SkyReels V1 | Human-Focused Video Creation

ComfyUI SkyReels V1 Workflow Description#

1. What is SkyReels V1 Workflow?#

SkyReels V1 is an advanced human-centric video generation workflow for ComfyUI, designed to create realistic human movements and expressions from text prompts or images. Built on the SkyReels-V1-Hunyuan-I2V model, which has been fine-tuned on over 10 million high-quality film and television clips, SkyReels generates Hollywood-caliber human videos with authentic facial expressions and natural motion at high resolution (544×960).

2. Benefits of SkyReels V1#

• SkyReels Cinema-quality output: Produces professional-grade videos with realistic human motion and high production value
• SkyReels Human expression support: Supports 33 distinct facial expressions for emotionally realistic characters
• Natural movement: SkyReels includes 400+ true-to-life human movements
• High resolution: SkyReels generates videos at 544×960 resolution with smooth 24fps playback
• Optimized performance: SkyReels features advanced compilation options for faster rendering
• Easy customization: SkyReels is fully compatible with visual workflows and adjustable parameters

3. How to Use SkyReels V1#

3.1 Generation with SkyReels#

Image-to-Video with Text Prompt#

• Inputs: Reference image and text prompt
• Best for: Generating SkyReels human-centric videos with lifelike expressions and movement
• Characteristics:
  • SkyReels realistic human motion and emotional expressions
  • Cinematic composition and professional framing
  • Stable SkyReels video generation with minimal artifacts

Generation Example#

1. Prepare SkyReels inputs:
   • In Load Image, upload your base image
   • In CLIPTextEncode, enter a prompt describing desired motions
2. SkyReels Refinement (Optional):
   • In EmptyHunyuanLatentVideo, set desired dimensions and frames (default: 544×544, 49 frames)
   • In FluxGuidance, adjust guidance values for stronger prompt adherence
3. Generate SkyReels output:
   • Click Queue Prompt to run the workflow
   • In Video Combine, check the preview output (also saved in ComfyUI > Output folder)

3.2 Parameter Reference for SkyReels V1#

• EmptyHunyuanLatentVideo:
  • length: [49] - Defines frame count
  • batch_size: [1] - Sets batch generation size
• SamplerCustom (First stage):
  • noise_seed: [15] - Controls randomness of output
  • cfg: [7.0] - Controls prompt adherence strength
• SamplerCustom (Second stage):
  • add_noise: [false] - No additional noise added during refinement

4. SkyReels Tips#

• Result variability: SkyReels can generate different results per seed. If unsatisfied, try changing the seed
• SamplerCustom settings: Run the workflow with default parameters (15 steps for the first sampler, 0 steps for the refining sampler)
• Negative prompts: Use negative prompts (e.g., "blurry, bad quality video, chaotic movement") to refine output quality

More Information#

For the latest updates and details on SkyReels V1, visit SkyReels V1 on Hugging Face.

Acknowledgments#

SkyReels V1 was developed by the SkyReels-AI team, based on the Hunyuan base model. The ComfyUI integration for SkyReels, created by *Kijai*, enables seamless workflow compatibility by converting the SkyReels model from diffusers format into HunyuanVideo format. Full credit goes to the original SkyReels developers for pioneering human-centric video generation.