Krea 2 Turbo - Style Reference
ComfyUI in the Cloud
Deploy Krea 2 Turbo - Style Reference, from
$0.60 / hour
Krea-2 Turbo Style Reference ComfyUI Workflow
A ComfyUI image generation workflow powered by Krea-2 Turbo, featuring INT8-quantized diffusion models, Qwen3-VL conditioning, and a Style Reference LoRA configuration for visual style customization.
Workflow Overview
This workflow utilizes the Krea-2 Turbo model for AI image generation within ComfyUI. It combines an INT8-quantized diffusion model, Qwen3-VL 4B conditioning, a dedicated Qwen Image VAE, and a Style Reference LoRA configuration.
The Style Reference LoRA provides an additional model configuration intended for visual style control. The selected LoRA is configured with a strength of 1.00, allowing the workflow to define the influence of the LoRA when it is applied during image generation.
The workflow also includes LLM-based prompt enhancement settings. Prompt enhancement and LLM thinking mode are disabled in the current configuration. The maximum LLM token value is set to 512.
Generation Settings
Prompt Enhancement
Disabled
LLM Thinking Mode
Disabled
LLM Max Tokens
512
LoRA Strength
1.00
Width
External Input
Height
External Input
Diffusion Precision
INT8
Conditioning Precision
FP8 Scaled
Model Configuration
Image Generation Model (UNet)
INT8krea2_turbo_int8_convrot
The Krea-2 Turbo model serves as the primary image generation component responsible for the denoising process. Its INT8-quantized configuration helps reduce memory requirements compared with higher-precision model weights. The diffusion model uses conditioning information to generate images based on the supplied instructions and configured generation parameters.
Text / Vision Conditioning Model
Qwen3-VL 4Bqwen3vl_4b_fp8_scaled
The Qwen3-VL 4B model provides text and vision conditioning for the image generation process. It helps interpret input instructions and visual information, producing conditioning representations that guide the diffusion model. The FP8-scaled configuration helps reduce model memory requirements compared with higher-precision implementations.
Variational Autoencoder (VAE)
VAEqwen_image_vae
The Qwen Image VAE converts generated latent representations into visible images. During the decoding stage, the VAE transforms latent outputs into images that can be previewed, saved, or processed further within ComfyUI. This component provides the final decoding stage for the image generation process.
Style Reference LoRA
Strength 1.00krea2_style_reference
The Style Reference LoRA is selected as an additional model component for visual style customization. LoRA models provide a way to modify the behavior of a diffusion model without replacing its primary model weights. The configured strength of 1.00 represents the selected LoRA influence when the component is applied. Its actual effect depends on the workflow connections and any reference-conditioning inputs used.
Style Reference Configuration
The workflow includes a Style Reference LoRA configuration intended for generating images with additional visual style guidance. This component provides a mechanism for customizing the appearance of generated images while retaining the Krea-2 Turbo diffusion model.
| LoRA Model | krea2_style_reference |
| LoRA Strength | 1.00 |
| LoRA Type | Style Reference |
| Base Model | Krea-2 Turbo |
The LoRA strength parameter controls the configured influence of the selected LoRA. A strength value of 1.00 represents the full configured LoRA weight when the model is applied. The exact style-reference behavior depends on the connected nodes and reference inputs available within the complete workflow.
Prompt Enhancement Configuration
The workflow includes an optional LLM-based prompt enhancement system that can be used to refine or expand image generation prompts. In the current configuration, both prompt enhancement and LLM thinking mode are disabled.
| Prompt Enhancement | Disabled |
| LLM Thinking Mode | Disabled |
| Maximum LLM Tokens | 512 |
With prompt enhancement disabled, the optional LLM-based rewriting stage is not enabled. The maximum token value remains configured at 512 and can be used when prompt enhancement is enabled.
Additional Parameters
| Seed | 251161685108094 |
| Image Width | External Input |
| Image Height | External Input |
| Diffusion Model Precision | INT8 |
| Conditioning Model Precision | FP8 Scaled |
| Style Reference Strength | 1.00 |
Surrounding Nodes and Workflow Integration
The Krea-2 Turbo Style Reference subgraph consolidates model selection, LoRA configuration, prompt enhancement settings, and image generation parameters into a reusable workflow component. Width and height are exposed as external inputs, allowing other nodes to control image dimensions.
In a typical ComfyUI image generation workflow, upstream nodes supply text prompts and image generation parameters. When style reference functionality is supported by the connected nodes, additional visual guidance may be supplied through reference-conditioning inputs.
The Qwen3-VL conditioning model processes the supplied instructions and provides conditioning representations for the Krea-2 Turbo diffusion model. The selected Style Reference LoRA can modify the diffusion model's behavior when applied through the appropriate workflow connections.
After generation, the Qwen Image VAE decodes the latent representation into a visible image. The resulting image can be connected to preview, saving, or additional processing nodes. Exact upstream and downstream connections depend on the complete ComfyUI workflow.
Workflow Highlights
- ✓ Krea-2 Turbo: Uses the Krea-2 Turbo diffusion model for AI image generation.
- ✓ INT8 Quantization: Helps reduce diffusion model memory requirements compared with higher-precision weights.
- ✓ Qwen3-VL Conditioning: Integrates the Qwen3-VL 4B FP8-scaled model for text and vision conditioning.
- ✓ Style Reference LoRA: Includes the krea2_style_reference model with a configured strength of 1.00.
- ✓ Prompt Enhancement: Supports optional LLM-based prompt processing, currently disabled.
- ✓ Flexible Image Dimensions: Width and height are controlled through external node connections.
- ✓ VAE Decoding: Uses the Qwen Image VAE to convert generated latent representations into images.
- ✓ Reusable Subgraph: Organizes model configuration, LoRA settings, and generation parameters for integration into larger ComfyUI workflows.
Workflow Summary
This ComfyUI workflow combines the Krea-2 Turbo INT8-quantized diffusion model with Qwen3-VL 4B FP8-scaled conditioning and the Qwen Image VAE. It includes a Style Reference LoRA configuration with a strength value of 1.00.
Prompt enhancement and LLM thinking mode are disabled, while the maximum LLM token value remains set to 512. The workflow uses externally connected width and height inputs for image dimension configuration.
The Style Reference LoRA provides an additional configuration option for controlling visual style when applied within the workflow. The subgraph organizes model components, prompt settings, and LoRA parameters into a reusable ComfyUI image generation configuration.