FLUX.2 Dev GGUF Workflow for ComfyUI, tested on RTX 3060 (12GB)
Main Diffusion Model (GGUF) Model: FLUX.2-dev-gguf Download:
https://huggingface.co/city96/FLUX.2-dev-gguf
Put it here: ComfyUI/models/diffusion_models/
Note: Choose the quantization that matches your GPU VRAM:
Q2_K → ~13 GB file — 4GB VRAM (slow, ~3 hours on laptop)
Q3_K_M → ~16 GB file — 6–8 GB VRAM
Q4_K_M → ~20 GB file — 8–10 GB VRAM
Q5_K_M → ~24 GB file — 12 GB VRAM (recommended for RTX 3060)
Q8 → ~38 GB file — 16 GB+ VRAM (best quality)VAE flux2-vae.safetensors Download: https://huggingface.co/Comfy-Org/flux2-dev/resolve/main/split_files/vae/flux2-vae.safetensors
Put it here: ComfyUI/models/vae/
Text Encoder mistral_3_small_flux2_fp8.safetensors Download: https://huggingface.co/Comfy-Org/flux2-dev/resolve/main/split_files/text_encoders/mistral_3_small_flux2_fp8.safetensors
Put it here: ComfyUI/models/text_encoders/
Note: Use the fp8 version to save VRAM. Use bf16 if you have headroom and want slightly better quality.
Required Custom Nodes Install via ComfyUI Manager or clone manually into ComfyUI/custom_nodes/
ComfyUI-GGUF https://github.com/city96/ComfyUI-GGUF
Description
v2.0.1: Notes only, graph unchanged (same run graph as v2.0.0). New "How to use" note: what it does, the inputs, settings, tips, attention launch flags, tested-on line. "Models and .whl files" note renamed "Downloads" and made to match downloads.md (old Q8_0 / bf16 file names, SageAttention .whl steps removed).
