Krea 2 workflows I actually use every day, tuned for 8 GB GPUs. Pick a version in the tabs above: every file loads and runs out of the box, with a free LoRA slot ready for yours.
The collection keeps evolving with it, so expect regular updates.
Description
Flow: generate (pass 1) -> latent 1.5x -> pass 2 HD -> color match -> save HD
-> USD 2.5x (group ②) -> color match -> save USD
Toggle each group on/off with the bars above the groups.
Made for photorealism: landscapes, wallpapers, street photos, amateur shots.
DesireX also knows Pokemon much better than base Krea2.
Settings:
- Pass 1: euler / beta, steps 0-8 of 8, CFG 1
- Pass 2: latent upscale 1.5x (bislerp), euler / simple, steps 6-10, CFG 1
(starting late keeps the details natural, no HDR / over-textured look)
- Color match (mkl 0.5) keeps the colors of pass 1
- USD: 2.5x with 4x Remacri, 3 steps, denoise 0.12, tiles auto-sized from the image (2x2), color match mkl 0.7
- Default size: 2:3 portrait, 1.0 MP. I recommend 1.5 MP (change it in "Resolution")
- Negative prompt is zeroed out (CFG 1, it has no effect)
LoRA slots: 2 Krea2 LoRA loaders, off by default (Ctrl+B to switch on).
The files inside are examples, put your own Krea2 LoRAs.
Speed: ~190s per image at 1.5 MP on RTX 4070 8GB
Models:
- BSSDesirexKrea2_x3BF16INT8NVFP4_int8.safetensors (UNET) -> (lien de la page DesireX)
- qwen3vl_4b_fp8_scaled.safetensors (text encoder, type krea2): https://huggingface.co/Comfy-Org/Krea-2/resolve/main/text_encoders/qwen3vl_4b_fp8_scaled.safetensors
- qwen_image_vae.safetensors (VAE): https://huggingface.co/Comfy-Org/Krea-2/resolve/main/vae/qwen_image_vae.safetensors
- 4x_foolhardy_Remacri.pth (upscaler)
Custom nodes:
- rgthree-comfy: https://github.com/rgthree/rgthree-comfy
- KJNodes: https://github.com/kijai/ComfyUI-KJNodes
- ComfyUI Essentials: https://github.com/cubiq/ComfyUI_essentials
- ComfyUI_UltimateSDUpscale: https://github.com/ssitu/ComfyUI_UltimateSDUpscal
