Changelog:

V1.1: Fixed some typos, added an image saver due to request(Connected to 2nd Group)

V1.0: Initial release.


⚠️⚠️User Guide for Lora Tester by Ardent⚠️⚠️

There are 2 Parts of this workflow; the main setup and a number of groups to differentiate the results(Which you can increase the number of as you like, simply grab and copy/paste but don't forget to connect the necessary nodes.)

Each setup groups result will be displayed in the preview node underneath it and the 1st and 2nd Groups results will be displayed on top of each other with a visual comparison slider underneath the Main Setup.

Generally don't touch anything that isn't dyed brown. Well, onto the quick explanation how to use this:

⚠️Main Setup⚠️

There are several things you can change here:

0)Prompt: Duh.

1)Latent Image Choice: If you follow the purple latent thread connecting to KSamplers, you'll see there is a switcher. The details on how to use is explained in the article but by default it is set to 3rd Static Latent(Don't forget to pick resolutions multiples of 64 for faster and reliable generations)

2)Checkpoint and VAE: You can pick whatever you prefer here.

3)Seed Generator: The seed generator is connected to all the KSamplers to have them all use the same seed. I assume you know how to use it. By default it is set to 0 and to increase by 1 each use.

⚠️Group Setup⚠️

I suggest leaving the 1st Group as the control group(The Vanilla Result without any LoRAs). Aside from changing LoRAs, you can also try out different CFG, Denoise, Auraflow, #Steps and Clip Skip from the nodes inside groups.

1)Clip Set Last Layer: Any number lower than -1 causes CLIP skipping by the difference.

2)# Group Clip: You can try different CLIPs if you like but I advise against it. Pick one you can rely on like 'ae' and stick to it.

3)LoRA Loaders: 2 nodes to load 8 is most likely more than necessary. Go wild. Watch things go wilder.

4)KSampler&AuraFlow: Steps increase quality, CFG is prompt adherence, Sampler and Scheduler is a bit complex matters and I am running out-of-space. Denoise is import for randomization so is the AuraFlow. You can simply set denoise to 1 and bypass AuraFlow for more rigid reliable results. Denoise is really impactful for I2I generation. More on the article.

This workflow doesn't save anything by default but you can simply add a 'Save Image' node and connect any KSampler to it, disable unnecessary groups and use it for regular image generation but I don't recommend it.

Feel free to follow me on [CivitAI](https://civarchive.com/user/Ardent)

Cheers,

Ardent.