Instructions to use Alissonerdx/LTX-LoRAs with libraries, inference providers, notebooks, and local apps. Follow these links to get started.
- Libraries
- LTX.io
How to use Alissonerdx/LTX-LoRAs with LTX.io:
# Install the LTX-2 pipelines git clone https://github.com/Lightricks/LTX-2.git cd LTX-2 uv sync --frozen
# Download the weights from this repo, plus the Gemma text encoder hf download Alissonerdx/LTX-LoRAs --local-dir models/LTX-LoRAs hf download google/gemma-3-12b-it-qat-q4_0-unquantized --local-dir models/gemma-3-12b
# Text/image-to-video with the LoRA on the HQ two-stage base pipeline uv run python -m ltx_pipelines.ti2vid_two_stages_hq \ --checkpoint-path path/to/checkpoint.safetensors \ --distilled-lora path/to/distilled_lora.safetensors 0.8 \ --spatial-upsampler-path path/to/spatial_upsampler.safetensors \ --gemma-root models/gemma-3-12b \ --lora models/LTX-LoRAs/<weights>.safetensors 1.0 \ --prompt "your prompt here" \ --output-path output.mp4 # For image-to-video, add: --image path/to/image.jpg 0 0.8 - Notebooks
- Google Colab
- Kaggle
nice work all results are same
give me a simple workflow of that which works for you, I have cloud gpu, I will download again everything as it is! and yes bro I tried like 2 lora's different ones for R2V masked one still same results, I got tired and gave up.
Okay, I found the reason. Because there were several frames in the mask that were not detected, the LTX model directly regenerated almost the same content as the original video based on the characters that were not detected in the mask
The mask must cover 100% of the character throughout the entire video. If even one frame of the character is visible, there will be a problem because the model is lazy and will pick the easiest information for it, which would be the frame where the character is uncovered.

