Generating tile-able patterns from text

Patent metadata

Field	Value
Publication number	US-12579608-B2
Application number	US-202318526855-A
Country	US
Kind code	B2
Filing date	Dec 1, 2023
Priority date	Jun 13, 2023
Publication date	Mar 17, 2026
Grant date	Mar 17, 2026

Abstract

Systems and methods for generating tile-able patterns from text include obtaining a text prompt and generating, by a generation prior model, a latent vector based on the text prompt, where the generation prior model is trained to output vectors within a distribution of tile-able patterns. An image generation model then generates an output image based on the latent vector. The output image comprises a tile-able pattern including an element from the text prompt.

First claim

What is claimed is:

1 . A method comprising:
obtaining a text prompt; generating, using a generation prior model, a latent vector based on the text prompt by sampling the latent vector from a learned cluster in an embedding space, wherein the generation prior model is trained to output vectors within a distribution of tile-able patterns corresponding to the learned cluster, and wherein the learned cluster includes vector representations of tile-able images; and generating, using an image generation model, an output image based on the latent vector, wherein the output image comprises a tile-able pattern including an element from the text prompt.

2 . The method of claim 1 , wherein generating the output image comprises:
performing a circular convolution.

3 . The method of claim 1 , further comprising:
encoding the text prompt to obtain a text embedding, wherein the latent vector is generated based on the text embedding.

4 . The method of claim 1 , further comprising:
applying the output image to an image canvas using a brush tool.

5 . The method of claim 1 , further comprising:
generating a plurality of latent vectors based on the text prompt; and computing a plurality of similarity scores corresponding to the plurality of latent vectors, respectively, wherein the latent vector is selected from the plurality of latent vectors based on the plurality of similarity scores.

6 . The method of claim 1 , wherein:
the generation prior model is trained based on a training set including a plurality of images depicting tile-able patterns.

7 . The method of claim 1 , wherein:
the output image comprises a plurality of repetitions of the tile-able pattern.

8 . The method of claim 1 , wherein generating the latent vector further comprises:
performing a reverse diffusion process.

9 . A method comprising:
obtaining training data including a plurality of images depicting tile-able patterns and a plurality of text descriptions corresponding to the plurality of images, respectively; and training a generation prior model to generate latent vectors for an image generation model within a distribution of tile-able patterns based on the training data by sampling the latent vectors from a learned cluster in an embedding space, wherein the learned cluster includes vector representations of tile-able images.

10 . The method of claim 9 , wherein training the generation prior model further comprises:
generating a predicted latent vector using the generation prior model; generating a predicted image based on the latent vector using the image generation model; and computing a loss function based on the predicted image.

11 . The method of claim 10 , wherein generating the predicted image comprises:
performing a circular convolution.

12 . The method of claim 9 , further comprising:
encoding each of the plurality of text descriptions to obtain a plurality of text embeddings, wherein the training is based on the plurality of text embeddings.

13 . The method of claim 9 , wherein obtaining the training data comprises:
obtaining an initial set of images; and classifying the initial set of images using a binary classifier model to obtain the plurality of images.

14 . An apparatus comprising:
at least one processor; at least one memory including instructions executable by the at least one processor; the apparatus further comprising a generation prior model including parameters stored in the at least one memory and trained to output vectors within a distribution of tile-able patterns by sampling the latent vector from a learned cluster in an embedding space, wherein the learned cluster includes vector representations of tile-able images; and an image generation model including parameters stored in the at least one memory and trained to generate tile-able patterns based on output from the generation prior model.

15 . The apparatus of claim 14 , wherein:
the generation prior model comprises a diffusion model.

16 . The apparatus of claim 14 , wherein:
the image generation model comprises a diffusion model.

17 . The apparatus of claim 14 , wherein:
the image generation model includes a circular convolution layer.

18 . The apparatus of claim 14 , further comprising:
a text encoder configured to encode a text prompt to obtain a text embedding.

19 . The apparatus of claim 14 , further comprising:
a training component configured to train the generation prior model.

20 . The apparatus of claim 19 , further comprising:
a binary classifier model configured to filter a dataset to obtain training data.

Assignees

• Adobe Inc

Inventors

• Batra Vineet
• Chaturvedi Sumit
• Rai Abhishek
• Aggarwal Pranav Vineet
• Kale Ajinkya Gorakhnath
• Jeph Aman
• Phogat Ankit
• Dhingra Sumit
• Chen Fengbin
• Garg Kshitiz
• Hasan Milos
• Harikumar Midhun
• Pathak Gaurav Suresh
• Chakraborty Souymodip

Classifications

• G06T11/10

  Texturing; Colouring; Generation of textures or colours (retouching, inpainting or scratch removal G06T5/77) · CPC title

• G06T2200/32

  involving image mosaicing · CPC title

• G06T3/40

  Scaling of whole images or parts thereof, e.g. expanding or contracting · CPC title

• G06T2200/24

  involving graphical user interfaces [GUIs] · CPC title

• G06F40/40

  Processing or translation of natural language (natural language analysis G06F40/20; semantic analysis G06F40/30) · CPC title