> 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/transfusion-objective.md).

# Transfusion: text and images in one model

Most multimodal systems use separate models for text and images, joined by an interface. The Transfusion-style objective trains a single transformer on both at once. This page explains how the training works.

## The two losses

* **Text** is trained with next-token prediction. The model learns to predict each token from the ones before it.
* **Images** are split into patches and trained with a diffusion loss. The model learns to denoise patches.

Both losses are applied to the same network in one training run, so the shared weights learn both.

## Why one backbone

When text and images share weights, the model can use structure learned from one modality in the other. The shared layers also reduce the number of separate systems to build and serve.

The method comes from [arXiv:2408.11039](https://arxiv.org/abs/2408.11039). Quantaria's version uses shared tensor-network cores, so the same low-rank structure carries both losses.

## Practical consequences

* Image quality and text quality compete for capacity in the shared layers. Balancing the two losses is a training decision.
* Image output is limited by patch size and resolution. Nano outputs 32×32 and Base outputs 256×256.

## Quantaria's use

The objective is a live part of Hilbert-1, as described in [Shared Multimodal Backbone](https://github.com/bhavy2908/dataai_docs/tree/main/architecture/multimodal-backbone.md#unified-objective). Training data and loss weights for each model are not published in these docs.

## Related

* [Diffusion on image patches](/deep-dives/diffusion-on-patches.md)


---

# 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/transfusion-objective.md?ask=<question>&goal=<user_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.