Method for generating images of synthetic products that are immediately manufacturable
The disclosed technology includes a computer-implemented technique for generating images of synthetic products that are manufacturable as physical products is disclosed. The system receives user input describing a conceptual product via a user interface, including text and/or image data. Based on the input, a manufacturer-specific or general generative model is selected and provided with a text-based prompt derived from the conceptual product description. The model generates one or more images of synthetic products, each conforming to manufacturing constraints. The images are presented to the user, each optionally associated with pricing and delivery information, and represent products that are manufacturable and acquirable via the interface. The system may incorporate manufacturability analysis, quality assurance, and model training using user feedback and product data. The invention further encompasses non-transitory storage media and systems implementing these methods for on-demand product visualization and acquisition.
1 . A computer-implemented method for generating images of synthetic products that are manufacturable as physical products, the method comprising:
receiving user input including a description of a conceptual product entered at a user interface,
wherein the user input includes at least one of text-based input or an image input;
analyzing the user input using a machine learning (ML) model to extract product attributes including at least one of product category, physical characteristics, or manufacturability constraints;
selecting, from a set of manufacturer models, a particular manufacturer model configured to generate images of synthetic products as output, wherein the selection comprises:
evaluating the extracted product attributes against metadata and learned feature vectors of each manufacturer model;
determining manufacturer suitability for the conceptual product by applying a classification algorithm trained on manufacturer-specific datasets including images and descriptions of products produced by the manufacturer, user feedback, and published product specifications; and
selecting the particular manufacturer model whose training data and manufacturability constraints, encoded in a latent space of each manufacturer model, most closely matches the extracted product attributes from among the set of manufacturer models,
wherein each manufacturer model comprises a generative AI architecture trained on a dataset comprising real and synthetic product images, manufacturer-specific production capabilities, and material constraints;
generating, based on the description of the conceptual product, a text-based prompt configured as input to direct the output of the particular manufacturer model;
generating, by using the text-based prompt with the particular manufacturer model, one or more images of one or more synthetic products representing the conceptual product; and
causing the user interface to present the one or more images of the one or more synthetic products representing the conceptual product,
wherein the one or more synthetic products depicted in the one or more images are manufacturable by a particular manufacturer of the particular manufacturer model into one or more physical products that are purchasable via the user interface upon presentation of the one or more images on the user interface.
2 . The method of claim 1 , wherein generating the text-based prompt further comprises:
generating a text-based description of an object in the image input; and
combining the text-based input of the user input and the text-based description of the object in the image input as a description of the conceptual product,
wherein the text-based prompt includes the description of the conceptual product.
3 . The method of claim 1 , wherein the particular manufacturer model is a first manufacturer model, and wherein selecting the first manufacturer model from among the set of manufacturer models comprises:
determining that the conceptual product belongs to a category of products;
determining that the first manufacturer model is associated with a first set of manufacturer constraints for physically manufacturing the category of products; and
determining that a second manufacturer model is associated with a second set of manufacturer constraints for physically manufacturing the category of products, the second set of manufacturing constraints being more restrictive than the first set of manufacturing constraints.
4 . The method of claim 1 , wherein generating the one or more images of the one or more synthetic products representing the conceptual product comprises:
determining one or more manufacturing constraints associated with the particular manufacturer model; and
generating the one or more images of the one or more synthetic products to conform with the one or more manufacturing constraints,
wherein the manufacturing constraints include constraints on equipment, material, or cost.
5 . The method of claim 1 , wherein generating the one or more images of the one or more synthetic products representing the conceptual product comprises:
determining that the conceptual product belongs to a category of products,
wherein each manufacturer model is specific to a product category such that synthetic products included in images for the particular manufacturer model are constrained by physical attributes of products that belong to the product category.
6 . The method of claim 1 , wherein generating the one or more images of the one or more synthetic products representing the conceptual product comprises:
generating multiple images of one synthetic product that represents the conceptual product.
7 . The method of claim 1 , wherein generating the one or more images of the one or more synthetic products representing the conceptual product comprises:
generating multiple images of one synthetic product that represents the conceptual product,
wherein the multiple images of the one synthetic product are generated by a manufacturer model for a particular manufacturer.
8 . The method of claim 1 , wherein generating the one or more images of the one or more synthetic products representing the conceptual product comprises:
generating multiple images of a synthetic product that represents the conceptual product,
wherein the multiple images of the synthetic product are generated by different manufacturer models trained for different manufacturers.
9 . The method of claim 1 , wherein generating the one or more images of the one or more synthetic products representing the conceptual product comprises:
generating multiple images of different synthetic products that represent the conceptual product,
wherein each of the different synthetic products is manufacturable by a manufacturer in accordance with a manufacturer model that is trained for that manufacturer.
10 . The method of claim 1 , wherein causing the user interface to present the one or more images of the one or more synthetic products representing the conceptual product comprises:
causing the user interface to present images of different synthetic products that each represent the conceptual product,
wherein each image is associated with a cost and a time parameter for completing manufacturing of the one or more physical products.
11 . The method of claim 1 , further comprising:
analyzing a manufacturability constraint associated with creation of a physical product based on a synthetic product of the conceptual product; and
incorporating an indication of the manufacturability constraint into the text-based prompt.
12 . The method of claim 1 , wherein the particular manufacturer model is trained for the particular manufacturer based on one or more data pipelines including:
user feedback,
published images of products with descriptions, and
images supplied by the particular manufacturer.
13 . The method of claim 1 , further comprising:
processing each of the one or more images with a quality assurance (QA) algorithm configured to assess metrics including:
reliability relative to the prompt,
design quality, or
image defects; and
training the set of manufacturer models based on the assessment of the metrics.