CivArchive
    Miko Mido - La Blue Girl (Anima) - v2.1
    NSFW
    Preview 144217657
    Preview 144217651
    Preview 144217654
    Preview 144217653
    Preview 144217655
    Preview 144217649
    Preview 144217652
    Preview 144217631
    Preview 144217633
    Preview 144217641
    Preview 144217639
    Preview 144217647
    Preview 144217638
    Preview 144217646
    Preview 144217644
    Preview 144219677
    Preview 144219704
    Preview 144219701
    Preview 144219700
    Preview 144219688

    v2 of my Anima LORA for the 90s La Blue Girl series' heroine, Miko Mido.

    • header : masterpiece, best quality, absurdres, [safety tag]

      • well, maybe not mandatory, but I included it in almost all my testing images so I should as much mention it. And the safety tag (general, questionable, nsfw, explicit) DOES have a huge impact on your image (which can be a problem if you forget about it).

    • main trigger : 1girl, solo, miko mido, medium breasts, brown eyes, pink bow

    • costumes (to add after the miko mido trigger word and after any breasts/hair/bow modifier if any) :

      • JK : himenogi school uniform, blue jacket, red necktie, white shirt, collared shirt, cyan skirt, pleated skirt, white socks, mary janes, brown shoes

      • Ninja : miroku ninja costume, green vest, sleeveless, open front, cleavage, pink scarf, red leotard, black leggings, red bracers, red boots, pink trim

      • trim the costume tags as needed for your scene framing

    • Optional hair : long sidelocks, very long hair

    • Style (at the end of your prompt) : retro artstyle, 1990s (style)

      • Retro anime style : cel, anime screenshot, retro artstyle, 1990s (style)

    Same dataset than v1, but with some heavy clean-up of the captions. As a result, the model should be more adaptable. It's easier to adjust hair or breasts size with this version for example.

    Strength of the model is generally kept between 0.8 and 1.0 - in most cases 0.9 should be a good starting point, adjust as needed. For tags combination, see the showcase for some simple use cases - the prompts can be pretty messy, but hopefully you will get the idea.

    Protip : artist @rin-sin works very well with this LORA (for obvious reasons), don't hesitate to give it a try if you are looking for some neat 90s ero-anime aesthetics.

    I would really love to work on a multi-characters v3 version with most of the main characters of the series next time, but depending on the complexity I don't know how much time it would take to bake.

    About the Softer and Stronger additional models

    Epoch 13 was the winner after my rounds of testing, but epochs 12 and 14 were almost as good, so I decided to package them along this release. 99% of the time you want to use the default BALANCE safetensor, but the other ones offer some slight advantages for niche use cases :

    • SOFTER is slightly more flexible - good if you need to soften the miko mido style.

    • STRONGER is a bit more rigit - you want to use it to accentuate the miko mido style.

    As a rule of thumb, consider using them only if you need to go lower than 0.8 or higher than 1.0 with the BALANCE lora to reach the style you are aiming for.

    Personally, I use the STRONGER one to generate my fake LBG screenshots.

    Model compatibility

    Testing/XYZ Plot has been done with base model, showcase has been realized with Hassaku v1.1.

    Some recommended models :

    • Hassaku v1.1 has been the most consistent one for me so far for my base line of retro artstyle, 1990s (style) tags + additional prompting for a highly detailed and colorful output.

      • with the cel, anime screenshot tags, it give a very nice 90s retro style - quite not like a cel, but more like a traditional media illustration.

      • with the STRONGER model, it's a pretty good way to generate fake anime screenshots.

    • base Anima is absolutely killing it for that vintage 90s screencap look that looks like it's coming straight from a VHS. Way more blurry than Hassaku v11 tho, but it can be a choice.

    • WaiAnima v10 is pretty good for a retro anime illustration style - the traditional heavy vanilla bent of the model serve us well here.

    • Hassaku v1.3 doesn't seem to work very well

    • Hoseki Lustrous is weird - it works well, but also looks sharper than other models I've tested, so if you want a retro style, the result can looks a bit strange - in an anime uncanny valley way.

    • NovaAnimeAM has the same problem - the lora works well but it absolutely feels too modern in style if you want a retro look.

    • One Obsession or MiaoMiaoHarem still completely refuse to work for me, but maybe it's just a skill issue ?

    DEPRECATED - v1 USAGE

    • header : masterpiece, best quality, absurdres, [safety tag]

    • trigger : 1girl, solo, miko mido

    • pink ribbon, medium breasts are optional, use them if you need to reinforce them.

    • style tag : retro artstyle (towards the end)

    • model strength : 0.8

    Description

    Changelog :

    • fixed a missing flag for caption shuffling in the training process

    LORA was still lacking in flexibility so I retrained it one last time.

    99% of the time you want to use Lora_Anima_Miko-v21-BALANCE.safetensors, but I have also included the safetensors for the previous epoch (Lora_Anima_Miko-v21-SOFTER.safetensors), and the one for the following epoch (Lora_Anima_Miko-v21-STRONGER.safetensors) just in case.

    FAQ

    LORA
    Anima

    Details

    Downloads
    248
    Platform
    CivitAI
    Platform Status
    Available
    Created
    9/29/2026
    Updated
    10/6/2026
    Deleted
    -

    Files

    Lora_Anima_Miko-v21-BALANCE.safetensors

    Mirrors

    Lora_Anima_Miko-v21-SOFTER.safetensors

    Mirrors

    Lora_Anima_Miko-v21-STRONGER.safetensors