Image training is a machine learning-based feature that enables learning from visual data inputs. We use advanced algorithms to identify patterns and features in images, improving the ability to generate new images based on provided input.
This is part of a bigger initiative to provide 'Brand Intelligence' for the project, allowing users to upload brand data and fine-tune outcomes based on this. Each model created within the project is available for this particular project.
To start training a model, navigate to the 'Brand Intelligence' page from the project homepage and add a new model in the 'Image model' section.
Training the model
Create a new model and fill out all required fields.
Upload 5-30 images to train the model.
Set up training settings using either a predefined set or by selecting custom settings. Once done, proceed with training the first version of the model, which will be set by default.
You can create up to 20 versions per each model.
Once the version is trained, you can test the outcomes using the test section on the right-hand side. You can test up to 5 versions at a time.
Select the best resulted version for the model by defining it as the default.
Use of trained model in image generation
You can use the trained model in all image generation places and tools (except Content matrix). To proceed with creating images with the trained model, navigate to the Image generation experience or tools and create in a prompt field description of the desired image using #model_name there, # defines what model should be used.
You can change the default version of the model in model settings to change the generated outcome.
All image generation settings and styles could also be applied while generating images with your trained model.
Enjoy your experiments and please let us know your feedback and outcomes in Intercoms chat.