Prone Face Cam

⚠️ Work in Progress (For testing only)

• I believe in open development. DO NOT expect the best result from this project.

• All images in the gallery are raw, unprocessed outputs directly from generation.

• The last 4 images in the gallery are I2V.

• Each image includes its attached workflow for full reproducibility.

• I know the lora is huge. Rank 16 results were not great. Any tips for lowering the size would be great!

• Any feedback is welcome.

Training Details

• Trainer: Directly from the LTX team. https://github.com/Lightricks/LTX-2

• Steps: 2,250

• Dataset: 12 videos

• Clip length: ~5 seconds

• Frame rate: 25 FPS

• Resolution buckets: 1024x576 - 121frames and 576x1024 - 121frames

  • Frames are required to divisible by 8+1

• Gemma3 abliterated used during training.

• NO audio training was done during this release.

Workflow & Settings

• Base workflows: ComfyUI default templates for LTX-2 (T2V & I2V)

• Tested on 19b-dev full and 19b-dev-FP8

• Sampler: Res2s

• Sampling steps: 20

• Additional LoRAs: ltx-2-ic-detailer-lora

• Gemma3 abliterated used during generation.