Abstract
What would the great poets of the past think and write about the objects of the twenty-first century? PoetrEye is an interactive experiment that generates abstract concrete poems, making poetic our everyday life. This project has been developed with the purpose of realizing a project in the context of creativity and visual expression using machine learning technology as an input or output.
Interface, Technology and Interaction Modality
PoetrEye is a tool that animates poems generated by ML: through a camera it detects any object and then creates a poetic composition about them using the style of the great poets of the past. The ML model Runway DenseCap detects the objects using as input the field of the view of a connected camera. The name of the objects are then used as input for creating poem throughout CharRNN, which uses a a model that we have trained with a selection of poems of the past. The output is a visual animation of the generated poem, line per line, in form of abstract concrete poems.
User experience
We realized a webpage on which there is just a button “Generate”. By clicking on it, then there will appear on the page an animation of the poem: each line compares, all the words that compose it start from a center point and then move around the space, creating a final abstract concrete poem. The only colors used in the website are white and black in order to focus the attention of the user exclusively on the text generated by ML, using a minimal and elegant style.
Research and Development Context
PoetrEye is an experiment that enlightens the power of ML, the capability of being more and more similar to humans. It could be interesting to train more the model that we have created using the poems of the past, in order to create poems that are more consistent and profound.
