romanri85/ltx2-19b-lora

LTX-2 19B LoRA

LoRA adapter trained on the LTX-2 19B model for personalized video generation.

Model Details

Parameter	Value
Base model	Lightricks/LTX-2 (ltx-2-19b-dev.safetensors)
Text encoder	google/gemma-3-12b-it-qat-q4_0-unquantized
LoRA rank	32
LoRA alpha	32
Target modules	to_k, to_q, to_v, to_out.0
Trigger word	t3es9s5a1

Training Configuration

Parameter	Value
Training steps	2000
Learning rate	1e-4
Optimizer	AdamW
Batch size	1
Mixed precision	bf16
Resolution	960x544x1
Checkpoint interval	500 steps
Scheduler	linear
Max grad norm	1.0
First frame conditioning	0.5

Checkpoints

File	Step
lora_weights_step_00500.safetensors	500
lora_weights_step_01000.safetensors	1000
lora_weights_step_01500.safetensors	1500
lora_weights_step_02000.safetensors	2000

Usage

Include the trigger word t3es9s5a1 in your prompt to activate the LoRA.

WaveSpeed API

curl -X POST "https://api.wavespeed.ai/api/v3/wavespeed-ai/ltx-2-19b/image-to-video-lora" \
  -H "Authorization: Bearer YOUR_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "image": "https://example.com/image.jpg",
    "prompt": "t3es9s5a1 walking through a park, cinematic lighting",
    "loras": [{"path": "https://huggingface.co/romanri85/ltx2-19b-lora/resolve/main/lora_weights_step_02000.safetensors", "scale": 1.0}],
    "duration": 5
  }'

Trainer

Trained using the official Lightricks/LTX-2 trainer (packages/ltx-trainer).