Is Chitra Tiny state of the art?Pending benchmarks.Announced Oct 12, 2026; no independent results yet.

The details so far

Lab

Udgam Labs

Announced

Oct 12, 2026

API

Not yet

Open weights

Published · Hugging Face (udgamlabs/Chitra-Tiny, about 0.85 GB) and code on GitHub (udgamlabs/chitra-tiny)

Parameters

212M (diffusion transformer, 16 layers, width 768)

Method

Latent rectified-flow diffusion transformer with classifier-free guidance; 50 steps by default

Training

Transformer trained from scratch (random weights) on about 4.7 million public-domain and CC0 images with automatically generated captions

Frozen components

google/flan-t5-base text encoder (Apache-2.0) and stabilityai/sd-vae-ft-mse decoder (MIT), used unchanged

Output

256×256 images from English prompts; no built-in safety filter

License

Apache-2.0 (weights and code)

Tags marked unsure are not stated on the lab's own pages.

What would let us call it

Verdicts use only independent boards we already track. Two consecutive snapshots are required before any crown.

• Publicly announced with a source
• Usable by the public (API or weights)
• Listed on an independent board we track
• Enough results for a rank in a category
• Holds across two snapshots (crowns only)

What the lab claims

Self-reported. Shown for context; not used in any verdict.

Lab claim: GenEval overall 0.219 with the official code, against 0.426 for Stable Diffusion 1.5 in the same harness.

Lab claim: FID 22.1 on 5,000 held-out images like its training data and CLIP score 26.2 (real held-out images: 27.5); Udgam notes both favour the model.

Lab claim: about three seconds per image on an Apple-silicon Mac.

Sources