Topic-driven, text-based image generation AI model.


Tutorials and examples in French.
Create your own images.


With Dreambooth specifying a description of the
context that I wanted, I was able to generate these images from only about ten photos of my face. Why not you ? Come and join the hundreds of people who have already trusted us .

Exemple d'images générées par Dreambooth

Order your styles

1 de 12


Use the small arrow on the right to open / close a text

Generate your images

Pictures of you

An important point to be able to use Dreambooth AI is to have a subject on several angles / positions.

At first, you can simply take about ten photos of your face by taking it from different sides.

You can also vary your state, smiling, sad, angry.

The more photos you have, the more beautiful images the AI ​​can generate in different contexts.

Use Birme (explain below) to edit your photos to 512x512px.


Birme allows you to easily generate your images in 512x512px, for this go to the link indicated above.

Click "Browse from your computer", choose the photos taken previously.

Once the photos are loaded, right click on the "Save files" button.

We will use these photos as the basis for our model later.

Hugging Face

First, Hugging Face is a prerequisite to start generating your images.

You need to click on the 2 links above and start creating an account on Hugging Face.

Once you have created your account, validate it using the email you received.

Return to the page linked above, you must have
a text with the title "You need to agree to share your contact
information to access this model".

At the bottom of this text, you have a checkbox to accept the conditions, check it and validate.

Then at the top right, click on your avatar, go to "Settings" and then on the right "Access Token".

Click on "New token" choose a name, for example your first name, choose "write" in "Role" and click on "Generate a token".

Copy it and put it aside, we will need it later.

Google Colab

The Google Colab proposed by ShivamShrirao is the place where everything
will happen, Google Colab if you don't know lets run
Python code on machines offered by Google. The power of
calculation is therefore important.

The Google Colab is split into several parts, each part
has a small icon at the top left "Play/Execute".

Start with the first "Check type of GPU and VRAM available".

Continue to "Install Requirements".

On "Login to HuggingFace" paste here the token we generated earlier.

Don't forget that for each part, you have to click on the "Play/Run" icon on the left.

In "Settings and run", click on "save_to_drive" and link your Google Drive account to the Google Colab.

Change the name "zwx" to your first name in "Output_dir".

On "Start training", modify the "concepts_list" part to
have in "instance_prompt" "photo of (your first name)" in
"class_prompt" change to "photo of a person" in "instance_data_dir"
modify "zwx" by your first name and in "class_data_dir" modify "dog"
by "person".

Then on "Upload your images by running this cell", run
well the code with "Read/Run" on the left, choose "Browse" and
send the photos of yourself that you took earlier.

Wait a few minutes while your images are processed.

In the meantime in the part below, you can modify "--save_sample_prompt" by "photo of (your first name)".

Once the processing on your images is complete, run this part and wait for several minutes.

Run "Specify the weights directory to use".

Run "Run to generate a grid of preview images from the last saved weights".

A grid with your photos should appear.

Run "Convert weights to ckpt to use in web UIs like AUTOMATIC1111".

Run "Inference".

Run "Can set random seed here for reproducibility" with a seed at "52362".

You are there, now you can generate your images, for
this, you just need to fill "prompt" by the description you
want, do not forget to always specify your context, either
"photo of (your first name)" and then whatever you want.

You can use Lexica (Link above) to find
inspirations and description examples, then you will understand
How does it work.

If you have any errors while processing a step, please feel free to contact me.

Errors encountered

ERROR: xformers-0.0.14.dev0-cp37-cp37m-linux_x86_64.whl is not a supported wheel on this platform.

If you encounter this error, you must update the step "Install xformers from precompiled wheel."

Replace line

%pip install -q https://github.com/metrolobo/xformers_wheels/releases/download/1d31a3ac_various_6/xformers-0.0.14.dev0-cp37-cp37m-linux_x86_64.whl


%pip install -q https://github.com/OMGhozlan/xformers_colab/releases/download/1.0.0/xformers-0.0.15.dev0+cu11.2-cp38-cp38-linux_x86_64.whl