This DORA is designed to recreate the style of the artist A6 (aka eeroku). Read the usage notes for details on each version.

HiRes Fix is recommended for the best results. Most sample images were generated with 1.5x Latent (Nearest) upscale with ~20 steps and 0.58-0.7 denoise. All samples were generated directly from ReForge with no use of inpainting or Adetailer. You may want to try "sketch" or "pencil outline" in the negative to get smoother outlines with a latent upscale.

Please support the original artist wherever possible so they can continue to produce their art. Without them, LORAs such as these would not be possible.

This model is also backed up on Huggingface and CivitAI: https://civitai.com/models/890946/a6-eeroku-style

USAGE NOTES:

For both versions, it is recommended to use it at full strength. If you are using a different checkpoint than it was trained on, you may need to increase the strength to retain the style. Both versions were trained with an additional tag of "pencil outline" that you may want to put in the negative if you are having issues with HiRes fix.

PONY:

This LORA was trained on AutismMix. Most images were tagged with "source_anime" and scores 6-9.

Recommended Positive prompt:

score_9, score_8_up, score_7_up, score_6_up, source_anime

Recommended Negative prompt:

source_pony, (hands:1.1)

NoobAI:

The NoobAI version was trained on NoobAI Epsilon v1.1. This should be compatible with NoobAI V-Pred, as well, as you can see in the sample gallery, just make sure you adjust your sampler/CFG accordingly. For NoobAI, you must use the trigger word "eeroku-style"

Recommended postive prompt:

masterpiece, best quality, eeroku-style

Recommended negative prompt: (you can add "hands" and "fingers" as needed)

worst quality, bad quality, screenshot, sketch, jpeg artifacts

TAGGING NOTES:

The dataset contained over 370 images that were ranked. Most dataset images also included a certain Japanese word for male youths as a tag, so that may contain some of the style data, as well, but it also makes the model more flexible.

Score 9-7/"masterpiece," "best quality," "good quality" images were the most recent and detailed, while score 7/"good quality" were older images or lower detail.

Score 6/"average quality" images are the least detailed and also include some images sourced from manga scans that have been tagged with "source_manga"/"manga" and "monochrome, screentones" tags.