This is how I made these freaky AI selfies, and now you can too

Oct 25, 2022

Udi Tirosh

Udi Tirosh is an entrepreneur, photography inventor, journalist, educator, and writer based in Israel. With over 25 years of experience in the photo-video industry, Udi has built and sold several photography-related brands. Udi has a double degree in mass media communications and computer science.

This is how I made these freaky AI selfies, and now you can too

Oct 25, 2022

Udi Tirosh

Udi Tirosh is an entrepreneur, photography inventor, journalist, educator, and writer based in Israel. With over 25 years of experience in the photo-video industry, Udi has built and sold several photography-related brands. Udi has a double degree in mass media communications and computer science.

Join the Discussion

Share on:

create your artificial intelligence image with stable diffusion - results 1

We have shared quite a few articles about AI and image generation recently. From art to deep fake to tool creation. Dall-E and Google’s Imagen have definitely been instrumental in bringing text-based image generation to the masses,  but have you considered some more fun applications? Like creating photos of yourself as a Viking, warrior, or James bond?

Asteria just released Dreambooth, a “Tailor-made AI image generation” tool. I played with it for a while, and let me tell you. It is a rabbit hole. Let me share a quick tutorial on how to use this tool to create the perfect collection of “you”.

If you create any good “prompts”, share them in the comments. Not sure what that means? You will at the end of the article.

What is DreamBooth?

create your artificial intelligence image with stable diffusion - results 2
These are  the results of several hours of playing with Dreambooth

Dreambooth is a model for generating images based on text prompts. The uniqueness of this model is its ability to place objects and humans into defined sets. If you want to read the full paper, click here. Be warned, though. It’s a heavy read if you are not in the image generation field.

How to use DreamBooth?

Using Dreambooth is quite simple. Just follow the steps below

Step 1 – Create a user

create-your-artificial-intelligence-image-account with stable diffusion

The first step would be to create a user. Head over to strmr.com and click login, this will start an account registration process, and it has the standard email and confirmation flow. Nothing special here.

Step 2 – Create a “Fine Tune”

Fine Tune is the name of your model. It helps to think about it as a digital, AI-generated model of the person you want to play with. There is a small fee for creating a model ($3 in October 2022), but the good news is that once you create this model (or fine-tune), you can create unlimited images with it.

To create a model, you need a few pictures of the subject. Astria recommends using between 10 and 20 images cropped to 1:1. They also suggest that at least the following photos will be present in your set:

  • Three pictures of full body or the entire object
  • Five medium-shot photos from the chest up
  • Ten close-ups

Here is the complete set of instructions for creating a good set:

  • Variation is key – Change body pose for every picture and show a variety of expressions and emotions. Make sure you capture the subject’s eyes looking in different directions for different images, take one with closed eyes. Every picture of your subject should introduce new information about your it.
  • Whatever you capture will be over-represented, so things you don’t want to get associated with your subject should change in every shot. Always pick a new background, even if that means just moving a little bit to shift the background.
  • Super important – Pick a good className. Something broad that you could associate your subject with. What works best? We don’t know! Please experiment and share! Some use “man” or “woman”, but we generally get better variations with “person”.
  • When constructing your prompts, always include “sks className” to represent the subject.

Here is part of the set that I used:

create your artificial intelligence image with stable diffusion - training set

You would need to name your set and give it a class like “man”, “woman”, “dog”, and so on.

Once the upload is complete, you will need to wait about 90 minutes, and TADA! You have created your first “fine tune”.

Step 3 – Create your AI prompts (a.k.a. images)

After about 90 minutes, the training is completed, and you will get a few default images. Here are a few of mine:

create your artificial intelligence image with stable diffusion - initial images

Now, it’s time to play. Use the textbox at the top of the site to enter different prompts and see what it generates. Just remember to enter sks man, sks woman, or sks <class> as part of your prompt. Don’t worry. The site will remind you if you forget with a red warning message.

It may take between one minute (during off times) and 20 minutes during peak time to generate a set of images. You will see the prompt status changing from enqueued to processing to a thumbnail which means that the image generation is done.

Where to find good descriptions

Once the tool is up and running, the challenge is finding good descriptions to generate images (called prompts). One place to find prompts is your imagination. You can think of a scenario and write in the line using sks <class> as the subject of the description. (I will share a few good prompts in a bit).

Your next step can be visiting the GitHub page for Dreambooth. It is filled with examples and clever keywords that can make an image look a certain way. Some words will make your model appear in a particular style, wear a certain piece of clothing, or place in a specific location.

create your artificial intelligence image with stable diffusion - lexica art
Lexica Art

You can also get inspiration from Lexica. Lexica is a website full of images and the prompts used to create them. You can use those directly or use them as a baseline for inspiration. You will soon find that  there are many keywords that you can use to create styles. Here are some ideas:

  • Style control: artstation, concept art, highly detailed, vintage, art nouveau, steampunk wizard, Alphonse Mucha, Louis Comfort Tiffany, by annie leibovitz, anime key visual, vector illustration, Oil painting
  • Photography direction: Rim Lighting, Cinematic Lighting, Rembrandt Lighting, glowing highlights, neon lights, photorealistic lighting
  • Artistic control: Retro comic style artwork,  comic book cover, symmetrical, vibrant, dynamic perspective, high angle
  • Facial control: Close-up, dirty face, beautiful, finely detailed eyes, moody.

Hugging Face

You can try Hugging Face if you want to test your prompts quickly. Hugging Face is not as detailed or high-resolution, but it’s a great place to practice variations until you get something you like.

Once you have a good prompt, you can go to DreamBooth and generate your image.

create your artificial intelligence image with stable diffusion - spaceman eating pizza

Some prompts to test

here are some “prompts” to try: (be warned, those are after my taste and may or may not be extremely worthy of a **Geek alert**.

  • Close-up of sks man in Hellblade: Senua’s Sacrifice, emerging from black mud, long hair with dreads, war blue paint, paint fading, angry expression, dirty face, finely detailed eyes, moody, viking clothes, epic scene, epic composition, Photography, Cinematic Lighting, Volumetric Lighting, ethereal light, intricate details, extremely detailed volumetric rays
  • symmetry ( badass ) cyberborg sks man ultra detailed, intricate, anime, dynamic lighting, digital art, digital painting, art station, wlop, sharp focus, illustration, art by artgerm and greg rutkowski and alphonse mucha
  • silhouette of a sks man illustration, vector art style, medium shot, intricate, elegant, highly detailed, digital art, ffffound, art by JC Leyendecker and sachin teng
  • a detailed animated portrait of sks man, wearing Starfleet uniform, by Alphonse Mucha and David Finch and Laurie Greasley, detailed, art nouveau

[BONUS] Generating multiple images while you are sleeping

If you are reading this, you have definitely fallen down the rabbit hole. At this point, there is a good chance that your contacts are either unfriending you because of all the weird images you send them, or asking you to create those images for them. You are probably frustrated with the amount of time it takes Dreambooth to generate your ideas.

There is a solution, though, and it requires some code writing, but fret not. I will make it super easy for you. This is a quick windows tutorial, but if you are using Linux, you probably don’t need this tutorial anyways :)

  1. Make sure you have curl installed on your computer. If you are running a recent windows edition, you’re set. If not, look for an online tutorial.
  2. Copy the text below into an empty file and call it “dreambooth.bat” and save it in an empty directory
    set TUNEID=<put your tine id here>
    set APIKEY=<put your tine id here>
    set FILENAME=prompts.txt
    
    for /f "tokens=*" %%a in (%FILENAME%) do call :process %%a
    
    goto thenextstep
    :process
    set VAR1=%*
    curl -X POST -H "Authorization: Bearer %APIKEY%" https://www.astria.ai/tunes/%TUNEID%/prompts -F prompt[text]="%VAR1%" 
    goto :EOF
  3. Copy the content below to another empty file and call it “prompts.txt” and save it to the same directory
    Highly detailed portrait of a hopeful pretty sks man painter, by farel dalrymple, 4k resolution, nier:automata inspired, bravely default inspired, vibrant but dreary but upflifting red, black and white color scheme!!! ((space nebula background))A picture of a sks man being very overdramatic, painting by artgerm and greg rutkowski and alphonse mucha
    extreme close up Illustration of sks man standing on a boat on a lake, book art, childrens book
    Little victorian sks man, black ink on paper, trending on artstation, beautiful, intricate, detailed
    Concept art character, very high angle view, book cover, sks man, walking in cyberpunk valley highly realistic, fine details, by borderland 3 sketch lineart for character design
    The portrait of an absurdly beautiful, graceful, elegant, sophisticated, fashionable sks man made of strawberries and white petals, an ultrafine hyperdetailed illustration by kim jung gi, irakli nadar, intricate linework, bright colors, octopath traveler, final fantasy, unreal engine 5 highly rendered, global illumination, radiant light, detailed and intricate environment 
    sks man, highly # detailed portrait of bulma, photographic realistic background, by jose torres, by royal jafarov, by dustin hobert, by joe fenton, by kaethe butcher, trending on instagram, award winning details 
    sks man as a pregnant man holding belly, intricate, elegant, highly detailed, digital painting, artstation, concept art, 8K, Refreshing, Rim Lighting, Cinematic Lighting, Rembrandt Lighting, highly detailed, vintage, art nouveau, steampunk wizard, Alphonse Mucha, Louis Comfort Tiffany, glowing highlights
    sks man in Retro comic style artwork, highly detailed "James Bond", comic book cover, symmetrical, vibrant
  4. edit dreambooth.bat and change the following:
    TUNEID to the ID of the model you want to play with (it is part of the URL when you click your model)
    APIKEY to an API key you can get here, in the API tab.
    FILENAME to prompts.txt
  5. now run the cmd command and cd to the directory where the files are saved, type dreambooth.bat and enjoy

If you prefer video form…

Here is a video tutorial showing how to create a Fine Tune and generate images:

https://www.youtube.com/watch?v=UN4r6BRDCYM

PS If you are concerned about using tools like this, you should check our article about AI and the photography industry.

Filed Under:

Tagged With:

Find this interesting? Share it with your friends!

Udi Tirosh

Udi Tirosh

Udi Tirosh is an entrepreneur, photography inventor, journalist, educator, and writer based in Israel. With over 25 years of experience in the photo-video industry, Udi has built and sold several photography-related brands. Udi has a double degree in mass media communications and computer science.

Join the Discussion

DIYP Comment Policy
Be nice, be on-topic, no personal information or flames.

Leave a Reply

Your email address will not be published. Required fields are marked *

8 responses to “This is how I made these freaky AI selfies, and now you can too”

  1. Tomer Daniel Avatar
    Tomer Daniel

    Amazing new world! Thank you for sharing :)

  2. Tomer Daniel Avatar
    Tomer Daniel

    Amazing new world! Thank you for sharing :)

  3. Joost Avatar
    Joost

    you have no idea how many laughs I did at this Udi

    1. Kaouthia Avatar
      Kaouthia

      You should’ve seen the images he DIDN’T post ;)

      1. udi tirosh Avatar
        udi tirosh

        @disqus_3VWXoe5ucH:disqus – oh, you should feel lucky!

  4. Joost Avatar
    Joost

    you have no idea how many laughs I did at this Udi

    1. Kaouthia Avatar
      Kaouthia

      You should’ve seen the images he DIDN’T post ;)

      1. udi tirosh Avatar
        udi tirosh

        @disqus_3VWXoe5ucH:disqus – oh, you should feel lucky!