Diffusion Transformer (DiT)

Tencent Hunyuan-DiT image detector: diffusion transformer forensics

Forensic examination of Tencent Hunyuan-DiT fine-grained image generation, Chinese/English bilingual text rendering, and transformer token spaces.

Verify media authenticity with Sealed Rose

Free · 3 scans a day · Deleted after your scan

Verify photo deepfake

Tencent Hunyuan-DiT is an open-source diffusion transformer model tailored for bilingual text-to-image synthesis. Scammers deploy Hunyuan-DiT to generate convincing identity documents, synthetic marketing collateral, and photorealistic social media personas. Sealed Rose examines Hunyuan-DiT’s transformer patch boundaries, high-frequency spatial noise residuals, and subsurface skin scattering.

Signs to look for

1DiT spatial patch grid boundaries

Microscopic inspection reveals subtle 2x2 or 4x4 latent patch boundary discontinuities in uniform textures like skies or walls.

2Bilingual typography kerning artifacts

Rendered text characters show subtle stroke irregularities and anti-aliasing inconsistencies along glyph edges.

3Iris and pupil reflection discrepancies

Corneal light highlights often reflect inconsistent light source vectors when examined under forensic zoom.

Tencent Hunyuan-DiT forensics methodology

  • Diffusion Transformer Latent Analysis: Identifies DiT cross-attention weight signatures and token-space boundary alignments.
  • Frequency Domain Residuals: Checks 2D FFT power spectrums for characteristic checkerboard patterns from upsampling layers.
  • Photometric Gradient Consistency: Validates directional shadow gradients against computed primary scene light vectors.

How to protect yourself and take action

Step 1

Upload image to Sealed Rose forensic analyzer

Run full-spectrum neural verification to detect Hunyuan-DiT transformer artifacts and synthetic diffusion markers.

Frequently asked questions

What is Tencent Hunyuan-DiT?▼

Hunyuan-DiT is an open-source image generation model built on a diffusion transformer architecture, featuring advanced multi-language text rendering.

Can Sealed Rose distinguish Hunyuan-DiT images from real photographs?▼

Yes. Even though the images look photorealistic, their underlying pixel probability distribution and transformer patch boundaries differ measurably from optical camera sensors.