Text image confrontation generation system, method and equipment based on multi-modal information prompt and medium

The invention provides a text image confrontation generation method based on multi-modal information prompt, and aims to synthesize text images with different character types and interference attributes under a specified style background. Rich interference attributes and character type information are introduced and serve as attribute prompts for text generation, a random character image is combined to be input into an attribute encoder and a character encoder for feature extraction, meanwhile, background style feature extraction is conducted on a background image based on a style encoder, and background style features are converted into background style features in a decoder part. Performing multi-level deep feature fusion on the attributes and the character features to generate a text image which corresponds to a background and has specified attributes and corresponding character contents and styles; multi-task adversarial training is carried out on the generated image, the authenticity of the text image ge