Generating digital images utilizing high-resolution sparse attention and semantic layout manipulation neural networks

A non-transitory computer-readable medium storing instructions that, when executed by at least one processor, cause a computing device to:

• generate a warped image from a digital image and an edited semantic layout utilizing a sparse attention warped image neural network;
• generate, from the warped image and the digital image, a locally warped image by replacing warped pixels outside of a masked portion of the warped image with pixels from the digital image;
• extract a warped image semantic feature set from the locally warped image utilizing a first encoder of a digital image layout neural network;
• generate an encoded feature set from the edited semantic layout, an image mask, and the warped image semantic feature set utilizing a second encoder of the digital image layout neural network;
• and
• generate a refined digital image from the encoded feature set utilizing a decoder of the digital image layout neural network.

A system for generating high-resolution digital images from edited semantic layouts, the system comprising:

• one or more memory devices comprising:
• a digital image layout neural network comprising an encoder, a coarse decoder, and a fine decoder;
• and
• a digital image, an edited semantic layout of the digital image, an image mask corresponding to the edited semantic layout, and a warped image of the digital image aligned to the edited semantic layout;
• and
• at least one processor configured to cause the system to:
• generate, from the warped image and the digital image, a locally warped image by replacing warped pixels outside of a masked portion of the warped image with pixels from the digital image;
• generate a set of encoded features from the edited semantic layout, the image mask, and the locally warped image utilizing the encoder;
• determine a first set of decoded features from the set of encoded features utilizing the coarse decoder;
• and
• generate a revised digital image from the first set of decoded features utilizing the fine decoder.