⚡ Check out my Free Patreon , with extra stuff
Laufzeitprofil
Quellenbeschreibung
⚡ Check out my Free Patreon , with extra stuff
Edit : kroma-v0.1.safetensors no longer exists; you can change it for AmateurSlider-KREA2_v1.safetensors in the workflow (Load Lora Node)
Hello! I'll be brief:
Before asking any questions, please read all this! This workflow will first use Krea2 Turbo with Qwen3 VL Abliterated as the text-encoder to create your stuff! If you don't know what that means, let's say that it's an uncensored version of Qwen3 VL 🔥! Your prompt will go to a Conditioner that will help Qwen3 to go VROOM VROOM . When the first KSampler finishes your image, it will go to VAE Utils to get a 2x mushroom 🍄! Since the upscale will have a few deformations, it will be re-processed with a second KSampler and saved 💾! The Kroma LoRA is optional; as a long-time Chroma user, I love it! ❤️
Quellen
1 QuelleQuellenauszüge
1 AuszugSource context: 1957 downloads · Type Workflows · Base model Krea 2
Designed and Tested with:
Intel Ultra 7, RTX 5060 Ti 16GB, 32 GB DDR5, ComfyUI (Portable) 0.29.0, and lots of caffeine ☕
BTW, it takes me about 73 seconds to generate a 2048 x 2048px image with this workflow!
Download links and locations:
Krea 2 Turbo Model
📂 models/ └── 📂 diffusion_models/ └── └krea2_turbo_fp8_scaled
Huihui Qwen3 VL 4B Instruct Abliterated
📂 models/ └── 📂 text-encoder/ └──└ Huihui-Qwen3-VL-4B-Instruct...safetensors
Wan2.1 VAE and Upscale 2x VAE
📂 models/ └── 📂 vae/ └──└ wan_2.1_vae.safetensors └── └Wan2.1_VAE_upscale2x_image...safetensors
Custom nodes:
ComfyUI Krea2 Conditioning
ComfyUI VAE Utils
Notes:
If the output is too glossy, you can either lower the Kroma LoRA to 0.6 and try going up, or you can totally disable it . Edit : Change it for Amateur Slider .
About the Upscaler (if it's taking too much time)
You can also disable the last KSampler and VAE Utils if you don't need the Upscaler, but then it's a regular Krea 2 workflow. If you need more help understanding the Krea 2 Conditioning Node , you can ask an LLM; or give them the JSON code; they totally know how to make adjustments.
Enjoy generating your cool images, or whatever you generate... pervert! 👀
Remaining hardware or download-size estimates are not available yet.
Anforderungen
Text encoder · 4.88 GB · Hugging Face · ahmed22xa/Huihui-Qwen3-VL-4B-Instruct-abliterated-comfy
krea2_turbo_fp8_scaled.safetensors
Datei ungeprüftCheckpoint · Hugging Face · Comfy-Org/Krea-2
LoRA · Civitai · 2773343
VAE · Hugging Face · Comfy-Org/Wan_2.2_ComfyUI_Repackaged
VAE · 484 MB · Hugging Face · spacepxl/Wan2.1-VAE-upscale2x
Knotenpaket · Registry