RIGen-Arena

Arena AR checkpoint with two selectable trained refiner variants.

Files

• weights/arena_ar_model.safetensors: Arena AR checkpoint.
• weights/causal_refiner_lora.safetensors: causal refiner LoRA.
• weights/bidirectional_refiner_lora.safetensors: bidirectional refiner LoRA.
• configs/causal_refiner_bundle.json: causal refiner configuration.
• configs/bidirectional_refiner_bundle.json: bidirectional refiner configuration.
• configs/arena_ar_config.yaml: Arena AR configuration.
• configs/arena_foundation_config.yaml: refiner foundation configuration.

Both refiner variants use the original LTX-2.3 foundation model with the selected trained LoRA. The LTX-2.3 backbone, official distilled LoRA, spatial upsampler, and text encoder are external dependencies and are not included here.

The refiner inference modes are causal and bidirectional.

License

The model is distributed under the LTX-2.x Community License Agreement in LICENSE.