Instructions to use MrReclusive/LTX-2.3-FP4 with libraries, inference providers, notebooks, and local apps. Follow these links to get started.
- Libraries
- Diffusers
How to use MrReclusive/LTX-2.3-FP4 with Diffusers:
pip install -U diffusers transformers accelerate
import torch from diffusers import DiffusionPipeline from diffusers.utils import load_image, export_to_video # switch to "mps" for apple devices pipe = DiffusionPipeline.from_pretrained("MrReclusive/LTX-2.3-FP4", dtype=torch.bfloat16, device_map="cuda") pipe.to("cuda") prompt = "A man with short gray hair plays a red electric guitar." image = load_image( "https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/diffusers/guitar-man.png" ) output = pipe(image=image, prompt=prompt).frames[0] export_to_video(output, "output.mp4") - Notebooks
- Google Colab
- Kaggle
New version LTX-2.3-22B-distilled-1.1
+1. This also works faster than both gguf's and fp8 with amd gpu's using rocm (with bitsandbytes) It is definately faster but the existing distill fpme is just a bit below what I get from the kijai's safetensors version.
This is a hidden gem! It is faster on a 3060 too (12GB VRAM).
I thought I would never touch FP4 (for 50XX), and here's a improved version!
just wanted to drop a note, been busy, bouncing in and out of other projects, i am building a new fp4me on distilled v1.1 tonight, so ill be getting that out.
i also do have one built on sulphur (the adult ltx2.3) that ill get posted, but i may adjust it, my problem with it is its too adult by default, it really doesn't like to do normal things, so ill try and get a half sulphur going.
i know i don't get on here much, this is hobby for me and i have other things that pay the bills that i need to focus on.
+1. This also works faster than both gguf's and fp8 with amd gpu's using rocm (with bitsandbytes) It is definately faster but the existing distill fpme is just a bit below what I get from the kijai's safetensors version.
and yes that's by design, not trying to compete with the kijai's or official fp4's in terms of quality, im trying to make them smaller at an acceptable quality loss, and using a good denoise clears up most the quality problems.