Supra2-IMG
A lightweight open-source text-to-image model released by SupraLabs, built on a Diffusion Transformer (DiT) architecture with 100 million parameters. It supports image generation at 256×256 resolution and is capable of local inference on both CPU and GPU.
Timeline (last 90 days)
The SupraLabs team has released a lightweight text-to-image model, Supra2-IMG, with 100M parameters, trained from scratch using a DiT architecture, with an output resolution of 256×256, and it is fully open-source.
The complete training process for Supra2-IMG was completed on a single H100 on the Runpod platform, taking less than 10 hours.
The sample images published for Supra2-IMG were not cherry-picked and were generated using uniform parameters: seed 0, 50 sampling steps, and CFG 3.0.
The publisher uses the term SOTA (state-of-the-art) to describe the quality performance of Supra2-IMG, but this claim should be understood in the context of the 100M parameter scale.
Supra2-IMG uses CFG 3.0 as the default display value, which is lower than the 7.0–9.0 commonly used by Stable Diffusion.
Supra2-IMG takes approximately 20 seconds to generate one image under CPU inference and approximately 2 seconds under GPU inference.
The Supra2-IMG model is hosted on Hugging Face (SupraLabs/Supra2-IMG) and is fully open for download and use.
Supra2-IMG's inference script supports customizing the prompt, random seed, CFG guidance strength, number of sampling steps, and output quantity via command-line arguments.
A resolution of 256×256 and a capacity of 100M parameters mean that Supra2-IMG is not suited for high-precision commercial image generation.
All Facts (9)
A resolution of 256×256 and a capacity of 100M parameters mean that Supra2-IMG is not suited for high-precision commercial image generation.
50%UnverifiedThe complete training process for Supra2-IMG was completed on a single H100 on the Runpod platform, taking less than 10 hours.
50%UnverifiedThe sample images published for Supra2-IMG were not cherry-picked and were generated using uniform parameters: seed 0, 50 sampling steps, and CFG 3.0.
50%UnverifiedThe publisher uses the term SOTA (state-of-the-art) to describe the quality performance of Supra2-IMG, but this claim should be understood in the context of the 100M parameter scale.
50%UnverifiedSupra2-IMG uses CFG 3.0 as the default display value, which is lower than the 7.0–9.0 commonly used by Stable Diffusion.
50%UnverifiedSupra2-IMG takes approximately 20 seconds to generate one image under CPU inference and approximately 2 seconds under GPU inference.
50%UnverifiedThe Supra2-IMG model is hosted on Hugging Face (SupraLabs/Supra2-IMG) and is fully open for download and use.
50%UnverifiedSupra2-IMG's inference script supports customizing the prompt, random seed, CFG guidance strength, number of sampling steps, and output quantity via command-line arguments.
50%UnverifiedThe SupraLabs team has released a lightweight text-to-image model, Supra2-IMG, with 100M parameters, trained from scratch using a DiT architecture, with an output resolution of 256×256, and it is fully open-source.
50%