neil1145/ShadowForge-models

ShadowForge Models

Pre-packaged model weights for ShadowForge — an AI horror content generation pipeline.

All models are selected, quantized, and tested for 16GB VRAM GPUs (RTX 4060 Ti 16GB, RTX 5060 Ti, etc.).

Models Included

Image Generation (Flux.1 Dev)

File	Size	Description
diffusion_models/flux1-dev-Q5_0.gguf	~7.8 GB	Flux.1 Dev UNET, GGUF Q5 quantized
text_encoders/clip_l.safetensors	~235 MB	CLIP-L text encoder
text_encoders/t5xxl_fp8_e4m3fn.safetensors	~4.6 GB	T5-XXL text encoder, fp8
vae/ae.safetensors	~320 MB	Flux autoencoder

Video Generation (Wan 2.1 I2V)

File	Size	Description
diffusion_models/wan2.1_i2v_480p_14B_fp8_e4m3fn.safetensors	~16 GB	Wan 2.1 I2V 14B, fp8
text_encoders/umt5_xxl_fp8_e4m3fn_scaled.safetensors	~6.3 GB	UMT5-XXL encoder, fp8
clip_vision/clip_vision_h.safetensors	~1.2 GB	CLIP Vision H encoder
vae/wan_2.1_vae.safetensors	~243 MB	Wan 2.1 video VAE

Total: ~37 GB

Quick Download

# Clone ShadowForge and download all models
git clone https://github.com/neil1145/ShadowForge.git
cd ShadowForge
pip install -r requirements.txt
python scripts/download_models.py --models-dir /path/to/comfyui/models

Or download individual files:

pip install huggingface_hub
huggingface-cli download neil1145/ShadowForge-models --local-dir /path/to/comfyui/models

Directory Structure

Files should be placed in your ComfyUI models directory:

models/
├── clip_vision/
│   └── clip_vision_h.safetensors
├── diffusion_models/
│   ├── flux1-dev-Q5_0.gguf
│   └── wan2.1_i2v_480p_14B_fp8_e4m3fn.safetensors
├── text_encoders/
│   ├── clip_l.safetensors
│   ├── t5xxl_fp8_e4m3fn.safetensors
│   └── umt5_xxl_fp8_e4m3fn_scaled.safetensors
└── vae/
    ├── ae.safetensors
    └── wan_2.1_vae.safetensors

VRAM Requirements

These models are designed to run one at a time on 16GB VRAM:

• Flux image generation: ~10 GB peak VRAM
• Wan video generation: ~14 GB peak VRAM

ShadowForge's VRAMManager handles automatic model swapping — it evicts the current model before loading the next one.

License

Individual models retain their original licenses:

• Flux.1 Dev: FLUX.1-dev Non-Commercial License
• Wan 2.1: Apache 2.0
• CLIP/T5 encoders: Various open licenses

This repo is a convenience bundle. Check individual model pages for license details.