AlayaWorld: Interactive Long-Horizon World Modeling - Full Technical Report (v1.1)
Paper • 2608.13492 • Published
How to use AlayaLab/AlayaWorld-v1.1-stage2b with Diffusers:
pip install -U diffusers transformers accelerate
import torch
from diffusers import DiffusionPipeline
from diffusers.utils import load_image, export_to_video
# switch to "mps" for apple devices
pipe = DiffusionPipeline.from_pretrained("AlayaLab/AlayaWorld-v1.1-stage2b", dtype=torch.bfloat16, device_map="cuda")
pipe.to("cuda")
prompt = "A man with short gray hair plays a red electric guitar."
image = load_image(
"https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/diffusers/guitar-man.png"
)
output = pipe(image=image, prompt=prompt).frames[0]
export_to_video(output, "output.mp4")Autoregressive SFT teacher of AlayaWorld (30-step inference / stage3 distillation base). Fine-tuned from LTX-2.3-22B; released under the LTX-2 Community License (academic & non-commercial; commercial use by entities with ≥ $10M annual revenue requires a license from Lightricks).
| File | Size | Content |
|---|---|---|
transformer.pt |
26.2 GB | full fine-tuned DiT (stage2b, ViGeo spatial memory) |
history_encoder.pt |
33 MB | compressed frame-history encoder |
Usage — point the inference/training configs of AlayaLab/AlayaWorld at this checkpoint:
paths:
resume_checkpoint: <this repo download dir> # transformer.pt
history_encoder: <this repo download dir>/history_encoder.pt
Run: VALIDATE_ONLY=1 CONFIG_PATH=configs/infer_i2v_camera_ar.yaml bash scripts/finetune/train.sh
See the repository README for the full pipeline (training stage0–3, three inference paths). Reports: intro · full · v1.1