Promptwaffle WF - Krea2 Turbo Uncensored Production Pipeline v3

A streamlined ComfyUI workflow for Krea 2 Turbo with LoRA stacking, automatic trigger words, resolution presets, reproducible seeds, and embedded generation metadata.

Choose Your Checkpoint

Uncensored Version

For unrestricted local generation, use:

My Krea2 Unlocked INT8 ConvRot

The workflow is configured for:

myKrea2UnlockedInt8_v10.safetensors

Place it in:

ComfyUI/models/diffusion_models/

Official Krea2 Version

Those who do not need the unlocked checkpoint can use the official ComfyUI Krea2 Turbo model:

Official Krea2 Turbo FP8 Scaled

Select the downloaded checkpoint in the Krea2 Diffusion Model loader.

Required Components

Text Encoder

qwen3vl_4b_fp8_scaled.safetensors

Place in:

ComfyUI/models/text_encoders/

VAE

qwen_image_vae.safetensors

Place in:

ComfyUI/models/vae/

The CLIP Loader type must be set to:

krea2

Required Custom Nodes

ComfyUI-Lora-Manager

https://github.com/willmiao/ComfyUI-Lora-Manager

Provides LoRA stacking, trigger-word controls, prompt encoding, metadata saving, and debugging.

Comfyui-Resolution-Master

https://github.com/Azornes/Comfyui-Resolution-Master

Provides resolution and aspect-ratio presets.

Workflow Features

• Separate style and character LoRA stacks

• Automatic LoRA trigger-word injection

• Trigger-word toggles

• Optional inline LoRA loading

• Resolution Master presets

• Shared seed for image and wildcard reproduction

• Final resolved-prompt preview

• Prompt, seed, model, LoRA, and workflow metadata embedded in saved PNGs

• Drag generated PNGs back into ComfyUI to restore the workflow

Recommended Settings

• Steps: 8

• CFG: 1.0

• Sampler: Euler

• Scheduler: Simple

• Denoise: 1.0

Krea2 Turbo is distilled, so increasing CFG above 1.0 usually makes results worse.

There is intentionally no traditional negative prompt. The workflow uses Conditioning Zero Out, which is appropriate for CFG 1.0.

Prompting

Edit the node labeled:

User Prompt — EDIT THIS

Krea2 works best with natural-language prompts describing the subject, action, environment, lighting, camera angle, mood, and important details.

Avoid older tag-heavy prompting styles such as masterpiece, score tags, or excessive weighting syntax.

LoRA Strengths

Useful starting values:

• Style LoRA: 0.75–0.9

• Character LoRA: 0.6–0.8

• Secondary stacked LoRA: lower it if the LoRAs begin competing

The included LoRAs are examples and can be replaced with your own Krea2-compatible LoRAs.

Resolution

Good starting resolutions include:

• 1024 × 1024

• 1216 × 832

• 832 × 1216

• 1328 × 1328

Generate near the model’s native resolution and upscale afterward for larger final images.

Built by Promptwaffle for fast Krea2 LoRA testing, repeatable generations, and everyday production use.