Instructions to use TenStrip/LTX2.3-10Eros with libraries, inference providers, notebooks, and local apps. Follow these links to get started.
- Libraries
- Diffusers
How to use TenStrip/LTX2.3-10Eros 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("TenStrip/LTX2.3-10Eros", 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
Is Eros capable of video2video?
I've been using Eros with the RuneXX workflows and no matter what combination of settings I use, video2video workflows (e.g. extend) always come out blurry, shiny, and with bad ghosting.
I don't see any reason Eros would be incapable of v2v, but I thought I'd check here to make sure.
Extremely capable. It will extend seamlessly. Instead of an image use a very normal i2v workflow with InPlace conditioning at max strength and upload a video loader that processes into image sequence at 24 fps, usually the VHS video loader; settings are 24 fps rate forced-frame cap of 49,73,96,121, etc (24+1 format) and then set your generation length so that the rest will be generated from the video. It will generated audio on the source and then extend it with prompt. If you have a low quality video don't condition the upcale pass and Eros will reskin the entire video which can be used to fix low quality looks or change the style of the video.
this was done with 73 start frames from an extremely low quality Grok gif, extremely compressed, but the upscale was unconditioned so it changed the style to Eros overall: https://civitai.red/images/129551783