Skip to content
latestai.eulatestai.eu
SI

Supra2 Img

by SupraLabs

Try itHugging Face

Tiny 100M-parameter open-weights text-to-image diffusion transformer from SupraLabs, generating 256×256 images.

Available in EUImage GenerationModel

Updated EU status checked Last shown

Supra2 Img at a glance

What is Supra2 Img?
Supra2 Img is a model from SupraLabs. Tiny 100M-parameter open-weights text-to-image diffusion transformer from SupraLabs, generating 256×256 images.
Who makes Supra2 Img?
Supra2 Img is made by SupraLabs.
Does Supra2 Img work in the EU?
Available in the EU. Weights are downloadable from Hugging Face under Apache-2.0 with no regional restrictions, so they can be run locally in the EU/EEA. Last verified 14 Nov 2026.
How much does Supra2 Img cost?
Open source. Free to download and run locally; no hosted service listed. Licence: Apache-2.0.
Where was Supra2 Img shown?
In 1 creator video by AI Search. Most recently by AI Search on 27 Sept 2026, at 36:53 in “Waifus incoming, GPT 6 Sol, Grok 4.7, Opus 5.5, Mimo 2.6, Step 5, OpenMuse: AI NEWS”.

Sources: the vendor's own pages (linked above) and creator videos at the linked timestamps. EU status is information, not legal advice; see how we check.

About

Supra2-IMG is a small text-to-image model from SupraLabs, built as a diffusion transformer (SupraDiT) with about 104M parameters. It was trained from scratch on the FLUX-Reason-6M dataset (5.6M images, 10 epochs) on a single H100 GPU in roughly 9 hours. It uses a frozen Flan-T5-Base text encoder (128-token context) and the SD-VAE-FT-MSE autoencoder, and outputs 256×256 images from 32×32 latents.

The model card reports FID 27.9 and CLIP score 29.4. A YouTube reviewer (AI Search) describes the checkpoint as about 417 MB, small enough to run on a phone and roughly 60 times smaller than Z Image, but notes image quality is low compared with larger models.

It is distributed as downloadable weights on Hugging Face under Apache-2.0 and run locally with a provided inference.py script (recommended: Euler flow sampling, 50 steps, CFG 3.0). It targets hobbyists and researchers interested in very small, low-resource image generators. Limits include the low resolution, English-oriented prompts and the custom code requirement; the card's "SOTA quality" claim refers to its size class only.

Key features
  • ~104M-parameter diffusion transformer (SupraDiT)
  • 256×256 output with Euler flow sampling
  • Frozen Flan-T5-Base text encoder and SD-VAE-FT-MSE
  • Trained from scratch on FLUX-Reason-6M (5.6M images)
  • Apache-2.0 weights; local inference script
  • Small checkpoint (~417 MB) suited to low-resource devices
1 moment

As seen on YouTube

Watch the moment: the exact second top creators show it. Videos load only when you press play, via youtube-nocookie.com.

Keep exploring

Similar AI tools

Closest neighbours by meaning, not just by category.

I1

i1

Model · Princeton

1

Fully open 3B text-to-image diffusion model from Princeton with released weights, training code, and data pipeline.

Available in EUImage Generation
Release date n/aOpen source
L2

L2P

Model

1

Pixel-space text-to-image diffusion model transferred from latent diffusion models, without a VAE, supporting 4K output.

Available in EUImage Generation
May 2026Open source
I3

Iris 3B

Model · Sperid Labs

1

Open 3B-parameter text-to-image flow-matching model that generates directly in pixel space, without a VAE.

Available in EUImage Generation
Oct 2026Open source
SV

Sana Video 2

Model · Nvidia

1

NVIDIA's 5B and 14B hybrid linear-attention video diffusion transformers generating up to 720p on a single GPU.

UnverifiedVideo Generation
Jul 2026
MH

MiniMax H3 Image

Model · MiniMax

3

Open-weights MiniMax image generation and editing model producing up to ~16 MP images in 6 sampling steps.

Available in EUImage Generation
3 Nov 2026Open source
ZI

Z-Image

Model

2

Open-weights 6B text-to-image model family from Alibaba's Tongyi-MAI lab, with a fast 8-step Turbo variant.

Available in EUImage Generation
26 Nov 2025Open source