π‘ Search "huggy" to find all my models on site.
π‘ Civitai switched its backend from A1111 to ComfyUI earlier this year (2024). That means you can't reuse prompts or seeds from last year to generate the same image. It's best to remove all weight values and start fresh.
π This is a big lora complex, read following table!
π Everything is an work in progress:
As you see, I uploaded lots of similar lora made with different epochs, LR and resource. Every lora is subtly unique as I will systematically compare one another by its quality (over sharpen / glitch / half baked), resemblance (to training resources) and character features (torso deformation, poor eyes, unnatural looks) in the future.
π Useful Tips:
β’ Use the Lora with epoch below 7 for a stable result, like V03E06, though lora with higher epoch like V03E10 has better resemblances to the training source;
β’ Most of my works were generated on Civitai (A1111 as a backend). Start from scratch is recommended, if you are using the prompt or parameters on a ComfyUI based generator (remove all weight values on tags).
β’ Check the base model first, a Pony based lora won't work on most SDXL checkpoints, vice versa.
β’ Some loras include multiple trigger codes (aaXXXxx, aaXXXxx, etc), representing different styles / characters.
β’ USE fewer triggering words each time for best result;
β’ Use CFG between 4.5 (complex scene) to 7.5 (simple);
β’ For anime start training steps from a lower value like 19 with Euler a sampler, if you are satisfied with the result, use hi-res fix / face fix (and set Euler to 29) to increase over all quality;
β’ For realistic photos, use DPM++2M Karra, 30+ steps for close up portrait, 35+ steps for full body shot.
β’ There's no universal prompt / lora weight for every generation. If you're using anime style / character lora with realistic checkpoint, starting with a lower lora weight like 0.7. Ignore the lora weight in sample images, due to a recent bug, image generated online will display lora we