sm079/supra2-img-web

Supra2-Img Web model files

Model files for Supra2-Img Web, which runs SupraLabs' Supra2-IMG text-to-image in the browser on WebGPU. They are unofficial, modified repackagings of the original models, made with tools/build_models.py:

file	contents	source	license
supra2-img-ema.safetensors	the Supra2-IMG diffusion transformer (104M parameters, float32), converted from the pickled checkpoint to safetensors; adds the checkpoint's stored unconditional text embedding as uncond_text (unmasked row only)	SupraLabs/Supra2-IMG model_final_ema.pt @ b22ffe6c	Apache-2.0
flan-t5-base-encoder.safetensors	the Flan-T5-Base encoder only (decoder tensors removed), float32, unchanged values	google/flan-t5-base model.safetensors @ 7bcac572	Apache-2.0
sd-vae-ft-mse-decoder.safetensors	the VAE decoder and post_quant_conv only (encoder removed), float32, unchanged values	stabilityai/sd-vae-ft-mse @ 31f26fde	MIT
tokenizer.json, tokenizer_config.json	the Flan-T5 tokenizer, unchanged	google/flan-t5-base @ 7bcac572	Apache-2.0

The weights keep their original licenses: Supra2-IMG and Flan-T5-Base are licensed under the Apache License 2.0 (copy in LICENSE-Apache-2.0.txt); sd-vae-ft-mse is licensed under the MIT license by Stability AI. All credit for the models goes to their authors: SupraLabs, Google and Stability AI. This repository is not affiliated with them.