---
title: Agent-GenPath
emoji: 🔬
colorFrom: blue
colorTo: green
sdk: gradio
sdk_version: 6.31.0
app_file: app.py
short_description: Pathology patch synthesis with the Agent-GenPath DiT
python_version: "3.12"
startup_duration_timeout: 30m
---

# Agent-GenPath — Pathology Patch Synthesis

Demo for **"Agent-Driven Generative Priors for Pathology Image Synthesis and Segmentation"**
([arXiv 2611.06701](https://arxiv.org/abs/2611.06701)).

Samples 256×256 synthetic H&E pathology patches from the paper's pretrained unconditional
flow-matching Diffusion Transformer ([`yscript/agent-genpath`](https://huggingface.co/yscript/agent-genpath)),
using the reference Euler sampler from the official
[Agent-GenPath repo](https://github.com/script-Yang/Agent-GenPath) (Apache-2.0).

The backbone is trained on **TCGA-200K**, a 217,030-patch dataset the paper's pathology agent
system curated from the public TCGA repository; the same backbone transfers to semantic
segmentation with a lightweight head (PathSeg-380K).

The model definition (`dit.py`) is copied verbatim from the official repository.