Pre-Publication V2. A software system trained on large datasets intended to that synthesizes original visual content from input prompts including natural language descriptions, reference images, sketches, or combinations thereof using generative machine learning models such as diffusion models, GANs, or multimodal transformers. Output may range from photorealistic imagery to stylized or abstract visuals, and can include still images, image variations, or edited versions of existing content. AI image generators may function as standalone tools or as embedded components within larger creative, production, or automated workflows.
Deliberation Summary:
(1) Objected to describing output as original, as diffusion models are trained to reconstruct their training set and it is unclear what counts as original in this context. (2) Proposed adding that these systems are trained on large datasets, consistent with the parallel entry for video generation. (3) Raised concern that a general reader may not appreciate the scale involved and asked whether that detail belongs here or in the separate entry addressing datasets.
Pre-Publication V1. A software system that synthesizes original visual content from input prompts including natural language descriptions, reference images, sketches, or combinations thereof using generative machine learning models such as diffusion models, GANs, or multimodal transformers. Output may range from photorealistic imagery to stylized or abstract visuals, and can include still images, image variations, or edited versions of existing content. AI image generators may function as standalone tools or as embedded components within larger creative, production, or automated workflows.