Guide
What is Exactly.ai
What is Exactly.ai?
Exactly.ai is an AI-powered tool for artists and visual creators. It lets you train your own AI model then create images in your style, in seconds, with a simple description. It allows artists to visualise elaborate scenes, which would normally require intensive work. We want to make powerful AI tools available to everyone. Our interface and workspace are user-friendly and intuitive. Here is a short explanation of some of the terms we use that you might find helpful.
What is a model?
An AI visual generative model is an algorithm that can create images based on written instructions. Exactly lets you train your AI model and create images in your style. When an image is created with an AI model, the creation process is sometimes called generation. When you want an AI model to create an image, you typically instruct it with a textual description of the desired image. This written instruction is sometimes called a prompt. Models are instructed with prompts. A prompt tells a model which elements and objects to include in the created image.
Legal and ownership
What are the legal principles of Exactly?
We would like you to read the Terms of Use and Copyright Policy to know all the policies we have in place. But here is a quick explainer of the principles:
- Images that you upload to Exactly belong to you.
- If you train an AI model on your work, it belongs to you.
- Images that you create with models trained on your work belong to you.
- Images created with AI models trained on someone else’s work do not belong to you.
- You may secure the rights to use these images from the original artists.
What is a copyright?
Copyright is an intellectual property right that protects the copyright owner by giving them a right to control their work and how it is used. In simple terms, it means that no one else can lawfully use or copy their work without their permission. Owners of copyright can use, sell or licence their work to a third party. To qualify for copyright protection, a work must have some skill, labour, or judgement in its creation.
What is a licence?
A copyright licence is an agreement, usually in writing, between a copyright owner and a third party that gives that third party the legal right to use the copyright-protected work for designated purposes, typically in exchange for payment.
What is the difference between a licence and a copyright?
Copyright is a legal right that protects the copyright owner and their work. In contrast, a copyright licence agreement is a contract by which the copyright owner gives a third party the right to use their copyright-protected work.
If I train a model on my artwork on Exactly, which rights do I have to that model?
If you create an AI model with your artwork, you own the model and any content generated by anyone using your model. You grant Exactly a licence to publish this model on the platform if you want to do so. If your model is published, you and other users can create images with your model. You don’t have to publish all of your models. But you might want to publish some of them, to see what other people can create with your style and technique. If other users create images with your model, you own the copyright to those images.
If I create an image on Exactly, which rights do I have to that image?
If you create an image using a model that belongs to you, you have a copyright to that image. If you create an image with a model that does not belong to you, then the model's owner has a copyright to that image.
Can I sell images that are created with my model to anyone?
Yes, you can. You can sell images created with your model because those images belong to you. You can determine the price and the terms of the agreement yourself. You don’t have to share any proceeds of this transaction with us.
According to the Terms of Use, you also grant us a right to sell licences to images produced with your public models on your behalf. The proceeds of that transaction will be transferred to you.
1. Creating models
Creating a Model
- Go to the Models Page in the Workspace.
- Press the Add Model Button, and name your model.
- Upload 5 to 25 images that showcase the specific style you want your model to learn.
- We recommend using square images sized 512x512 to 8000x8000 in the jpeg and png formats.
- Write short content descriptions for your images with their subject, medium, and style. Wait 10-15 minutes, and you will receive an email when your model is ready.
Can I use more than 25 images to train my model?
The maximum number of images you can train a model on is 25. The more images you use, the better your model will be. However, 10-15 is enough to produce a functional model.
What images should I use to train my model?
When choosing the images, it’s important to make sure they are consistent in style and subject matter. Avoid mixing images that are very different. This will guarantee the best result and create a model that can accurately depict the nuances of your input artwork.
How should I describe the image I use to train my model?
Exactly automatically suggests descriptions for the images you upload. You can then edit and improve the descriptions. Use simple language to make them effective. Mention everything you think is important. E.g., “An oil painting of a landscape with oak trees, green hills, and blue sky.”
What should I do if I accidentally close the page or turn off my device while the model is being trained?
Don’t worry! The model will continue to train and we will send you an email when it’s ready to use.
2. Creating Images
How do I create Images?
Select the model you want to use in the Workspace. Write a description of the image you want to create. Press “Create”. Press multiple times on the Create button to see different versions, experiment, and produce that perfect image.
How to write a good description to instruct my model?
The best kind of input instruction is clear and descriptive. For example, A watercolor painting of a blonde woman wearing a red dress drinking champagne. We recommend the following order when writing a description: the medium of the artwork (watercolor painting, illustration, closeup photograph, etc.), the name and description of the image’s subject, and any other details you would like to see in the image (colors, textures, background, angle, amount of details, style/genre, etc.).
The images I’ve generated don’t look like the ones I’ve uploaded. Why?
If you instruct your model with a description that is drastically different from your uploaded artworks and their descriptions, the images created by your model will look different from your original set style. It’s a good idea to write descriptions similar to the descriptions of the artworks used to train your model. For example, if you have a model trained on images of animals it will easily create a new animal in your style. But if you instruct this model to create a landscape, it will still create a landscape, but its style might differ from the original one.
Can I try to instruct my model to generate something it wasn’t trained to create?
Yes, you can always experiment with your prompt descriptions. A model will work with any description you give it. Change the subject matter, the medium, or the perspective. The results will be unpredictable and not always representative of your style, but they might turn out to be inspiring and useful for your project.
3. Tailoring Images
How to use the Avoid feature?
- Press the More button and select the Avoid option
- Write down the things you would like to avoid in the created image
How can I improve my image’s details?
Press on the Add Details button under the created image and Exactly will automatically improve the overall quality of your image by adding and improving its details and their quality.
How to download images in High Resolution?
- Select your desired resolution under the created image
- Click the Download button
How to remove a watermark from an Image?
Press the More button and select the Hide Watermark option.
4. Feedback and Issues
Where can I leave my feedback, and how can I ask for assistance if I encounter an issue?
You can always tell us about your experience and concerns using the support chat in the Workspace.
What is the difference between a Private and Public Model?
A Private model is a tool that you have created and trained on your artwork. Only you can use it.
A Public model is a tool that has been created and trained by another user. Everyone on Exactly can use it.
Do I need to create multiple models if I work in different styles?
Most of the Exactly community members have dozens of different specialised models they use in their work on different projects. We recommend this practice because the more specialised your model, the more it will be helpful in working on your project. It is best to specifically train a model for a specific purpose.
5. Delete an Account
To make a deletion request, please contact us by support@exactly.ai or via chat in your Workspace. To protect your privacy, we will take steps to verify your identity before fulfilling your request.