Simple LTX Video Generator
No idea? Try leaving the prompt empty.
🎞️ LTX-2.3 LoRA Library
- Transition: Designed for smooth morphing or scene changes.
- HardCut: Generates convincing cinematic hard-cut transitions within a single video. Recommended strength: 0.2–1.0.
- VBR: Helps the model follow complex prompt instructions and spatial logic.
- BetterPhysics: Simulates physical collisions, impact, and friction.
- Msr: Maintains consistency across multiple subjects in reference-to-video workflows.
- DemoPose: Holds a specific character pose throughout the generation.
- Singularity: Comprehensive I2V optimizer—improves anatomy stability, facial expressions, and style compatibility. Weight: 0.8–1.0.
- Water: Rivers, surf, rain, waterfalls, floods, splashes, spray, and wet-surface specularities — while keeping the subject's identity.
- Omni: Enhances subject fidelity, intricate details, and stylized rendering for unique character or object generations.
- JoyAI: Generate highly coherent, synchronized, multi-shot video and audio stories up to 5 minutes long from structured text prompts.
⚡ LTX-2.3 Distilled LoRA Library
Tip: If your prompt contains complex instructions (e.g., "a blue cat sitting on a red chair eating a fish") and the model is omitting elements, the LoRA may be overfitting. Switching to a lower rank or CondSafe variant lets the base model focus on your prompt again.
CondSafe (Ceil 32 / 52 / 72): Hard-capped ranks with conditioning safety. Prevents color burning and blow-ups at higher CFG scales.
Dynamic Avg Rank (105 / 111): Smart dynamic rank allocation. Light on VRAM while keeping details sharp.
Energy Optimized (Rank 72 / 96): Balances power efficiency and speed without major quality loss.
Fro90 Ceil (36 / 72): Frobenius-regularized weights with dynamic ceilings. Keeps motion fluid, but sensitive to sampler tuning.
High-Rank 384 (v1.0 / v1.1): Uncompressed weights for maximum fidelity. Best for high-VRAM setups; use cautiously with fine-tuned base models.