CivArchive
    Preview undefined
    Preview undefined
    Preview undefined
    Preview undefined
    Preview undefined
    Preview undefined
    Preview undefined
    Preview undefined
    Preview undefined
    Preview undefined
    Preview undefined
    Preview undefined
    Preview undefined
    Preview undefined
    Preview undefined
    Preview undefined
    Preview undefined
    Preview undefined
    Preview undefined
    Preview undefined

    Try my new Checkpoint for much better Quality and additional concepts

    TgirlsXL

    TgirlsXL is a step up from the SD1.5 version in realism and reliablility

    • Can create flaccid penises in almost every pose

    • follows prompts much better (large/small or flaccid/erect)

    • better looking penises

    • much better at side poses and poses from behind

    • produces much less body horror like multiple dicks...

    • The penis bulge concept from V4 is not working in this version

    Works great in combination with my other XL Lora

    Planned improvements:

    • Uncut penises

    • Tgirl on Tgirl and Tgirl on Woman poses

    The Model has been trained on default SDXL and is primarily intended for realistic models but I will probably also work for anime models.

    General Tips:

    • The recommended strength is 0.5-0.9, a higher strengths sometimes creates more realistic results.

    • The model seems to follow size and state (erect/flaccid) better if they come close to the beginning of the prompt

    • size and state (erect/flaccid) works better if the prompt is shorter

    • Use highres fix!!! The upsampler does not really matter. Denoise should be at least 0.3.

    • Use the regular model for impainting penises. Select "only mask" set Padding to a >50px and set denoiseing to ~0.6

    Look at the samples for prompts

    Description

    • Better image quality

    • Additional concepts including flaccid penis and bulges

    • Works better in even more poses now

    FAQ

    Comments (19)

    Lbjnumber1Aug 15, 2023· 2 reactions
    CivitAI

    Thanks so much, you are the GOAT. Is there any possibility for a LoRA version as well?

    ffjggrtbjibv
    Author
    Aug 22, 2023

    I uploaded an older Lora version. Sadly it's not as good as V4 and does not support bulges, but it should still be an upgrade over V3b.

    Lbjnumber1Aug 22, 2023

    @ffjggrtbjibv thank you! Much appreciated

    ttb2112Aug 15, 2023· 3 reactions
    CivitAI

    v4 is vastly improved! Nice work! Thank you!

    RomazeoAug 15, 2023· 1 reaction
    CivitAI

    Very impressive :D

    screamlouderAug 16, 2023
    CivitAI

    Amazing results! How did you manage to fit in not only erect and flaccid but also bulge concept? When I try to merge all those concepts into my own lora/locon it becomes worse so I had to split mine into two loras.

    ffjggrtbjibv
    Author
    Aug 16, 2023

    Using a fairly large LR and Batchsize in combination with allot of training steps helped. Strangely adding the bulge concept actually made prompting to other concepts more consistent.

    Miru00Aug 16, 2023
    CivitAI

    Thank you! V4 is great update, provide decent adaptation for many kinds of non-Futa based model. But one thing i thought could be revised later that the details of skin and light seems to be lower than the original model when TgirlV4 is activated. If there is a way to merge the tgirl Loha model into checkpoint, this flaw may be resolved properly! (Realistic Futa came from your TgirlV3 :)

    wang89xAug 16, 2023
    CivitAI

    Looks great, can this be used in inpainting?

    ffjggrtbjibv
    Author
    Aug 16, 2023

    Inpainting works great with the regular epic realism model (not the inpainting model). Denoising should be set to ~0.60.

    alona_chemerysAug 22, 2023

    can confirm it works inpainting a non futa image into one

    374754276130Aug 18, 2023
    CivitAI

    I like penis bulges very much. Could you add a penis under pantyhose

    Otaku4LifeAug 18, 2023· 1 reaction
    CivitAI

    You are a golden god. Thank you for the LYCO. Looking forward to future updates!

    DurudurAug 20, 2023
    CivitAI

    On EasyDiffusion, error says :

    Error: Could not load the lora model! Reason: Could not find text_encoder.text_model.encoder.layers.0.mlp.fc1.hada.w1 in the given object!

    How can i fix it?

    Always thank you for the Lora.

    ffjggrtbjibv
    Author
    Aug 20, 2023· 1 reaction

    You are trying to load the Loha as a Lora, make sure you have the Lycoris extension installed

    EggbenaAug 21, 2023
    CivitAI

    Damn it. The service I use doesn't support other lycoris. Another futa lora down the dang drain for me.

    alona_chemerysAug 22, 2023

    Just start..saving for a gpu ..?

    ffjggrtbjibv
    Author
    Aug 22, 2023· 1 reaction

    I uploaded an old testing Version thats a regular Lora and better then V3 but not a good as V4

    SilasAI6609Aug 22, 2023
    CivitAI

    Amazing work! I would like to share notes with you sometime. I have my reddit link in my profile, I also go by the same name as here on Discord as well. I am a model developer and am doing work with SDXL mostly lately. If you are on the unstablediffusion group in Discord, you can see some of my progress.