> For the complete documentation index, see [llms.txt](https://docs.quantaria.xyz/llms.txt). Markdown versions of documentation pages are available by appending `.md` to page URLs; this page is available as [Markdown](https://docs.quantaria.xyz/deep-dives/diffusion-on-patches.md). # Diffusion on image patches Diffusion models generate images by learning to reverse a noising process. This page explains the idea and how it applies to patches inside a shared transformer. ## The forward process Start with an image and add noise in small steps until it is pure noise. This process is fixed and needs no learning. ## The reverse process The model learns to predict the noise that was added, or equivalently to remove it. Starting from random noise, repeated denoising steps produce an image. ## Patches The image is split into patches, and each patch is treated as a token-like input. The noise is added to the patches, and the loss is computed on the predicted noise for each patch. This is what lets the same transformer handle image patches alongside text tokens. ## Output size The number of patches sets the image resolution. Hilbert-1 Nano outputs 32×32 images and Base outputs 256×256. Larger images need more patches and more compute per step. ## Evaluation Image quality is commonly measured with FID, which compares generated and real image statistics. Lower is better. Quantaria plans to report FID for each model. See [Evaluating against matched dense baselines](/deep-dives/evaluation-methodology.md). ## Quantaria's use Diffusion loss on patches is part of the Transfusion-style objective. The number of sampling steps and the noise schedule are not published in these docs. ## Related * [Transfusion: text and images in one model](/deep-dives/transfusion-objective.md) * [Born machines](/deep-dives/born-machines.md): a different route to sampling images --- # Agent Instructions This documentation is published with GitBook. GitBook is the documentation platform designed so that both humans and AI agents can read, navigate, and reason over technical content effectively. Learn more at gitbook.com. ## Querying This Documentation If you need additional information that is not directly available in this page, you can query the documentation by asking a question. Perform an HTTP GET request on the following URL with the `ask` and `goal` query parameters: ``` GET https://docs.quantaria.xyz/deep-dives/diffusion-on-patches.md?ask=&goal= ``` `ask` is the immediate question: it should be specific, self-contained, and written in natural language. `goal` is what the user is ultimately trying to achieve, the reason they need the answer. Sharing it helps GitBook give you a better, more relevant answer. A goal is most helpful when it describes the outcome the user wants rather than restating the question. For example, with `ask=how do I create an API token`, a goal like `automate deployments from our CI pipeline` lets GitBook tailor the answer to that use case. The response will contain a direct answer to the question and relevant excerpts and sources from the documentation. Use this mechanism when the answer is not explicitly present in the current page, you need clarification or additional context, or you want to retrieve related documentation sections.