Not so simple (or is it?) Krea 2 Turbo worklow

Not so simple (or is it?) Krea 2 Turbo worklow

This is my personal workflow for Krea 2 Turbo text-to-image generation.

My workflow is very basic. Krea 2 Turbo is already good out of the box and doesn't need an excessive amount of features to get good results.

This workflow includes standard features such as:

Diffusion Model, CLIP, VAE, and lora loaders:

• Why mention this? - This workflow requires your models to be placed in the model/diffusion_models folder.

• AIO models seem kind of pointless when the model files are already big and most of us probably have the CLIP and VAE models installed already. As a result, there is no standard Load Checkpoint node in this workflow.

No CFG setting on the initial image

• This workflow is setup under the impression that you are using CFG 1. The Basic Guider node handles this. If you need to be able to adjust the CFG for some reason, then the Basic Guider node will need to be swapped for a CFG Guider node or something similar.

Positive Prompt

• The node for your positive prompt has three text boxes available and will combine all three before it is sent to the CLIP Text Encode (Prompt) node.

• Why use this? - This makes it easy for me when I have specific prompts that I like to reuse.

Wildcards

• Ever feel lazy and just want to use some ready made prompts for clothing, backgrounds, etc.?

• The ImpactWildcardProcessor node is here for you. Just save your wildcard files to the custom_nodes/comfyui_impact_pact/wildcards folder and now you can be lazy like me.

• Connect the output of the ImpactWildcardProcessor node to one of the text boxes on the Positive Prompt node and you're set to use them.

Upscale

• Upscaling is done via the Upscale Image node with the option to use an upscale model.

• After the upscale, the image can optionally be refined in Ultimate SD Upscale.

Detailers

• Three detailer groups have been added as of v1a and use the standard detection and SAM models. By default, each detailer node will handle everything that was detected in on go. Toggle off 'combined' on the MASK to SEGS node if you want it to do one detected item at a time.

• Not always needed and will increase the overall generation time if used.

Save Image via Image Saver

• The saved image includes metadata that CivitAI can properly read including resources, prompts, sampler, scheduler, resolution, etc.

Other stuff

• I included CFGZeroStar for the initial image. It can be toggled off without any issues.

• Krea 2 Conditioning Control is included as a test and can be toggled off without any issue.

• A markdown note with links and URLs to all the custom nodes used in this workflow can be found at the start of the workflow. If you are unsure of how to use any of the related nodes, check out the related project page or have your preferred AI chat bot do it for you.

• Runtime from start to finish (while using INT8 versions) at SDXL resolutions 1024x1024 is about 1 minute on a 5060ti 16GB. Bumping up the initial resolution 1.5x increases this to 3 minutes.

Questions I will probably get asked:

Will I add <insert feature here>?

• Short answer: No.

• Long answer: If it's something I decide to use, then yes.

Why do I use Ultimate SD Upscale instead of another KSampler/SamplerCustomAdvanced?

• Answer: I like the results better and I'm just used to using it.

Will I post my workflow to <insert some other platform here>?

• Short answer: No.

• Long answer: If I ever lose enough social credit here to be banned, then yes.