Master Krea 2 Workflows in ComfyUI: Text-to-Image, Image-to-Image & Builder Scenes
This guide walks you through the Krea 2 image generation workflow inside ComfyUI – covering text-to-image, image-to-image, and the Ideogram-style builder for creating complex compositions. You’ll also learn how to apply LoRAs, enhancers, and upscalers to produce commercial-quality images up to 4K.
What You'll Learn
Prerequisites
Before starting, ensure you have:
- A working ComfyUI installation (see our ComfyUI Installation Guide)
- Sufficient VRAM (16 GB recommended for the Turbo BF16 model; 8–12 GB users should use the MXFB8 model)
- The Krea 2 Turbo model, Qwen 3 VL 4B FB8 scaled CLIP, and Qwen image VAE – all links are provided in the video description
- The following custom nodes installed via ComfyUI Manager:
- Conditioning Krea 2 Rebalance (for bypassing filters)
- RGB Smart Seed Variants (for more diverse outputs)
- ComfyUI Krea 2 T Enhancer (for subtle variations)
- Krea 2 Enhancer (separate workflow for higher detail)
Commercial Licensing
Krea 2 has a very permissive license. All images can be used for commercial purposes without issue if your business earns under $1 million annually. Above that threshold, you must purchase a separate license. This makes it ideal for client work and product marketing.
If your hardware isn’t powerful enough or you want to speed up generations, consider using RunPod’s cloud GPU service. You can get a $5 credit to start generating up to 9,000 images and 300 videos.
Special Offer - $5 Credit Included!
When you sign up for RunPod using our affiliate link, you'll receive a $5 credit that can be used to generate up to 9,000 images and 300 videos. This gives you plenty of resources to explore ComfyUI and AI image/video generation without any upfront cost!
Step-by-Step Process
Step 1: Download and Install the Models & Custom Nodes
- Download the Krea 2 Turbo model (BF16 if you have 16+ GB VRAM, otherwise MXFB8)
- Download the CLIP – Qwen 3 VL 4B FB8 scaled
- Download the VAE – Qwen image VAE
- Place all model files in the appropriate ComfyUI
models/subfolders (checkpoints incheckpoints/, CLIP inclip/, VAE invae/) - Install missing custom nodes:
- Go to ComfyUI Manager → Custom Nodes Manager → Install Missing Custom Nodes
- Restart ComfyUI after installation
- If some nodes still show red, download them manually from their GitHub repositories (e.g.,
Conditioning Krea 2 Rebalance,RGB Smart Seed Variants)
Pro Tip
The first generation will take longer (about 60 seconds on 16 GB VRAM). Subsequent generations average 30 seconds per image.
Model Compatibility
If you have 8–12 GB VRAM, must use the MXFB8 version of the Turbo model. Using the BF16 model on lower VRAM will cause out‑of‑memory errors.
Step 2: Understand the Workflow Switches
The template includes three switches on the top‑left corner:
- Text to Image – generate from scratch
- Image to Image – transform an existing image
- Builder to Image – use the Ideogram‑style canvas builder
Switch Logic
- For Text to Image only: enable just that switch.
- For Image to Image: enable that switch (and optionally disable text‑to‑image).
- For Builder to Image: enable both Builder to Image and Text to Image simultaneously – the builder uses the aspect ratio from the text‑to‑image resolution dropdown.
You can also choose a resolution from the dropdown: square, landscape, portrait, and various cinematic views.
Step 3: Configure Sampler, CFG, and Steps
The Turbo model uses only 8 steps. Set the following defaults:
- CFG: 1 (leave at 1)
- Sampler:
- For complex scenes: RS Multi-Steps or DRSD
- For portraits/simple generations: Euler Simple (most tested)
- Alternatives: Euler Ancestral, ERSD (slight variations)
Pro Tip
If you want more variations of the same prompt, enable the RGB Smart Seed Variants node. Krea 2 tends to produce very similar outputs; this node forces more diversity.
Step 4: Generate Images with Text-to-Image
- Enable the Text to Image switch
- Enter your positive prompt (e.g., a detailed portrait description)
- Enter a negative prompt – always include one to improve precision
- Press Run (the queue button)
Example prompt used in the video:
Close‑up natural beauty portrait of a South Korean girl, warm dark eyes, soft window light from left, plain warm neutral background, realistic skin texture, visible pores, tiny imperfections, natural lips, minimal makeup, relaxed expression, detailed eyelashes, subtle catchlights, 85mm portrait lens, shallow depth of field, editorial but honest photography, realistic color, no airbrushed skin, no doll‑like face.
Negative prompt: a paragraph listing everything you don’t want.
First Generation
You’ll get a 1440×2560 image with impressive realism. The model handles skin texture, lighting, and prompt adherence remarkably well.
Step 5: Apply LoRAs and Enhancers
The workflow includes a LoRA loader section. You can add any LoRA (e.g., the Krea 2 Enhancer LoRA) at a strength of 0.70.
- Without LoRA: the image is clean but slightly flat
- With Enhancer LoRA: adds shadows, skin pores, more contour – more realistic
Additionally, there is a Krea 2 T Enhancer node (separate group) that you can enable with a value from 0.05 to 2.0. This adds subtle variations (moles, wrinkles, etc.) and also helps bypass content filters.
Censorship Bypass
Use the Conditioning Krea 2 Rebalance node to bypass filters. Set a value above 1 (e.g., 2, 3, 4) to make the model follow your prompt more accurately for unconventional content. The Krea 2 T Enhancer node also acts as a bypass tool.
Step 6: Image-to-Image with Masking
- Enable the Image to Image switch; disable Text to Image
- Drag and drop an image into the image drop box
- Right‑click on the image → Open in Mask Editor
- Brush over the area you want to change (e.g., text on a sneaker)
- Save the mask
- Write a prompt describing what to replace (e.g., “Remove only the visible text… reconstruct the original sneaker…”)
- Set the denoising strength (in the sampler) to 0.50 (range 0.10–0.80). Lower values preserve more of the original image
- Press Run
Pro Tip
For best results when removing text or small elements, use a denoising strength between 0.50 and 0.70. If you set it to 1, it will regenerate the entire image.
Alternative Method
If you need to preserve the image exactly and only remove elements, consider using Canvas or Photoshop outside ComfyUI, or use Qwen Edit Image (a separate node).
Step 7: Builder-to-Image – Creating Advertisements
- Enable both Builder to Image and Text to Image switches
- Choose a square resolution (e.g., 2000×2000) from the dropdown – the builder will use these dimensions
- Enter your main prompt (e.g., “premium futuristic advertising campaign for an ultra light performance shoe”)
- Add a further description (e.g., “clean minimalist studio environment with soft white, silver, and pale gray tones…”)
- Use the builder canvas:
- Click and drag to create boxes on the canvas
- Double‑click a box to edit its prompt
- For text boxes, you can specify exact text to render (e.g., “Unleash your pace”) and a style description
- You can also paste a JSON generated by ChatGPT (ask it to create a builder JSON for Krea 2)
- Press Run – the builder will generate the composition with text and images
Using ChatGPT for JSON
You can ask ChatGPT to generate a JSON file for the builder. Copy the JSON, switch import mode to Always, paste it into the builder, and run. The builder will populate the canvas with elements.
Pro Tip
If the text hallucinates (e.g., unintended text appears), you can easily edit it in post‑production using Photoshop, Canva, or Qwen Edit Image.
Step 8: Upscale to 4K+
The workflow includes a dedicated Krea 2 Upscaler section. To use it:
- Drag and drop an image into the upscaler input
- Leave all parameters as default
- Press Run
- The upscaler will produce a 4K+ version with sharper details and improved text legibility
Before vs After
The upscaled image shows significantly higher definition, with crisper text and finer textures. It’s ideal for final output.
Bypass Unused Groups
When you are not using the enhancer or upscaler, bypass those groups (right‑click → Bypass) to avoid unnecessary processing.
Related Guides
- ComfyUI Installation Guide – Complete setup of ComfyUI, Manager, and custom nodes
- Running ComfyUI on RunPod – Cloud GPU setup for faster generations
- Deploy Your First RunPod – Basic RunPod account and pod deployment
Next Steps
Now that you have mastered the Krea 2 workflow, you can:
- Experiment with different LoRAs (e.g., style, character, or detail enhancers)
- Try more complex builder layouts using ChatGPT‑generated JSON
- Combine the Krea 2 Enhancer with the upscaler for maximum quality
- Explore further editing techniques using Qwen Edit Image or other in‑paint nodes
If you have questions about this template, leave a comment on the video. Subscribe for more ComfyUI workflows and advanced techniques.