Hotaru Blend

This Illustrious based checkpoint merge produces 2D Anime pictures with a clean and detailed style.
It is the result of merging 2 of my current favorite anime checkpoints.

Update

V-pred Version 1.0
This a Noob AI V-pred based version of Hotaru Blend. It is still work in progress and the style is not yet where I want it.
But it is already working very well. It has a good prompt adherence and results are reproducible to some degree.

Recommended settings:
Sampler: Euler A + SGM Uniform or DDPM + Align your Steps
CFG: 3-5 (I prefer 3.5)
Steps: 25

merged models
copycat-noob obsV1(vpred)
hikariNoobVPred_121
+ additional styling loras

Version 4.0 DMD2
This release is the same model as Version 4 with an additional DMD2 lora merged into for faster image generation. The quality is nearly identical to the main model, except that this version has a bit more saturated colors.

Recommended settings:

Sampler: Euler A + SGM Uniform
Steps: 8-12 (recommended 10)
CFG: 1.5 - 2 (a bit higher should also work)

Version 4.0

For this version I added an additional personal merge model which improves the creativity of the model. The colors are now a bit less vibrant and instead have "smoother" look now.

Version 3.0

This release provides some overall improvements. I tried to make the style more like 2D Anime illustrations.

I made 2 quality embeddings for positive and negative prompt, that can be used an an alternative to the below mentioned Quality tags.

You can find them here: https://civitai.com/models/1821398?modelVersionId=2061138

Version 2.0
This release improves the overall details a bit, most noticeable with the hair and reflections. The prompt adherence should also be better then the previous version.

The recommended settings are the same as for Version 1.

Recommended settings

Sampler: Euler A / Scheduler: SGM Uniform

Steps: 25-30

CFG: 3-5

Resolutions: SDXL recommended resolutions (832x1216, 896x1152...)

Highres fix depending on use case

Quality tags:

Positive:

masterpiece, best quality, amazing quality, very aesthetic, absurdres, cute face, slender, detailed eyes, detailed hair

Negative:

lowres, bad quality, worst quality, bad anatomy, sketch, jpeg artifacts, ugly, poorly drawn, blurry, watermark, bad eyes, bright pupils, white pupils, bad hands, bad face, censored, mosaic censoring, bar censor, censor bar, text, signature

merged models
HDA_Rainbow_Illus v1.4
HotaruBreedXL_Anime_v20
DPO full original checkpoint (for DPO fix)

+additional style loras

Permission: no commercial use, no selling, use only on Civitai (please also check the sources for their respective permissions)