{
  "title": "Finetune Stable Diffusion",
  "video_id": "RJbnIndOfVw",
  "published": "2024-04-18T19:16:20+0000",
  "complexity": "Complex",
  "old_url": "https://workflows-of-ai.com/workflow/finetune-stable-diffusion",
  "description": "Fine-tune Stable Diffusion from Drupal with Dreambooth on Runpod, host the model on Hugging Face and generate images with the AI Interpolator.",
  "subtitle": "Fine-tune Stable Diffusion on your own images from the Drupal UI, then generate new images with the model through the AI Interpolator.",
  "summary": "The Hugging Face module starts a Runpod GPU machine that fine-tunes Stable Diffusion with Dreambooth on images uploaded and cropped in Drupal, then pushes the model to Hugging Face. Marcus trains it on rendered images of rumoured AirPods with a display and compares the result with the base model. The model is then used through the AI Interpolator, first with the free inference API and then with a dedicated inference endpoint.",
  "categories": [
    "Drupal AI"
  ],
  "intro_html": "<p>This video will showcase how you can use Drupal as your end-to-end shop for taking a Stable Diffusion (text-to-image) model, fine-tuning it with a concept, like your own face for instance, and reusing that for generation. All done in Drupal. No Python or scripting is needed!</p><p>This opens up a whole new world of personalization for you and your company. These models can then be used either to generate via the inference endpoints on Hugging Face, or you can download them to other providers or run them completely locally.</p><p>The two services you need are Hugging Face (free) for hosting the model and inference, and Runpod to host GPU machines with enough juice to fine-tune.</p><p>The price for an SD 1.4 model is around $0.5 and for an SD 2 model $1.</p><p>If you sign up for Runpod and like my videos, please use the following link so I can get access to more credits to play around with even more.</p><p>Links:<br>Drupal Hugging Face module: <a href=\"https://www.drupal.org/project/huggingface\">https://www.drupal.org/project/huggingface</a><br>Test my generated AirPods model: <a href=\"https://huggingface.co/ivanboring/airpods\">https://huggingface.co/ivanboring/airpods</a><br>Runpod to credit me: <a href=\"https://github.com/ivanboring/drupal-dreambooth-finetune\">https://github.com/ivanboring/drupal-dreambooth-finetune</a><br>Docker Hub: <a href=\"https://hub.docker.com/r/ivanboring/drupal-dreambooth-finetune\">https://hub.docker.com/r/ivanboring/drupal-dreambooth-finetune</a><br>Repo for Docker: <a href=\"https://github.com/ivanboring/drupal-dreambooth-finetune\">https://github.com/ivanboring/drupal-dreambooth-finetune</a><br>Hugging Face: <a href=\"https://huggingface.co/\">https://huggingface.co/</a></p>",
  "intro_added": false,
  "chapters": [
    "0:00 Fine-tuning without a GPU",
    "2:18 How it works for you",
    "2:54 Runpod and the credit system",
    "8:21 Installing the Hugging Face module",
    "9:47 Hugging Face and Runpod tokens",
    "10:54 Creating the Dreambooth fine-tune",
    "13:27 Uploading and cropping images",
    "16:47 Testing the model on Hugging Face",
    "18:54 Generating images with AI Interpolator",
    "22:20 A dedicated inference endpoint"
  ],
  "corrections": [
    [
      "druple",
      "Drupal"
    ],
    [
      "Dro B",
      "Drupal"
    ],
    [
      "droup B",
      "Drupal"
    ],
    [
      "hugging phase",
      "Hugging Face"
    ],
    [
      "hugging face",
      "Hugging Face"
    ],
    [
      "huging face",
      "Hugging Face"
    ],
    [
      "runp pod",
      "Runpod"
    ],
    [
      "runp POD",
      "Runpod"
    ],
    [
      "Run pod",
      "Runpod"
    ],
    [
      "run pod",
      "Runpod"
    ],
    [
      "runpod",
      "Runpod"
    ],
    [
      "with rmod",
      "with Runpod"
    ],
    [
      "dream booth",
      "Dreambooth"
    ],
    [
      "dream Booth",
      "Dreambooth"
    ],
    [
      "stable infusion",
      "Stable Diffusion"
    ],
    [
      "stable diffusion",
      "Stable Diffusion"
    ],
    [
      "touch Mahal",
      "Taj Mahal"
    ],
    [
      "AI interpolator",
      "AI Interpolator"
    ],
    [
      "open AI",
      "OpenAI"
    ]
  ],
  "ai_note": "The video and the introduction are Marcus's own, first published on workflows-of-ai.com. AI wrote the subtitle, the summary and the video chapters from the transcript, and corrected misheard names in YouTube's automatic captions."
}