Textual design agent

One or more computer-storage media having computer-executable instructions embodied thereon that, when executed by a computing system having a processor and memory, cause the computing system to perform a method of generating a textual design, the method comprising:

• receiving a text to place on a digital canvas having a background image;
• generating a stacked input feature map of the background image using a neural network, wherein the stacked input feature map includes a plurality of layers representing different image features, wherein at least one of the plurality of layers is a masking layer;
• generating, with a location prediction model, a recommended location to place the text on the digital canvas, wherein the stacked input feature map is an input to the location-prediction model;
• generating a recommended font, a recommended font size, and a recommended font color for a text design, wherein the recommended location is used as input to generate the recommended font and the recommended font size;
• building a plurality of text designs with a text design engine, each text design in the plurality of text designs comprising a unique combination of the recommended font, the recommended font color, and the recommended font size added to the digital canvas at the recommended location;
• ranking the plurality of text designs using a text-design ranking model that evaluates images of the plurality of text designs;
• and
• outputting from the plurality of text designs a recommended text design having above a threshold ranking.

A method of generating a textual design comprising:

• receiving through a graphical user interface a text to add to a digital canvas having a background design;
• generating a stacked input feature map of the background design using a neural network, wherein the stacked input feature map includes a plurality of layers representing different image features, wherein at least one of the plurality of layers is a masking layer;
• generating, with a location prediction model, a location to place the text on the digital canvas, wherein the stacked input feature map is an input to the location-prediction model;
• generating, with a font prediction model, a plurality of predicted fonts for the text, wherein the location is an input to the font prediction model;
• generating, with a font-size prediction model, a plurality of predicted font sizes for the text, wherein the location is an input to the font-size prediction model;
• generating, with a font-color prediction model, a plurality of predicted font colors for the text, wherein the location is an input to the font-color prediction model;
• and
• outputting for display in the graphical user interface a recommended text design comprising the text inserted into the digital canvas at the location, wherein the text is displayed with a font from the plurality of predicted fonts, a font-size from the plurality of predicted font sizes, and a font-color from the plurality of predicted font colors.