Lumina-Next-SFT

Lumina-Next-SFT is a 2B Next-DiT model with Gemma-2B serving as the text encoder, enhanced through high-quality supervised fine-tuning (SFT).

Demo current model: Lumina-Next-SFT 1k Resolution

Lumina-Next-T2I enables zero-shot resolution extrapolation to 2k.

Lumina-Next supports higher-order solvers ["euler", "midpoint"].

It can generate images with merely 10 steps without any distillation for 1K resolution generation.

Tip: For improved human portrait generation, please choose resolution at 1024x2048.

To reduce waiting times, we are offering three parallel demos:

Lumina-T2I 2B model: [demo (supported 2k inference)] [demo (supported 2k inference)] [demo (supported 2k inference)] [demo (compositional generation)]