Unofficial implementation of HelloWorld (arXiv 2608.05070) on LTX-2.3. Code: GitHub · LoRA: Hugging Face
⭐ NEW — Generalization wall (v2): 10 brand-new charactersBuilt from scratch after training — fresh stills, fresh warps, unseen prompts. Base vs v2@2500. 40 pairs — base vs trained, side by sideSame seed, prompt, warp, press window. Only the LoRA differs. The 40 bench charactersFirst frames + what each was directed to do. The model learning to read camera directionLeft: warp ghost input. Right: generated. Compare step_000000 vs step_001250.