_to_cv2(sample, type=np.float32), frame_idx\n",
        "            frame_idx += turbo_steps\n",
        "        else:    \n",
        "            filename = f\"{args.timestring}_{frame_idx:05}.png\"\n",
        "            image.save(os.path.join(args.outdir, filename))\n",
        "            if anim_args.save_depth_maps:\n",
        "                if depth is None:\n",
        "                    depth = depth_model.predict(sample_to_cv2(sample), anim_args)\n",
        "                depth_model.save(os.path.join(args.outdir, f\"{args.timestring}_depth_{frame_idx:05}.png\"), depth)\n",
        "            frame_idx += 1\n",
        "\n",
        "        display.clear_output(wait=True)\n",
        "        display.display(image)\n",
        "\n",
        "        args.seed = next_seed(args)\n",
        "\n",
        "def render_input_video(args, anim_args):\n",
        "    # create a folder for the video input frames to live in\n",
        "    video_in_frame_path = os.path.join(args.outdir, 'inputframes') \n",
        "    os.makedirs(video_in_frame_path, exist_ok=True)\n",
        "    \n",
        "    # save the video frames from input video\n",
        "    print(f\"Exporting Video Frames (1 every {anim_args.extract_nth_frame}) frames to {video_in_frame_path}...\")\n",
        "    try:\n",
        "        for f in pathlib.Path(video_in_frame_path).glob('*.jpg'):\n",
        "            f.unlink()\n",
        "    except:\n",
        "        pass\n",
        "    vf = r'select=not(mod(n\\,'+str(anim_args.extract_nth_frame)+'))'\n",
        "    subprocess.run([\n",
        "        'ffmpeg', '-i', f'{anim_args.video_init_path}', \n",
        "        '-vf', f'{vf}', '-vsync', 'vfr', '-q:v', '2', \n",
        "        '-loglevel', 'error', '-stats',  \n",
        "        os.path.join(video_in_frame_path, '%04d.jpg')\n",
        "    ], stdout=subprocess.PIPE).stdout.decode('utf-8')\n",
        "\n",
        "    # determine max frames from length of input frames\n",
        "    anim_args.max_frames = len([f for f in pathlib.Path(video_in_frame_path).glob('*.jpg')])\n",
        "\n",
        "    args.use_init = True\n",
        "    print(f\"Loading {anim_args.max_frames} input frames from {video_in_frame_path} and saving video frames to {args.outdir}\")\n",
        "    render_animation(args, anim_args)\n",
        "\n",
        "def render_interpolation(args, anim_args):\n",
        "    # animations use key framed prompts\n",
        "    args.prompts = animation_prompts\n",
        "\n",
        "    # create output folder for the batch\n",
        "    os.makedirs(args.outdir, exist_ok=True)\n",
        "    print(f\"Saving animation frames to {args.outdir}\")\n",
        "\n",
        "    # save settings for the batch\n",
        "    settings_filename = os.path.join(args.outdir, f\"{args.timestring}_settings.txt\")\n",
        "    with open(settings_filename, \"w+\", encoding=\"utf-8\") as f:\n",
        "        s = {**dict(args.__dict__), **dict(anim_args.__dict__)}\n",
        "        json.dump(s, f, ensure_ascii=False, indent=4)\n",
        "    \n",
        "    # Interpolation Settings\n",
        "    args.n_samples = 1\n",
        "    args.seed_behavior = 'fixed' # force fix seed at the moment bc only 1 seed is available\n",
        "    prompts_c_s = [] # cache all the text embeddings\n",
        "\n",
        "    print(f\"Preparing for interpolation of the following...\")\n",
        "\n",
        "    for i, prompt in animation_prompts.items():\n",
        "      args.prompt = prompt\n",
        "\n",
        "      # sample the diffusion model\n",
        "      results = generate(args, return_c=True)\n",
        "      c, image = results[0], results[1]\n",
        "      prompts_c_s.append(c) \n",
        "      \n",
        "      # display.clear_output(wait=True)\n",
        "      display.display(image)\n",
        "      \n",
        "      args.seed = next_seed(args)\n",
        "\n",
        "    display.clear_output(wait=True)\n",
        "    print(f\"Interpolation start...\")\n",
        "\n",
        "    frame_idx = 0\n",
        "\n",
        "    if anim_args.interpolate_key_frames:\n",
        "      for i in range(len(prompts_c_s)-1):\n",
        "        dist_frames = list(animation_prompts.items())[i+1][0] - list(animation_prompts.items())[i][0]\n",
        "        if dist_frames <= 0:\n",
        "          print(\"key frames duplicated or reversed. interpolation skipped.\")\n",
        "          return\n",
        "        else:\n",
        "          for j in range(dist_frames):\n",
        "            # interpolate the text embedding\n",
        "            prompt1_c = prompts_c_s[i]\n",
        "            prompt2_c = prompts_c_s[i+1]  \n",
        "            args.init_c = prompt1_c.add(prompt2_c.sub(prompt1_c).mul(j * 1/dist_frames))\n",
        "\n",
        "            # sample the diffusion model\n",
        "            results = generate(args)\n",
        "            image = results[0]\n",
        "\n",
        "            filename = f\"{args.timestring}_{frame_idx:05}.png\"\n",
        "            image.save(os.path.join(args.outdir, filename))\n",
        "            frame_idx += 1\n",
        "\n",
        "            display.clear_output(wait=True)\n",
        "            display.display(image)\n",
        "\n",
        "            args.seed = next_seed(args)\n",
        "\n",
        "    else:\n",
        "      for i in range(len(prompts_c_s)-1):\n",
        "        for j in range(anim_args.interpolate_x_frames+1):\n",
        "          # interpolate the text embedding\n",
        "          prompt1_c = prompts_c_s[i]\n",
        "          prompt2_c = prompts_c_s[i+1]  \n",
        "          args.init_c = prompt1_c.add(prompt2_c.sub(prompt1_c).mul(j * 1/(anim_args.interpolate_x_frames+1)))\n",
        "\n",
        "          # sample the diffusion model\n",
        "          results = generate(args)\n",
        "          image = results[0]\n",
        "\n",
        "          filename = f\"{args.timestring}_{frame_idx:05}.png\"\n",
        "          image.save(os.path.join(args.outdir, filename))\n",
        "          frame_idx += 1\n",
        "\n",
        "          display.clear_output(wait=True)\n",
        "          display.display(image)\n",
        "\n",
        "          args.seed = next_seed(args)\n",
        "\n",
        "    # generate the last prompt\n",
        "    args.init_c = prompts_c_s[-1]\n",
        "    results = generate(args)\n",
        "    image = results[0]\n",
        "    filename = f\"{args.timestring}_{frame_idx:05}.png\"\n",
        "    image.save(os.path.join(args.outdir, filename))\n",
        "\n",
        "    display.clear_output(wait=True)\n",
        "    display.display(image)\n",
        "    args.seed = next_seed(args)\n",
        "\n",
        "    #clear init_c\n",
        "    args.init_c = None\n",
        "\n",
        "\n",
        "args = SimpleNamespace(**DeforumArgs())\n",
        "anim_args = SimpleNamespace(**DeforumAnimArgs())\n",
        "\n",
        "args.timestring = time.strftime('%Y%m%d%H%M%S')\n",
        "args.strength = max(0.0, min(1.0, args.strength))\n",
        "\n",
        "if args.seed == -1:\n",
        "    args.seed = random.randint(0, 2**32 - 1)\n",
        "if not args.use_init:\n",
        "    args.init_image = None\n",
        "if args.sampler == 'plms' and (args.use_init or anim_args.animation_mode != 'None'):\n",
        "    print(f\"Init images aren't supported with PLMS yet, switching to KLMS\")\n",
        "    args.sampler = 'klms'\n",
        "if args.sampler != 'ddim':\n",
        "    args.ddim_eta = 0\n",
        "\n",
        "if anim_args.animation_mode == 'None':\n",
        "    anim_args.max_frames = 1\n",
        "elif anim_args.animation_mode == 'Video Input':\n",
        "    args.use_init = True\n",
        "\n",
        "# clean up unused memory\n",
        "gc.collect()\n",
        "torch.cuda.empty_cache()\n",
        "\n",
        "# dispatch to appropriate renderer\n",
        "if anim_args.animation_mode == '2D' or anim_args.animation_mode == '3D':\n",
        "    render_animation(args, anim_args)\n",
        "elif anim_args.animation_mode == 'Video Input':\n",
        "    render_input_video(args, anim_args)\n",
        "elif anim_args.animation_mode == 'Interpolation':\n",
        "    render_interpolation(args, anim_args)\n",
        "else:\n",
        "    render_image_batch(args)    "
      ],
      "outputs": [],
      "execution_count": null
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "4zV0J_YbMCTx"
      },
      "source": [
        "# Create video from frames"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "cellView": "form",
        "id": "no2jP8HTMBM0"
      },
      "source": [
        "skip_video_for_run_all = True #@param {type: 'boolean'}\n",
        "fps = 12 #@param {type:\"number\"}\n",
        "#@markdown **Manual Settings**\n",
        "use_manual_settings = False #@param {type:\"boolean\"}\n",
        "image_path = \"/content/drive/MyDrive/AI/StableDiffusion/2022-09/20220903000939_%05d.png\" #@param {type:\"string\"}\n",
        "mp4_path = \"/content/drive/MyDrive/AI/StableDiffusion/2022-09/20220903000939.mp4\" #@param {type:\"string\"}\n",
        "\n",
        "\n",
        "if skip_video_for_run_all == True:\n",
        "    print('Skipping video creation, uncheck skip_video_for_run_all if you want to run it')\n",
        "else:\n",
        "    import os\n",
        "    import subprocess\n",
        "    from base64 import b64encode\n",
        "\n",
        "    print(f\"{image_path} -> {mp4_path}\")\n",
        "\n",
        "    if use_manual_settings:\n",
        "        max_frames = \"200\" #@param {type:\"string\"}\n",
        "    else:\n",
        "        image_path = os.path.join(args.outdir, f\"{args.timestring}_%05d.png\")\n",
        "        mp4_path = os.path.join(args.outdir, f\"{args.timestring}.mp4\")\n",
        "        max_frames = str(anim_args.max_frames)\n",
        "\n",
        "    # make video\n",
        "    cmd = [\n",
        "        'ffmpeg',\n",
        "        '-y',\n",
        "        '-vcodec', 'png',\n",
        "        '-r', str(fps),\n",
        "        '-start_number', str(0),\n",
        "        '-i', image_path,\n",
        "        '-frames:v', max_frames,\n",
        "        '-c:v', 'libx264',\n",
        "        '-vf',\n",
        "        f'fps={fps}',\n",
        "        '-pix_fmt', 'yuv420p',\n",
        "        '-crf', '17',\n",
        "        '-preset', 'veryfast',\n",
        "        mp4_path\n",
        "    ]\n",
        "    process = subprocess.Popen(cmd, stdout=subprocess.PIPE, stderr=subprocess.PIPE)\n",
        "    stdout, stderr = process.communicate()\n",
        "    if process.returncode != 0:\n",
        "        print(stderr)\n",
        "        raise RuntimeError(stderr)\n",
        "\n",
        "    mp4 = open(mp4_path,'rb').read()\n",
        "    data_url = \"data:video/mp4;base64,\" + b64encode(mp4).decode()\n",
        "    display.display( display.HTML(f'<video controls loop><source src=\"{data_url}\" type=\"video/mp4\"></video>') )"
      ],
      "outputs": [],
      "execution_count": null
    }
  ],
  "metadata": {
    "accelerator": "GPU",
    "colab": {
      "collapsed_sections": [],
      "provenance": [],
      "private_outputs": true
    },
    "gpuClass": "standard",
    "kernelspec": {
      "display_name": "Python 3",
      "name": "python3"
    },
    "language_info": {
      "name": "python"
    }
  },
  "nbformat": 4,
  "nbformat_minor": 4
}