DESC:FIELD OF INVENTION
[0001] The present invention is directed to artificial intelligence (AI) based system and method for garment generation, and more particularly to AI based garment generation system and method at a low computational cost with UV fitting for print ready format and 3D preview.
BACKGROUND OF INVENTION
[0002] People love being unique. Even if a person picks a garment today, he can find at least 10-50 people owning the same dress the other day. What if each customer is provisioned with a unique garment to a customer that no one owns. This is particularly relevant in today’s scenario where various industries are thriving on uniqueness of individual appearances, fashion trends, customized apparels and the like. Moreover, seldom is one interested in repeating one’s outfit. Further, after buying the garment of one particular design it is rarely sold to another person.
[0003] Furthermore, getting custom designed unique outfits made each time from designer is too heavy on pockets. In the background of foregoing limitation, there exists a need for an advanced and intelligent system and method that can design unique garments for unique individuals and meet growing demand of fashion and garment industry at low cost.

OBJECT OF THE INVENTION
[0004] The primary objective of the present disclosure is to provide an intelligent and advanced artificial intelligence (AI) based system and method for unique garment generation for unique individuals.
[0005] In one other objective of the present disclosure, an AI based garment generation system and method is provided at a low computational cost.
[0006] Another objective of the present disclosure is to provide a cost-effective AI based garment generation system and method that is UV fitted for being print ready and for 3D preview.
[0007] In yet another objective of the present disclosure, an automated AI based garment generation system and method that provides users with multiple design options to choose from.
SUMMARY OF THE INVENTION
[0008] The various embodiments of the present invention provide an artificial intelligence (AI)-based system designed for generating unique garment designs. This system includes a prompt engineering module that generates design prompts from text and image embeddings. It is closely integrated with a texture map generation module, which takes these prompts and images, encodes them into sequences of tokens, and then decodes these tokens into initial garment design images. Following this, a region customization module applies specific design elements to designated areas of the garment, such as logos or styles on sleeves and shoulders. An upscaling module enhances the resolution of these garment design images to ensure high-quality print readiness and 3D visualization. The final step involves a UV generation module that prepares these images for UV fitting and print readiness, culminating in a printing module that produces the final garment design in a physical format. The system uniquely features generative models in the prompt engineering module for creating unsupervised, context-specific design prompts, employs bidirectional transformer encoders and decoders in the texture map generation, and utilizes a Residual Dense Network in the upscaling module for high-quality image enhancement.
[0009] The various embodiments of the present invention also provide a method for generating unique garment designs based on an AI-based system. The method starts by generating design prompts using a prompt engineering module that leverages text and image embeddings to create context-specific prompts. These prompts, along with images, are encoded into sequences of tokens and decoded back into initial garment design images by a texture map generation module. Specific design elements are then applied to designated regions of the garment, such as adding logos or stylizing sleeves and shoulders, through a region customization module. To ensure the designs are of high quality, an upscaling module enhances the resolution of the images, preparing them for UV fitting and print readiness with the help of a UV generation module. Finally, the designs are brought into the physical world by a printing module that produces the final garment. This method notably involves the use of a generative model for creating design prompts, the application of bidirectional transformer encoders and decoders for image processing, and the enhancement of image resolution through a Residual Dense Network to achieve print and visualization-ready designs without losing image quality.
BRIEF DESCRIPTION OF THE DRAWINGS
[0008] Fig. 1 shows set of UV ready formats for 3D model fitting and garment printing, in accordance with one exemplary embodiment of present disclosure.
[0009] Fig. 2 shows flow chart and method for AI based garment generation, in accordance with one exemplary embodiment of present disclosure.
DETAILED DESCRIPTION OF THE EMBODIMENTS
[0010] Before the present working principle of AI based garment generation is described, it is to be understood that this disclosure is not limited to the particular means and mode for achieving so, as described, since it may vary within the specification indicated. It is also to be understood that the terminology used in the description is for the purpose of describing the particular versions or embodiments only, and is not intended to limit the scope of the present invention, which will be limited only by the appended claims. The words "comprising," "having," "containing," and "including," and other forms thereof, are intended to be equivalent in meaning and be open ended in that an item or items following any one of these words is not meant to be an exhaustive listing of such item or items, or meant to be limited to only the listed item or items. The disclosed embodiments are merely exemplary methods of the invention, which may be embodied in various forms.
[0011] The system and method for present disclosure provides an artificial intelligence (AI) based garment generation at a low computational cost with UV fitting for print ready format and for 3D preview. Once the garment is sold or picked by a consumer, it is not sold again. In one preferred aspect of present disclosure, this unique design generation is based on diffusion model which is converted into 3D UV format and printable formats for completely automated production. Fig. 1 shows a set of UV ready formats for 3D model fitting (enabling visualization during shopping) and garment printing (UV ready print files).
[0012] Accordingly the present disclosure by way of developing unique AI generated designs for garments, makes it easy for users to select garment designs from millions of available designs that gets generated by AI based system. In one preferred embodiment, machine learning is used for art generation along with usage of UV creation algorithms and image manipulation algorithms to accomplish unique features of present invention. In one exemplary embodiment, the unique features achieved by aforementioned combination of algorithms includes, though not limited to:
- Hyper resolution boost for printer formats
- Logo placements (in case of sports)
- Name and number placements (in case of sports)
- Region customization such as stylized sleeves, shoulder designs.
[0013] For the purposes of present disclosure, appropriate machine learning is used to design garments in a unique fashion in countless numbers, still being non-repetitive. Referring to Fig. 2, the system 1000 uses vector quantized adversarial networks to generate designs followed by a region level customization and UV fitting to finally form garments that can be visualized in 3D and also pushed for production prints. In one working embodiment of present disclosure, the components of present system 1000 comprises of:
Prompt engineering (100)
Texture map generation (200)
Region customization (300)
Upscaling (400)
UV generation (500)
Printing (600)
[0014] Drawing from above, referring to first component, prompt engineering (100) works based on text and image embedding that is trained and modeled with millions of image and text pairs that form the basis for the ML model to engineer a given prompt in its own way. The model is a generative model that builds a prompt result that is truly unsupervised yet sticks to the context.
The prompt engineering (100) works for the main
- Garment design,
- Region Customization such as sleeves and shoulders
- Logos
[0015] Now, re-referring to Fig. 2 and next component i.e. Texture map generation (200), the input prompt is encoded via a bidirectional regressive transformer encoder. The output of the encoder and the images which are encoded tokens by the vector quantized adversarial encoder (which turns images into a sequence of tokens) are then sent through the bidirectional transformer decoder, which is an auto-regressive model whose goal is to predict the next token. Then the sequences of predicted image tokens are decoded through the vector quantized decoder.
[0016] The text-to-image encoding model guides the Adversarial network's decoded output images to the best match for the given input prompt. The similarity between the image and input prompt can be represented by the Cosine similarity of the learnt feature vector. The generated output image is then sent to the upscaled via a residual dense network. All the images are generated in a resolution of 256*256. This is done to reduce memory constraints.
[0017] In accordance with one exemplary embodiment, it is pertinent to note that the two UV files are always generated- a) one for the front; b) one for the rear. Now referring to third component, region customization (300) is attempted whereby UV files are generated for the garments. The purpose is to use the image noise and background removal libraries and front libraries to:
- Superimpose logo on the desired location in the garment with a shape whose background is inversion of average color scale across the UV texture generated.
- Add styles of different styles to sleeves and shoulders.
- Add names to the rear UV whole color is an inversion of the average color like the above.
[0018] Now, Upscaling (400) is performed through hyper resolution process. This step creates a high resolution UV map that is to be fed on a printer or a 3D rendering engine. Upscaling (400) is achieved through Residual Dense Network, which does the process of upscaling and enhancing an image. Thus, the system (1000) improves the image resolution without any loss in the generated output image. Hyper resolution process is multi-stage with a pixel level feedback loop repeated till a 2k image resolution is obtained for printing and visualization.
[0026] Lastly, in the final step, UV formal fitting (500) is applied whereby UV format fitting assembles and fits the dimensions as required by the printer (600) to print the garment. Separate formats are generated for the printer (600) and one for the 3D visualization process. Every garment generated creates two folders with a printer necessary files and a web folder with necessary files for 3D preview, capable of running in a web server like apache.
[0027] The foregoing description is a specific embodiment of the present disclosure. It should be appreciated that this embodiment is described for purpose of illustration only, and that numerous alterations and modifications may be practiced by those skilled in the art without departing from the spirit and scope of the invention. It is intended that all such modifications and alterations be included insofar as they come within the scope of the invention as claimed or the equivalents thereof.
,CLAIMS:We claim:
1. A system for automated generation of unique garment designs, the system comprising:
a prompt engineering module (100) configured to generate design prompts based on text and image embeddings;
a texture map generation module (200) operatively connected to the prompt engineering module (100), configured to encode input prompts and images into sequences of tokens and decode these tokens into initial garment design images;
a region customization module (300) operatively connected to the texture map generation module (200), configured to apply specific design elements to designated regions of the garment design;
an upscaling module (400) operatively connected to the region customization module (300), configured to enhance the resolution of the garment design images;
a UV generation module (500) operatively connected to the upscaling module (400), configured to prepare the enhanced garment design images for UV fitting and print readiness; and
a printing module (600) operatively connected to the UV generation module (500), configured to produce the final garment design in a physical format.

2. The system as claimed in claim 1, wherein the prompt engineering module (100) utilizes a generative model to create unsupervised, context-specific design prompts from millions of image and text pair embeddings.

3. The system as claimed in claim 1, wherein the texture map generation module (200) employs a bidirectional transformer encoder and decoder for the conversion of encoded prompts and images into initial design images.

4. The system as claimed in claim 1, wherein the region customization module (300) includes image noise and background removal libraries for superimposing logos and applying styles to specific parts of the garment design such as sleeves and shoulders.

5. The system as claimed in claim 1, wherein the upscaling module (400) utilizes a Residual Dense Network for enhancing the resolution of the garment design images without loss in image quality, achieving a resolution suitable for high-quality printing and 3D visualization.

6. A method for automated generation of unique garment designs, the method comprising:
generating design prompts based on text and image embeddings using a prompt engineering module (100);
encoding the generated prompts and images into sequences of tokens and decoding these tokens into initial garment design images using a texture map generation module (200);
applying specific design elements to designated regions of the garment design using a region customization module (300);
enhancing the resolution of the garment design images using an upscaling module (400);
preparing the enhanced garment design images for UV fitting and print readiness using a UV generation module (500); and
producing the final garment design in a physical format using a printing module (600).

7. The method as claimed in claim 6, further comprising the step of utilizing a generative model within the prompt engineering module (100) to create unsupervised, context-specific design prompts.

8. The method as claimed in claim 6, further including employing a bidirectional transformer encoder and decoder within the texture map generation module (200) for the conversion of encoded prompts and images into initial design images.

9. The method as claimed in claim 6, further including utilizing image noise and background removal technologies within the region customization module (300) for superimposing logos and applying styles to specific parts of the garment design.

10. The method as claimed in claim 6, wherein enhancing the resolution of the garment design images includes using a Residual Dense Network within the upscaling module (400) to achieve a resolution suitable for high-quality printing and 3D visualization without loss in image quality.