tags:
- text-to-video
- ltx-video
- lora
- realism
- skin-texture
- hair-rendering
base_model: Lightricks/LTX-Video
LTX-2.3-skin-hair
This LoRA is designed to refine the texture and physical properties of skin and hair within the LTX-Video framework. It focuses on reducing "plastic" skin artifacts and improving the specular highlights and movement of hair strands.
Training Progression Comparison
The video below demonstrates the training impact across different intervals.
Sequence: Base Model (No LoRA) $\rightarrow$ 1,000 steps $\rightarrow$ 2,000 steps $\rightarrow$ 3,000 steps $\rightarrow$ 4,000 steps.
Model Details
- Developer: o-8-o
- Architecture: LTX-Video (Transformer-based)
- Checkpoint Sizes: 353 MB (.safetensors)
- Training focus: High-fidelity human textures.
Usage Instructions
ComfyUI
- Download your preferred step checkpoint.
- Place it in your
models/loras folder.
- Use the LoraLoader node and connect it between your Model and CLIP inputs.
- Recommended strength: 0.75 - 0.9 for best results without over-sharpening.
Trigger Words
While this is a fine-tune of the latent space, using descriptive prompts helps guide the LoRA:
extreme close up, pores, detailed skin, individual hair strands, photorealistic.
