AI Photoshoot / DreamBooth

DreamBooth

AI Photoshoot

With this node you can train AI to generate images with any subject from person to your product

Quick menu: click on any section to quickly go to it

Key features

Step by step tutorial

How to use trained model

Recommendations for photos

Parameters description

Key features

Train on anything
Supports training on any subject: people, animals, products, illustrations, styles
Customizable training
Set up training for anything using the necessary parameters: perfect for beginners and advanced users
Available to the subscribers
All users with Premium and Plus subscriptions get access to AI training node
Custom models training
Advanced feature for the experienced AI users: choose from 30+ Stable Diffusion checkpoints to train on
Presets of parameters
Quick setup with the best settings for each use case

How to use the node

Click on each step to see the corresponded image as an example and tutorial to follow

1. Upload images to the workspace
The easiest way is to drag and drop them on the workspace. You can also add Import files node and click on Select files.
2. Connect images to the node
Add the node AI photoshoot by clicking on it in the left panel, or by right clicking on the empty workspace, selecting Train model in the dropdown menu and choosing AI photoshoot.
Connect images by dragging the prolonged oval to the orange circle of AI photoshoot node
3. Choose the category
Select the option from dropdown list.
4. Type unique name
The name should be unique. It should be something that AI doesn't know yet. Avoid simple names (use DariaWnd, not just Daria) and black spaces (DariaWnd, not Daria Wnd).
Good examples: DariaWnd, AlxDrin
5. Press Launch and wait
Press Start and wait for the training to be completed (usually takes up to 30 minutes). After it's done, you'll see an image in the node saying that Training is done.

How to use the trained model

1. Create Stable Diffusion 1.5 node
You can either
- click on Stable Diffusion 1.5 on the left menu in AI tools
- or right click on the empty workspace, choose Generate images category and click on Stable Diffusion 1.5
2. Select My models
By default, when you create a node, a random style will be chosen.
In order to use your model, choose --- in the Styles.
Then Choose corresponding newly trained model in My models. It will have the name which you have given it while training + category

After you select a model, you will get the Subject automatically put in the prompt.
Generally speaking, in order to generate with your model, you need to type the unique name which you have given your model, so it can replicate the trained subject. In Phygital+ it's done automatically.
3. Generate
Type your prompt and launch!

Need inspiration for prompts?

Check out our prompt collection!

Explore prompts

Which photos should you use for training on people?

Rule of Thumb: more photos = better results. But photos should be different from one another.

Requirements for photos of people
• all parts of the face should be visible
• good lighting and quality on the image
• different clothing
• different background
• different angles
• no far away shots

Parameters for training

Generally you would need only to type Unique name and select Category, but here are the detailed description of parameters

Subject

It should describe what is shown on your uploaded images. Later you will use what you write in this text field as a prompt in Stable Diffusion.
It has to be unique, something that Stable Diffusion doesn't know yet. It must contain two things:
1) unique name of person. Avoid simple names (use DariaWnd, not just Daria) and blank spaces (DariaWind, not Daria Wnd) for better results
2) subject type (man, woman, shoe, etc.)
For example, 'DariaWnd Woman', 'EkonikaBt Shoe'

Class Prompt

This prompt is used by AI to generate additional images. These additional images will make the training results better.
These prompts work great for training:

on people: 'a photo of woman, ultra detailed' or 'a photo of man, ultra detailed'
on animals: photo of cat, ultra detailed
on products/objects: photo of bag, ultra detailed
on style: artistic style (game style, illustration style, etc.)

Steps

We recommend to use around 1200-1500 steps. It will take longer (around 30 min), but you will get more similarities to original person.

Recommended values for training:

on people: 1500 for 10 photos
on objects: 1000 for 10 photos
on animals: 1200 for 10 photos
on style: 1000 for 25 images

Style

Defines on which model the training will be done. Generally speaking, it's advised to use base model (plain Stable Diffusion 1.5), as changing this value may worsen the results.
However, if you feel confident in your AI experience, you can try choosing some style.
There are more than 30 models available, both photorealistic and stylized. If you want to dive deeper into this topic, we highly recommend checking out our article on training on custom checkpoints

Class Images

Defines how many images will be generated in addition to your loaded dataset. The images are generated using Class prompt.
Recommended values for training:

on people: 100
on objects: 0
on animals: 100
on style: 0

Training seed

Defines the starting point for training. You can try adjusting this value and use different seed every time, but it won't affect the training that much