One Krea 2 workflow for every download. A switch picks .safetensors or .gguf, so it also runs on AMD and small cards.
The loader you do not pick never opens, so it costs no memory.
Watch the video: https://www.youtube.com/watch?v=rst_h25FAXM
What you need
- Stable Yogi Nodes (SY Nodes) in ComfyUI/custom_nodes (loaders, seed, switches, save)
- ComfyUI 0.42.1 or newer
- A Krea 2 model, .safetensors or .gguf, in models/diffusion_models (Muse by Stable Yogi Krea2 is free here on CivitAI)
- For .gguf only: the small gguf Python package (python -m pip install gguf, with ComfyUI's python)
- qwen3vl_4b_fp8_scaled.safetensors in models/text_encoders (Krea 2 licence)
- qwen_image_vae.safetensors in models/vae (not the Wan VAE: that saves coloured static)
How to use it
- Look at the end of your model file: .safetensors or .gguf.
- Set the MODEL FILE switch: 1 for .safetensors, 2 for .gguf, and pick your model in that box.
- INPUT switch: 1 = text to image, 2 = start from your own picture.
- It opens on 16 steps, CFG 2 (Extended model). Turbo: 8 steps, CFG 1. Stock Krea 2 Base: 30 steps, CFG 3.
- Small card? Q8_0 GGUF (about 13.6 GB) is the safe pick, Q4_0 (7.8 GB) and Q2_K (5 GB) are smaller.
The step-by-step guide, the video and more free workflows are on my portal.

