Instructions to use gregt/T2_IC_SDR2HDR_LTX_LoRA with libraries, inference providers, notebooks, and local apps. Follow these links to get started.
- Libraries
- Diffusers
How to use gregt/T2_IC_SDR2HDR_LTX_LoRA with Diffusers:
pip install -U diffusers transformers accelerate
import torch from diffusers import DiffusionPipeline # switch to "mps" for apple devices pipe = DiffusionPipeline.from_pretrained("Lightricks/LTX-Video", dtype=torch.bfloat16, device_map="cuda") pipe.load_lora_weights("gregt/T2_IC_SDR2HDR_LTX_LoRA") prompt = "Astronaut in a jungle, cold color palette, muted colors, detailed, 8k" image = pipe(prompt).images[0] - Notebooks
- Google Colab
- Kaggle
- Local Apps Settings
- Draw Things
Commit History
-
Add model card
Verified signature5f703d3
-
Upload T2_IC_SDR2HDR_LTX_LoRA.safetensors
Verified signaturefa83420 -
Delete T2_SDR--->HDR_LTX_LoRA.safetensors
Verified signature20978cc -
Upload T2_SDR--->HDR_LTX_LoRA.safetensors
Verified signatureModel Description LTX HDR LoRA (LumiVid-inspired) A lightweight IC-LoRA for Lightricks LTX-2.3 that enables true high dynamic range video generation and SDR → HDR conversion. Trained using logarithmic encoding (LogC3) techniques from the official LumiVid research paper, this LoRA adapts the base LTX model to output scene-linear HDR with significantly improved highlight roll-off, shadow recovery, and specular detail while maintaining strong temporal consistency. Key Features • True HDR output – Produces clean 16-bit scene-linear results (ideal for EXR sequences) • Excellent SDR → HDR up-conversion – Recovers blown-out highlights and crushed shadows from standard LTX generations • Filmic highlight behavior – Natural, delicate roll-off on bright light sources, lens flares, and specular reflections (sun, water, chrome, etc.) • Strong shadow detail – Recoverable information in deep shadows without excessive noise • Lightweight – Only ~<1% of parameters, fast inference, works well at st…b668d1c -
Delete lora_weights_step_07000.safetensors
Verified signature142cf42 -
Upload lora_weights_step_07000.safetensors
Verified signature3efb42b -
Update README.md
Verified signatureThis lora is based on the LumiVid paper LTX released. It was trained on 5200 paired video files form the Polyhaven HDR 4K latlongs. It will take an SDR video or still image and convert it to an HDR video/ still with real highlights in the otherwise clipped highlights. Use a standard VHS saver node and set it to Prores for a 4444 12bit file. You can use the Comfy EXR saver nodes to get a 16bit float, though I haven't tested that yet but have been told it works.0ec88e3 -
initial commit
Verified signature34cc268