Instructions to use MaxedOut/ComfyUI-Starter-Packs with libraries, inference providers, notebooks, and local apps. Follow these links to get started.
- Libraries
- llama-cpp-python
How to use MaxedOut/ComfyUI-Starter-Packs with llama-cpp-python:
# !pip install llama-cpp-python from llama_cpp import Llama llm = Llama.from_pretrained( repo_id="MaxedOut/ComfyUI-Starter-Packs", filename="Flux1/clip/GGUF/t5xxl_Q5_K_M.gguf", )
output = llm( "Once upon a time,", max_tokens=512, echo=True ) print(output)
- Notebooks
- Google Colab
- Kaggle
- Local Apps Settings
- llama.cpp
How to use MaxedOut/ComfyUI-Starter-Packs with llama.cpp:
Install from brew
brew install llama.cpp # Start a local OpenAI-compatible server with a web UI: llama-server -hf MaxedOut/ComfyUI-Starter-Packs:Q4_K_M # Run inference directly in the terminal: llama-cli -hf MaxedOut/ComfyUI-Starter-Packs:Q4_K_M
Install from WinGet (Windows)
winget install llama.cpp # Start a local OpenAI-compatible server with a web UI: llama-server -hf MaxedOut/ComfyUI-Starter-Packs:Q4_K_M # Run inference directly in the terminal: llama-cli -hf MaxedOut/ComfyUI-Starter-Packs:Q4_K_M
Use pre-built binary
# Download pre-built binary from: # https://github.com/ggerganov/llama.cpp/releases # Start a local OpenAI-compatible server with a web UI: ./llama-server -hf MaxedOut/ComfyUI-Starter-Packs:Q4_K_M # Run inference directly in the terminal: ./llama-cli -hf MaxedOut/ComfyUI-Starter-Packs:Q4_K_M
Build from source code
git clone https://github.com/ggerganov/llama.cpp.git cd llama.cpp cmake -B build cmake --build build -j --target llama-server llama-cli # Start a local OpenAI-compatible server with a web UI: ./build/bin/llama-server -hf MaxedOut/ComfyUI-Starter-Packs:Q4_K_M # Run inference directly in the terminal: ./build/bin/llama-cli -hf MaxedOut/ComfyUI-Starter-Packs:Q4_K_M
Use Docker
docker model run hf.co/MaxedOut/ComfyUI-Starter-Packs:Q4_K_M
- LM Studio
- Jan
- Ollama
How to use MaxedOut/ComfyUI-Starter-Packs with Ollama:
ollama run hf.co/MaxedOut/ComfyUI-Starter-Packs:Q4_K_M
- Unsloth Studio
How to use MaxedOut/ComfyUI-Starter-Packs with Unsloth Studio:
Install Unsloth Studio (macOS, Linux, WSL)
curl -fsSL https://unsloth.ai/install.sh | sh # Run unsloth studio unsloth studio -H 0.0.0.0 -p 8888 # Then open http://localhost:8888 in your browser # Search for MaxedOut/ComfyUI-Starter-Packs to start chatting
Install Unsloth Studio (Windows)
irm https://unsloth.ai/install.ps1 | iex # Run unsloth studio unsloth studio -H 0.0.0.0 -p 8888 # Then open http://localhost:8888 in your browser # Search for MaxedOut/ComfyUI-Starter-Packs to start chatting
Using HuggingFace Spaces for Unsloth
# No setup required # Open https://huggingface.co/spaces/unsloth/studio in your browser # Search for MaxedOut/ComfyUI-Starter-Packs to start chatting
- Docker Model Runner
How to use MaxedOut/ComfyUI-Starter-Packs with Docker Model Runner:
docker model run hf.co/MaxedOut/ComfyUI-Starter-Packs:Q4_K_M
- Lemonade
How to use MaxedOut/ComfyUI-Starter-Packs with Lemonade:
Pull the model
# Download Lemonade from https://lemonade-server.ai/ lemonade pull MaxedOut/ComfyUI-Starter-Packs:Q4_K_M
Run and chat with the model
lemonade run user.ComfyUI-Starter-Packs-Q4_K_M
List all available models
lemonade list
Create README.md
Browse filesπ§ Best Models for ComfyUI
Welcome to the ultimate curated ComfyUI model hub β an all-in-one repository containing the most significant and useful models for real-world workflows. This isnβt just a grab bag of everything floating around online; itβs a distilled collection of only the most impactful, tested, and workflow-proven models across Flux1, SDXL, and beyond.
This repo covers Flux1 (Dev, Schnell, Fill, Depth, Canny, Redux, and more), GGUF variants for low-VRAM systems, T5XXL + CLIP text encoders, and a growing handpicked list of SDXL tools, controlnets, and top-performing models from Civitai.
π Not Everything, Just the Best
We intentionally avoid flooding this repo with every model in existence. Instead, this collection:
Focuses on essential, practical models youβll actually use
Prioritizes multiple VRAM tiers (e.g., Q3, Q5, FP8)
Skips Q8 GGUF unless justified (too big for low-spec systems, too slow for high-end ones)
Excludes LoRAs unless they offer real benefit over full models
π Want One-Click Simplicity?
If you're looking for an even easier way to use these models, check out my Patreon. I create:
β
One-click installers that detect your VRAM and download the right files
βοΈ Plug-and-play workflows built for ComfyUI (from beginner to advanced)
π§ͺ Deep-dive guides, model breakdowns, and private toolkits
No fluff. Just clean, well-organized systems designed to actually work. Perfect for both noobs and ComfyUI veterans.
π¨ Visual Examples (Coming Soon)
We'll be adding concise example blocks so you can see exactly what each model type does. Expect:
Flux1 Core Models
Dev: High-quality output (slower)
Schnell: Lower-quality but fast generation
Flux1 Tools
Depth: Real image β depth map β new image
Canny: Real image β edge map β new image
Fill (Inpaint): Original β masked β regenerated
These will be organized into clean XY grids so you're not scrolling through an image tsunami.
SDXL Models (Preview)
Best realism & anime models from Civitai
Refiner-enabled and ControlNet-ready options
π Folder Structure
Best-Models-For-ComfyUI/
Flux1/
ββ unet/
β ββ Dev/
β β ββ flux1-dev-fp8.safetensors
β β ββ GGUF/
β β β ββ flux1-dev-Q3_K_S.gguf
β β β ββ flux1-dev-Q5_K_S.gguf
β β β ββ flux1-dev-Q6_K_S.gguf
β ββ Schnell/
β β ββ flux1-schnell-fp8-e4m3fn.safetensors
β β ββ GGUF/
β β β ββ flux1-schnell-Q3_K_S.gguf
β β β ββ flux1-schnell-Q5_K_S.gguf
β β β ββ flux1-schnell-Q6_K_S.gguf
β ββ Canny/
β β ββ flux1-canny-dev-fp8.safetensors
β β ββ GGUF/
β β β ββ flux1-canny-dev-fp16-Q4_0-GGUF.gguf
β β β ββ flux1-canny-dev-fp16-Q5_0-GGUF.gguf
β ββ Depth/
β β ββ flux1-depth-dev-fp8.safetensors
β β ββ GGUF/
β β β ββ flux1-depth-dev-fp16-Q4_0-GGUF.gguf
β β β ββ flux1-depth-dev-fp16-Q5_0-GGUF.gguf
ββ clip/
β ββ t5xxl_fp8_e4m3fn.safetensors
β ββ t5xxl_fp8_e4m3fn_scaled.safetensors
β ββ t5xxl_fp16.safetensors
β ββ clip_l.safetensors
ββ loras/
β ββ flux1-canny-lora.safetensors *(optional)*
β ββ flux1-depth-lora.safetensors *(optional)*
β¨ Final Notes
This is a living repo. New models and categories will be added over time.
Stay comfy. Stay maxed out.
β Max