Run UNCENSORED Image-to-Video Locally: LOW VRAM GPUs SUPPORTED | Hunyuan SkyReels Tutorial

Date Updated: 26 February 2025

Open source AI models have been rapidly improving, following the success of Flux Dev and Hunyuan’s video capabilities. SkyReels, a Hunyuan-based image-to-video model, now delivers quality that rivals major platforms like Sora, Veo, and Kling. In this guide, we’ll walk through installing and running SkyReels locally on Windows across low, medium, and high-end consumer GPUs.

Prerequisites

Before installation, you'll need the following tools:

1 . Microsoft Visual Studio Installer

https://visualstudio.microsoft.com/downloads/

Add to Environment Variables

Verify installation by typing “CL” in Command Prompt (should display Microsoft C++ message)

2 . CUDA Toolkit

https://developer.nvidia.com/cuda-toolkit-archive

3 . Python 3.12

https://www.python.org/downloads/release/python-3128/

4 . Git

https://git-scm.com/downloads

Installation Steps

1 – Download the batch script from our website

2 – Place the file in the location where you want to install ComfyUI

3 – Run the batch file

– It will clone ComfyUI from the comfyanonymous repository
– The script will scan for Python versions
– Select the option for Python 3.12

4 – Wait for ComfyUI dependencies installation

– Press enter when prompted with “press any key to continue”

5 – Select Triton version:

– Triton 3.2 works with PyTorch 2.6+ (recommended for torch.compile improvements)
– Triton 3.1 works with PyTorch 2.4-2.5
– PyTorch 2.3 and older versions are not supported

6 – Select option 1 for the latest Triton 3.2.0
7 – Wait for SageAttention installation to complete
8 – Press enter to continue and complete the installation

Setting Up ComfyUI

1 – Run ComfyUI by clicking on the “run ComfyUI” batch file
2 – Load the image-to-video workflow
3 – Install the 3 missing custom nodes from ComfyUI manager

– Always select the latest versions
– Wait for each installation to complete
– Restart ComfyUI when done

4 – Refresh the browser to confirm all nodes are working correctly

Required Models

1 . Kijai VAE

https://huggingface.co/Kijai/HunyuanVideo_comfy/tree/main

2 . SkyReels Model

https://huggingface.co/Kijai/SkyReels-V1-Hunyuan_comfy/tree/main

3 . SkyReels Lora

https://huggingface.co/spacepxl/skyreels-i2v-smooth-lora/tree/main

Performance Notes

– RTX 3090 (24GB VRAM): ~20 minutes for 3-4 seconds of video
– RTX 4090: ~10 minutes
– Lower GPUs with quantized models: 20+ minutes
– First run requires LLM download (a few minutes depending on internet speed)
– Initial iteration speed may be high (e.g., 119 seconds) but will decrease (to 40-50 seconds)
– To reduce generation time, lower steps from 30 to 25 or 20 (affects quality)

Video Workflow & Installer

Image-to-Video Skyreels

FREE DOWNLOAD

Send download link to:

Master AI Influencer with Our Expert Course!

Learn to create stunning AI Influencer with ComfyUI. This beginner-friendly course will guide you through every step—from basics to advanced techniques.