CivArchive
    沢田さん - Teacher Sawada Isekai de Cheat Skill wo Te ni Shita Ore wa / 沢田 (異世界でチート能力を手にした俺は) - V4
    NSFW
    Preview 933149
    Preview 933147
    Preview 933150
    Preview 933152
    Preview 933151

    30th Release, 沢田さん - Teacher Sawada From Isekai De Cheat Skill - 異世界でチート能力を手にした俺は.

    You can now request a Character's LORA. CLICK HERE

    Put everything in Filter for seeing all images!

    Click on Show more for Weight & Trigger recommendation, Also click on About This Version at Right Corner for more Info.

    IDK, Since it's a New Anime and Art on it is Inexistant, So i decided to create a lora of Sawada.

    About this Lora:

    • Lora trained on 59+ Images that i sorted manually, at the max quality possible (Anime Screencaps 1080P Quality), Were used in the datasets) SFW+.

    • Can do Official Default Skin very precisely.

    • Flexibility is Very good.

    • Can do Anime TV Like Sawada & Original Art. (If you lower the weight to 0.7-0.8)

    • NAI Used, Work with Majority of the Anime Model's.

    • Work's Perfectly at Weight 1.

    • Work's greatly with Detail Tweaker Lora. (I Recommend to use it).

    For Prompts & Weight:

    Flexibility is Very Good. (See examples images).

    About Trigger Words & Prompts: SawadaV4 (For Trigger) & for prompts just refer to examples images.

    Weight recommended: 0.6-1. - Best Result 0.7-1.

    Leave a Review & Share you're image in comments if you appreciate this Work :).

    PS: Will probably update when Anime is Finish.

    Description

    I pruned Purple eyes, & Purple Hair, so normally you don't need to put in prompt, but if they don't appear Brown put: ((Purple Eyes, Purple Hair)) prompt.

    For default skin, put 0.7-1 Weight, See examples Image for this. (Depends if you want Anime TV Like or Original Art). Original Art = 0.6-0.8

    For Flexibility for like clothes that don't exist put 0.6 - 0.8 Weight, See my Examples images for this.

    Also if you don't want something to appear, put it in Negative Prompt.

    FAQ

    Comments (7)

    LittleJellyMay 30, 2023· 1 reaction
    CivitAI

    Where the heck did you get 59 images of her? That must be a lie... I screen capped literally every shot of her in anime and only got 11 caps :sus:

    KXRO
    Author
    May 30, 2023

    I never lie on what i said, all screencaps All come from FanCaps.

    11 ? Weird ahah, I did dl screencaps when 8 episode where released.

    Also i picked some on google and used a technique in 3-4 images (+Upscale 4x-AnimeSharp For Quality), in FullBody because otherwise no fullbody where in AnimeScreenCaps, but otherwise a solid 50+ Images are all AnimeScreencaps, i can upload datasets if you want ahah.

    LittleJellyMay 30, 2023

    @KXRO I'm curious about the dataset 🤔 so please upload it if you don't mind

    KXRO
    Author
    May 31, 2023

    @LittleJelly Np, here's the Datasets(59+images): https://www.mediafire.com/file/trb4kxnqfmmx91y/SawadaV4.rar/file 

    I sorted all manually like in all my Lora's, also some i have cropped them for quality & angle precision, other not for others reasons, some are upscaled too, i do this when datasets is too low for flexibility for different angles, with Maximum quality Possible Ahah.

    LittleJellyMay 31, 2023

    @KXRO Ah right... 4 differences compared to mine:
    I don't use:
    1. A lot of face close ups as reference images because it sometimes tends to mess up results.
    2. Screen caps with other people
    3. Mirroring images as it messes character design
    4. I don't use same screen cap when character just changes face
    That's why I ended only with 11 images, but it seems like I missed 3-4 scenes with her.

    KXRO
    Author
    May 31, 2023

    @LittleJelly Yea if you have too much of face close up it's not good, but if you mix with upper body + full body, that's will don't matter from my experience since AI know what to put at these distance (Need minimum of 3-4 FB + Upper body same), Screencaps with other people i tryed to minimize it, it's not too much, mirroring is a simple technique when you have very low datasets, you just have to crop or reverse image and the AI will think it's another image atm it's not very popular but, (Very good to do when you're character as very low Art on it, like my Sushang,Cocolia ect). Well all of this come from my experience creating Lora with little or inexistant Art datasets.

    After all What is matter is the Result, if the result was bad i will don't post it obs, for the moment i'm satisfied about my result on this Lora, He's Not Bad at Flexibility and can do some different Angles with different face with a not bad quality, Specially in 0.7-0.8 (Where you can create original Art with it).

    QuinceyForderOct 15, 2023· 1 reaction
    CivitAI

    Perfect! I'm working on a fanfic for Cheat Skill Level Up, focused on Sawada sensei, and this LoRA is exactly what I needed. Thanks!