🛠️ Manual Download – Flux Level 1, 2, & Mega

This guide covers everything, whether you're a free user here for Level 1 & 2, or a Power User / Keyholder setting up Mega manually.

⚠️ Mega users: You get a One‑Click Installer that auto-detects RAM & VRAM and downloads exactly what you need. No manual setup needed.
But if you want to do it manually, this guide walks you through every required file.

How this works:

• Free users → Go through everything up to the "⚡ Only for Mega Workflow" section.

• Mega users → Everything, plus the Mega-only section.
  Mega includes Level 1 & 2, so don’t skip ahead.

📥 Core Files

💾 Right-click any file link → choose "Save Link As..." → Save into the mentioned folder. All folders are in ComfyUI/models/. Workflow Level 2 needs everything from Level 1.

🎯 Always Download:

1️⃣ Bootcamp (Level 1):

• navi_flux_v1.safetensors ➡️ loras (Test out my avatar Lora 🙂)

• RealESRGAN_x2plus.pth ➡️ upscale_models

• 4x-UltraSharp.pth ➡️ upscale_models

• ae.safetensors ➡️ vae

2️⃣ Advanced (Level 2):

• sam_vit_b_01ec64.pth ➡️ sams

• face_yolov8m.pt ➡️ ultralytics/bbox

• hand_yolov8s.pt ➡️ ultralytics/bbox

📎 CLIP Models

💾 Save all into models/clip

🎯 Always Download:

• clip_l.safetensors

🖱️ Choose One Tier:

🔶 Tier A (32GB+ RAM, not VRAM)

• t5xxl_fp16.safetensors

🔶 Tier B (16GB RAM, not VRAM)

• t5xxl_fp8_scaled.safetensors

🔶 Tier C (less than 16GB RAM, not VRAM)

• t5xxl_Q5_K_M.gguf

🧠 Diffusion Models (Choose One Tier)

💾 Save these files into models/diffusion_models.

🔶 Tier A (16GB+ VRAM – fp8)

1️⃣ Bootcamp (Level 1):

• flux1-dev-fp8.safetensors

2️⃣ Advanced (Level 2):

• flux1-fill-dev-fp8.safetensors

🔶 Tier B (12 GB VRAM – Q5 GGUF)

1️⃣ Bootcamp (Level 1):

• flux1-dev-Q5_K_S.gguf

2️⃣ Advanced (Level 2):

• flux1-fill-dev-Q5_K_S.gguf

🔶 Tier C (Less than 12GB VRAM – Q3/Q4 GGUF)

1️⃣ Bootcamp (Level 1):

• flux1-dev-Q3_K_S.gguf

2️⃣ Advanced (Level 2):

• flux1-fill-dev-Q3_K_S.gguf

⚡ Only for Mega Workflow (Level 3)

🎯 Always Download:

• sigclip_vision_patch14_384.safetensors ➡️ clip_vision

• flux_shakker_labs_union_pro-fp8.safetensors ➡️ controlnet

• codeformer.pth ➡️ facerestore_models

• GFPGANv1.4.pth ➡️ facerestore_models

• pulid_flux_v0.9.1.safetensors ➡️ pulid

• flux1-redux-dev.safetensors ➡️ style_models

🖱️ Choose One Tier:

🔶 Tier A (16GB+ VRAM – fp8)

• flux1-canny-dev-fp8.safetensors ➡️ diffusion_models

• flux1-depth-dev-fp8.safetensors ➡️ diffusion_models

🔶 Tier B (12 GB VRAM – Q5 GGUF)

• flux1-canny-dev-Q5_0.gguf ➡️ diffusion_models

• flux1-depth-dev-Q5_0.gguf ➡️ diffusion_models

🔶 Tier C (Less than 12GB VRAM – Q3/Q4 GGUF)

• flux1-canny-dev-Q4_0.gguf ➡️ diffusion_models

• flux1-depth-dev-Q4_0.gguf ➡️ diffusion_models

✅ Done!

📦 Extra Dependencies (Mega Only)

Some Mega-only nodes (Reactor, Pulid, etc.) need extra libraries.

• Open your ComfyUI folder (where .venv is located).

• Right-click in the folder → Open in Terminal.

• Run these commands in order:

1. Activate venv:

.venv\Scripts\activate

2. Install facexlib:

python -m pip install facexlib

3. Install insightface (Windows wheel):

python -m pip install "https://huggingface.co/MaxedOut/ComfyUI-Starter-Packs/resolve/main/Misc/insightface-0.7.3-cp312-cp312-win_amd64.whl"

🔷 How to Use My LoRA (Avatar Test)

LoRA: navi_flux_v1.safetensors
Recommended Strength: 1.0
Trigger Words:

• na'vi (apostrophe is crucial)

• blue skin

• male or female (each image was tagged with male or female. Use one.)

Example: na'vi female riding a bike, blue skin

LoRA guide coming soon, you ask? Perhaps. 🙂