Search

CN-122023856-A - Image generation method, device, electronic equipment, storage medium and product

CN122023856ACN 122023856 ACN122023856 ACN 122023856ACN-122023856-A

Abstract

The application relates to an image generation method, an image generation device, electronic equipment, a storage medium and a product. The method comprises the steps of obtaining a first reference image, wherein the first reference image meets a preset image quality requirement, obtaining at least one piece of first preference information indicating a target community channel, generating a first target image corresponding to the first reference image by using an image processing model as a guide, wherein the first target image is to-be-issued content of a target object, the target object issues content in a cold start stage of the target community channel, the image processing model indicates training results aiming at candidate generation type models based on a plurality of sample images, and the candidate generation type models pre-train and determine the preset generation type models based on a plurality of sample contents. The application provides an image generation scheme which takes product characteristics and generation efficiency into consideration in a cold start stage of a target community channel.

Inventors

  • LIU GANG

Assignees

  • 腾讯科技（深圳）有限公司

Dates

Publication Date
20260512
Application Date
20241111

Claims (13)

  1. 1. An image generation method, applied to a target application, the method comprising: Acquiring a first reference image, wherein the first reference image meets the preset image quality requirement; acquiring at least one first preference information indicating a target community channel, wherein the target community channel is a newly created community channel in the target application; Generating a first target image corresponding to the first reference image by using the image processing model as guidance, wherein the first target image is matched with the first reference image in content, the first target image has an association relationship with the at least one first preference information, the first target image is to-be-distributed content of a target object, the target object performs content distribution in a cold start stage of the target community channel, the image processing model indicates training results of candidate generation models based on a plurality of sample images, the candidate generation models perform pre-training determination on the preset generation models based on a plurality of sample contents, the plurality of sample contents are a plurality of to-be-selected contents meeting preset requirements in a to-be-selected content set, the to-be-selected content set is composed of historical distribution contents in the cold start stage of each of a plurality of first association community channels, and the first association community channels are community channels which have completed cold start in the target application.
  2. 2. The method according to claim 1, wherein the method further comprises: determining the target number of interactive operations for the first to-be-selected content in a preset duration by taking the release time of the first to-be-selected content as a time starting point for each first to-be-selected content in the to-be-selected content set; Determining a plurality of second to-be-selected contents, of which the target number is larger than a first preset number, from the plurality of first to-be-selected contents; for each content topic type, determining the sample content conforming to the content topic type from the second candidate contents, wherein the content topic type is determined based on content topic preference information of the target community channel.
  3. 3. The method according to claim 1, wherein the method further comprises: In the cold start stage of the target community channel, determining whether the target community channel is cold started or not according to the object condition of content release in the current time period and the interaction condition aiming at the released content; Under the condition that the target community channel does not complete cold start, determining sub-interaction conditions corresponding to the multiple preference information types based on the interaction conditions, and generating adjustment information based on the sub-interaction conditions corresponding to the multiple preference information types, wherein the adjustment information is used for guiding adjustment of preference information to be adopted in the next time period.
  4. 4. A method according to any one of claims 1-3, wherein in case the pre-set generative model is a first text-to-graph diffusion model, the image processing model is obtained by training the following steps: Acquiring the first text-generated graph diffusion model and a preset low-rank matrix to construct the candidate generation model, wherein the preset low-rank matrix is used for representing the weight of key parameters in the first text-generated graph diffusion model; taking the sample image and preset guide information as inputs, and outputting a predicted image corresponding to the sample image by using the candidate generation type model; And adjusting the preset low-rank matrix based on a first difference or a second difference to obtain the image processing model, wherein the first difference indicates the difference between the predicted image and the sample image, the second difference indicates the difference between the predicted image and a labeling image, and the labeling image indicates the prediction result of a large language model aiming at the sample image and the preset guide information.
  5. 5. The method of any of claims 1-3, wherein the obtaining at least one first preference information indicative of a target community channel comprises: Determining the at least one first preference information based on the first characterization information of the target community channel, wherein the first preference information corresponds to any one of an image theme dimension, an image style dimension and an image detail dimension, and the first characterization information of the target community channel comprises at least one of a service type of the target community channel, at least one target image label corresponding to the target community channel and representative content in similar community channels of the target community channel, and the at least one target image label is determined based on at least one preset image label corresponding to the target community channel and an image label carried by at least one released image.
  6. 6. The method of claim 5, wherein after the determining the at least one first preference information based on the first characterization information of the target community channel, the method further comprises: constructing a plurality of preference information groups based on at least two of the first preference information in the presence of the at least two of the first preference information; The generating, by using the at least one first preference information as a guide and using an image processing model, a first target image corresponding to the first reference image includes: generating a candidate image corresponding to the first reference image by using the image processing model by taking the preference information group as a guide for each preference information group in the plurality of preference information groups to obtain a candidate image set, wherein the candidate image is matched with the first reference image in content, and the candidate image has an association relationship with the preference information group; And carrying out image filtering on the candidate image set based on a first similarity threshold value to obtain a target image set, wherein the similarity between any two first target images in the target image set is smaller than the first similarity threshold value.
  7. 7. The method of claim 5, wherein the method further comprises: acquiring a reference text conforming to image subject information in the case where the first preference information is the image subject information; And taking the reference text as input, and outputting a corresponding second target image by using a second text graph diffusion model.
  8. 8. A method according to any one of claims 1-3, wherein before generating the first target image corresponding to the first reference image using the image processing model, the method further comprises: determining at least one second preference information based on the second characterization information of the target community channel, wherein the second preference information corresponds to any one of an image theme dimension, an image style dimension and an image detail dimension, and the second characterization information of the target community channel comprises at least one of object information of each of a plurality of target community member objects of the target community channel and similar content preference corresponding to at least two associated community member objects, the at least two associated community member objects and the target community member objects indicate the same identity account, the at least two associated community member objects correspond to at least two second associated community channels in the target application, and the content preference similarity of the at least two associated community member objects in the at least two second associated community channels is larger than a second similarity threshold; The generating, by using the at least one first preference information as a guide and using an image processing model, a first target image corresponding to the first reference image includes: And generating the first target image corresponding to the first reference image by using the image processing model with the at least one piece of first preference information and the at least one piece of second preference information as guidance, wherein the first target image has an association relationship with the at least one piece of first preference information and the at least one piece of second preference information.
  9. 9. A method according to any one of claims 1-3, wherein the method further comprises: Acquiring an image release request sent by a preset object, wherein the image release request comprises a second reference image, and the preset object is any object of a plurality of target community member objects of the target community channel; determining at least one piece of third preference information based on the object information of the preset object, wherein the third preference information corresponds to any one of an image theme dimension, an image style dimension and an image detail dimension; And generating a third target image corresponding to the second reference image by using the image processing model and issuing the third target image, wherein the third target image is matched with the second reference image in content, and the second target image and the at least one third preference information have an association relation.
  10. 10. An image generation apparatus for application to a target application, the apparatus comprising: The image acquisition module is used for acquiring a first reference image, and the first reference image accords with the preset image quality requirement; The first preference information acquisition module is used for acquiring at least one first preference information indicating a target community channel, wherein the target community channel is a newly created community channel in the target application; The first image generation module is used for generating a first target image corresponding to the first reference image by taking the at least one first preference information as a guide and utilizing an image processing model, wherein the first target image is matched with the first reference image in content, the first target image has an association relationship with the at least one first preference information, the first target image is to-be-issued content of a target object, the target object issues content in a cold start stage of the target community channel, the image processing model indicates training results of candidate generation models based on a plurality of sample images, the candidate generation models pre-train and determine a preset generation model based on a plurality of sample contents, the plurality of sample contents are a plurality of to-be-selected contents meeting preset requirements in a to-be-selected content set, the to-be-selected content set is composed of historical issued contents in respective cold start stages of a plurality of first association community channels, and the first association community channels are community channels which have completed cold start in the target application.
  11. 11. An electronic device comprising at least one processor and a memory communicatively coupled to the at least one processor, wherein the memory stores at least one instruction or at least one program loaded and executed by the at least one processor to implement the image generation method of any of claims 1-9.
  12. 12. A computer readable storage medium having stored therein at least one instruction or at least one program loaded and executed by a processor to implement the image generation method of any of claims 1-9.
  13. 13. A computer program product comprising at least one instruction or at least one program loaded and executed by a processor to implement the image generation method of any of claims 1-9.

Description

Image generation method, device, electronic equipment, storage medium and product Technical Field The present application relates to the field of internet communications technologies, and in particular, to an image generating method, an image generating device, an electronic device, a storage medium, and a product. Background With the development of internet communication technology, various internet products are layered endlessly. Taking a community product as an example, the community product aims to promote communication, interaction and collaboration among users. In the related art, the cooperative professional production content (Professional Generated Content, PGC) is often relied on in the cold start phase of the community product. However, such a content generation method is inefficient and costly. Accordingly, there is a need to provide a more efficient content generation scheme. Disclosure of Invention In order to solve at least one technical problem set forth above, the present application provides an image generating method, an apparatus, an electronic device, a storage medium, and a product: according to a first aspect of the present application, there is provided an image generation method applied to a target application, the method comprising: Acquiring a first reference image, wherein the first reference image meets the preset image quality requirement; acquiring at least one first preference information indicating a target community channel, wherein the target community channel is a newly created community channel in the target application; Generating a first target image corresponding to the first reference image by using the image processing model as guidance, wherein the first target image is matched with the first reference image in content, the first target image has an association relationship with the at least one first preference information, the first target image is to-be-distributed content of a target object, the target object performs content distribution in a cold start stage of the target community channel, the image processing model indicates training results of candidate generation models based on a plurality of sample images, the candidate generation models perform pre-training determination on the preset generation models based on a plurality of sample contents, the plurality of sample contents are a plurality of to-be-selected contents meeting preset requirements in a to-be-selected content set, the to-be-selected content set is composed of historical distribution contents in the cold start stage of each of a plurality of first association community channels, and the first association community channels are community channels which have completed cold start in the target application. According to a second aspect of the present application, there is provided an image generation apparatus for application to a target application, the apparatus comprising: The image acquisition module is used for acquiring a first reference image, and the first reference image accords with the preset image quality requirement; The first preference information acquisition module is used for acquiring at least one first preference information indicating a target community channel, wherein the target community channel is a newly created community channel in the target application; The first image generation module is used for generating a first target image corresponding to the first reference image by taking the at least one first preference information as a guide and utilizing an image processing model, wherein the first target image is matched with the first reference image in content, the first target image has an association relationship with the at least one first preference information, the first target image is to-be-issued content of a target object, the target object issues content in a cold start stage of the target community channel, the image processing model indicates training results of candidate generation models based on a plurality of sample images, the candidate generation models pre-train and determine a preset generation model based on a plurality of sample contents, the plurality of sample contents are a plurality of to-be-selected contents meeting preset requirements in a to-be-selected content set, the to-be-selected content set is composed of historical issued contents in respective cold start stages of a plurality of first association community channels, and the first association community channels are community channels which have completed cold start in the target application. According to a third aspect of the present application, there is provided an electronic device comprising at least one processor and a memory communicatively coupled to the at least one processor, wherein the memory has stored therein at least one instruction or at least one program that is loaded and executed by the at least one processor to implement the image generation method as described i