I hope you all enjoy the LoRA. Remember you can use my LoRAs in your checkpoint models long as you give credit. I cannot guarantee this works the same on every model.

I'm working on a small AI Discord Community you can join here: https://discord.gg/ehhMrD4PQT

DISCLAIMER - WHAT'S BELOW IS NOT THE RECOMMENDED SETTINGS, YOU WILL HAVE TO FIGURE THAT OUT YOURSELF!

( W.I.P ) V0.5 Version Information:

• Dataset includes images made from the previous datasets and cleaned it up to just the images I want for the style

• Dataset includes images from only the LoRA mix used to make the beta version.

• Dataset also includes images using the LoRA mix from the Mix I used with Beta plus V0.1 of the LoRA added into said mix.

For those who it may concern.

The checkpoint used to generate images in the dataset: Hex Furry and More V0.5 since it likes the LoRA somewhat also used Crystal VAE V3.

And these for the prompts and settings:

Sampler Euler

Scheduler: Normal

Steps: 20

CFG: 4

Positive:

[PROMPT], illustration, anime coloring, 2GDA1, zzkr_9528, N0xfallRev, animalshapes, ep1c7, arctenox, r1ng3k0, masterpiece, best quality, highest quality, absurdres, highres, extremely detailed, mature, official art, official style, [(artist:nt00:0.01), (artist:fuzichoco:0.01), (artist:swav:0.03), (artist:crote:0.03), (artist:unfairr:0.04), (fate_/(series/):0.2), (bleach:0.5), (black_clover:0.3)], usnr, Flatline, Flat vector illustration, 0kaz3, shi3j_illu, e0ce69, skyger_style

Negative:

worst quality, low quality, normal quality, lowres, blurry, out of focus, jpeg artifacts, compression artifacts, ugly, bad anatomy, bad proportions, bad hands, deformed, mutated, loli, shota, aged down, child

Training Params:

V0.1 Version Information:

• Dataset Includes images made from only the Beta Version.

• Dataset includes images made from the Beta Version in a Small LoRA Style Mix.

• Dataset includes images from the dataset that made the Beta Version.

• Dataset includes a little bit of images I fixed through Nano Banana.

• Dataset includes images from the LoRA Mix that made the Beta Version.
  

For those who it may concern.

The checkpoint used to generate images in the dataset: Hex Ionsyx V1.0, and Atlas Core V0.6, Hyphoria V0.01 also used Crystal VAE V3.

This LoRA Mix in question uses:

• Arctenox's Styles - Ouroboros | Series 2 [Illustrious] on 58%

• Lobsteranian style - Illustrious on 11%

• Bleach Anime Style LoRA on 10%

And these for the prompts and settings:

Sampler: Euler Ancestral / Euler

Scheduler: Normal

Steps: 22 / 20 / 25

CFG: 3.7 / 4

Positive for the LoRA Mix made images inside of the Dataset:

illustration, anime, 2390241, [PROMPT], masterpiece, best quality, highest quality, absurdres, highres, extremely detailed, mature, official art, official style, [(artist:nt00:0.01), (artist:fuzichoco:0.01), (artist:swav:0.03), (artist:crote:0.03), (artist:unfairr:0.04), (fate_/(series/):0.2), (black_clover:0.3)], bleach_tybw_style, lobsteranian

Positive for the Beta Version of the LoRA only made images inside of the Dataset:
A few images use this:

illustration, anime, 2390241, masterpiece, best quality, highest quality, absurdres, highres, extremely detailed, mature, official art, official style, [PROMPT]

The rest use this:

[PROMPT] 2390241, masterpiece, best quality, highest quality, absurdres, highres, extremely detailed, mature, official art, official style, [(artist:nt00:0.01), (artist:fuzichoco:0.01), (artist:swav:0.03), (artist:crote:0.03), (artist:unfairr:0.04)]

Negative used for everything:

worst quality, low quality, normal quality, lowres, blurry, out of focus, jpeg artifacts, compression artifacts, ugly, bad anatomy, bad proportions, bad hands, deformed, mutated, loli, shota, aged down, child, extra digits, fewer digits, 6 fingers, 4 fingers, 3D, semirealistic, photorealistic, realistic

Training Params:
{ "engine": "kohya", "unetLR": 0.0001, "clipSkip": 2, "loraType": "lora", "keepTokens": 1, "networkDim": 64, "numRepeats": 1, "resolution": 1024, "lrScheduler": "cosine_with_restarts", "minSnrGamma": 5, "noiseOffset": 0.1, "targetSteps": 925, "enableBucket": true, "networkAlpha": 32, "optimizerType": "AdamW8Bit", "textEncoderLR": 0.00003, "maxTrainEpochs": 10, "shuffleCaption": true, "trainBatchSize": 4, "flipAugmentation": false, "lrSchedulerNumCycles": 3 }

Beta Version Information:

This is a style mix made using:

• Dataset includes images using a LoRA mix.

For those who it may concern.

The checkpoint used to generate images in the dataset: Hex Ionsyx V1.0

The LoRA Mix in question uses:

1. Ringeko | by UTdT69 on 78%

2. Epic Seven - Official Art Style | Series 2 [Illustrious] on 62%

3. Animal Shapes Style on 50%

4. Zackary911 Style illustrious on 50%

5. Arctenox's Styles - My Style | Series 1 [Illustrious] on 37%

6. Flatline - Illus x Niji [BETA] on 32%

7. Arctenox's Styles - Noxfall | Series 1 [Illustrious] on 30%

8. Arctenox's Styles - Celfade | Series 1 [Illustrious] on 28%

9. Okaze | by UTdT69 on 22%

10. Arctenox's Styles - Detailed Anime | Series 1 [Illustrious] on 10%

11. I can't believe is not Kegant on 5%

12. 薄塗り / USNR STYLE on 2%

13. Shiej | Shiiro's Styles on %

And these for the prompts and settings:

Sampler Euler

Scheduler: Normal

Steps: 20 - 35

CFG: 4

Positive:

[PROMPT], illustration, anime coloring, 2GDA1, zzkr_9528, N0xfallRev, animalshapes, ep1c7, arctenox, r1ng3k0, masterpiece, best quality, highest quality, absurdres, highres, extremely detailed, mature, official art, official style, [(artist:nt00:0.01), (artist:fuzichoco:0.01), (artist:swav:0.03), (artist:crote:0.03), (artist:unfairr:0.04), (fate_/(series/):0.2), (bleach:0.5), (black_clover:0.3)], usnr, Flatline, Flat vector illustration, 0kaz3, shi3j_illu, e0ce69, skyger_style

Negative:

worst quality, low quality, normal quality, lowres, blurry, out of focus, jpeg artifacts, compression artifacts, ugly, bad anatomy, bad proportions, bad hands, deformed, mutated, loli, shota, aged down, child

Training Params:
{ "engine": "kohya", "unetLR": 0.0001, "clipSkip": 2, "loraType": "lora", "keepTokens": 0, "networkDim": 32, "numRepeats": 1, "resolution": 1024, "lrScheduler": "cosine_with_restarts", "minSnrGamma": 5, "noiseOffset": 0.1, "targetSteps": 555, "enableBucket": true, "networkAlpha": 16, "optimizerType": "AdamW8Bit", "textEncoderLR": 0.00005, "maxTrainEpochs": 10, "shuffleCaption": true, "trainBatchSize": 4, "flipAugmentation": false, "lrSchedulerNumCycles": 3 }