CivArchive
    Krea 2 Turbo 18+ for Character LoRAs - v1
    NSFW
    Preview 139439204
    Preview 139439191
    Preview 139439185
    Preview 139439187
    Preview 139439196
    Preview 139439179
    Preview 139439180
    Preview 139439182
    Preview 139439188
    Preview 139439186
    Preview 139439189
    Preview 139439198
    Preview 139439194
    Preview 139439193
    Preview 139439197
    Preview 139439192
    Preview 139439184
    Preview 139439199
    Preview 139439200
    Preview 139439207

    Krea 2 Turbo 18+ for Character LoRAs

    For all those who want NSFW content with Character LoRAS, no Face and Body change

    Test this Model in your Browser:

    https://www.seaart.ai/models/detail/d9u0r4te878c73b9ldmg

    Sample Image Settings

    Tool: Forge Neo 2.26

    Steps: 8

    Sampler: Euler A

    Schedule type: Simple / Beta

    CFG scale: 1

    Size: 768x1024

    Model: This

    Module 1: Qwen_Image-VAE.vae

    Module 2: qwen3vl_4b_fp8_scaled

    Diffusion in Low Bits: Automatic (fp16 LoRA)

    Description

    FAQ

    Comments (31)

    KazuhiraAug 12, 2026· 3 reactions
    CivitAI

    Thank you! Can you also make a int8 version please?

    Peli86
    Author
    Aug 12, 2026

    @Kazuhira Added,but untested. Enjoy!

    KazuhiraAug 12, 2026

    @Peli86 Thanks!

    ledeni80Aug 12, 2026· 4 reactions
    CivitAI

    Hands down the best Krea 2 outhere.

    Peli86
    Author
    Aug 15, 2026

    Thanks! Enjoy!

    narugokuAug 12, 2026
    CivitAI

    An INT8 Version Please 🙏🏼

    Peli86
    Author
    Aug 12, 2026

    @narugoku Added,but untested. Enjoy!

    mj_2695635Aug 12, 2026
    CivitAI

    BF16 version please so we can quant ourselves

    Peli86
    Author
    Aug 12, 2026

    @mj_2695635 i am uploading all kind of variations and quants. tell me wich one you want the most, i'll upload it with priority.

    mj_2695635Aug 14, 2026

    @Peli86 Whatever the original checkpoint was, fp32 or bf16 I'm assuming? I'll try to quant to w4a8 but I need the original

    chamo9009Aug 15, 2026

    @Peli86 Int 4?

    Peli86
    Author
    Aug 15, 2026

    @mj_2695635 isn't that int8?

    mj_2695635Aug 16, 2026

    @Peli86 It's mixed precision int4/int8 convrot hybrid. Thank you for the FP32

    xFennec777Aug 12, 2026
    CivitAI

    Great work! Solid for character LoRA, though wet body apparently = cum covered.

    Peli86
    Author
    Aug 12, 2026· 1 reaction

    not trained for cum, it changed char too much, main goal was to enable nsfw and KEEP char. cum is next in v2, not easy.

    xFennec777Aug 12, 2026

    Odd. Any time I say a character is wet or has water on them, it’s white.

    Don’t need cum, just noting wet/water looks odd

    Peli86
    Author
    Aug 13, 2026

    @xFennec777

    https://civitai.red/images/139500460

    https://civitai.red/images/139439184

    https://civitai.red/images/139439182

    xFennec777Aug 13, 2026· 1 reaction

    @Peli86 Interesting. I will play a bit more with my prompts - Thanks!

    AnteMaxxAug 13, 2026
    CivitAI

    Could you explain what LoRas you merged into this one to enable NSFW content? This seems to does it with better then other models I've used. Unless you wish to keep this knowledge to yourself, I understand.

    Peli86
    Author
    Aug 15, 2026

    i mostly train a thing that i want to add to a base model

    opie1Aug 13, 2026· 6 reactions
    CivitAI

    i dont know what you did, i dont know how you did it.
    But it works! as of today this is the best checkpoint which keeps characters consistent!
    i converted the checkpoint to wa4a4 svd quant rank 256 even after that it keeps characters consistent as hell!!!
    after a lot of time i feel like i am using PONY but most if not all the issues of PONY fixed. ofcourse it is still hard to beat PONY when it comes to hardcore stuff but yes, it works really well when it comes to lora compatibility.

    Peli86
    Author
    Aug 15, 2026

    That's very nicely written, thank you so much! Most people only take the time to complain, but rarely to say when something is good. I've never used Pony, I always found SDXL too weak and inconsistent for realism. All the NSFW stuff was just copies of images, no prompt freedom.

    niudanyuleAug 15, 2026· 3 reactions
    CivitAI

    NSFW content cannot be generated.

    Peli86
    Author
    Aug 15, 2026

    Yeah, all the images lie...

    boulbi78Aug 15, 2026

    u got plenty of nsfw outputs using it on the model page...

    niudanyuleAug 16, 2026

    @Peli86 My fault. I was able to successfully generate the content I wanted. The facial features are quite consistent, but the anatomical structure is quite ordinary. Could you please offer some suggestions? Thank you.

    LeemnyxAug 15, 2026· 3 reactions
    CivitAI

    Title did not lie! Working very well with a characters lora, also good with nsfw, and other Krea loras. 🥰

    Peli86
    Author
    Aug 15, 2026

    Enjoy! :)

    Peli86
    Author
    Aug 15, 2026· 4 reactions
    CivitAI

    I've found a vulnerability that causes very long and contradictory prompts to become unstable. I'm currently working on it, but it'll definitely take another week. With my 4070 Ti Super, things don't go that fast.

    niudanyuleAug 16, 2026

    Yes, after testing, some of the prompt words did not work. Thank you.

    LeemnyxAug 16, 2026

    Hope you'll manage to fix it my brother, with same gpu as mine 😅😅😅

    Checkpoint
    Krea 2
    by Peli86
    Download (Beta) View on CivitAI
    woman
    base model
    male
    girls
    sexy
    man
    realistic
    sfw
    game character
    nsfw
    photorealistic
    video game
    person

    Details

    Downloads
    1,457
    Platform
    CivitAI
    Platform Status
    Available
    Created
    8/12/2026
    Updated
    8/17/2026
    Deleted
    -

    Files

    krea2Turbo18For_v1.gguf

    Size:
    9.02 GB
    SHA256:
    03c35e514a30fe5c8cec3cb3eb2ee4fddfcf5609aeb8fab74d4e884ba29d3b84

    Mirrors

    CivitAI (1 mirrors)

    krea2Turbo18For_v1.safetensors

    Size:
    23.26 GB
    SHA256:
    e88d171fa365c1ea60949a4adddf9e49d5c5fe804a365c23c1486682adce6abc

    Mirrors

    CivitAI (1 mirrors)

    krea2Turbo18For_v1.safetensors

    Size:
    23.26 GB
    SHA256:
    1a90fac730239d96bd25827524bd861cd9f10ad7e5ca60c4d3b8b8196f30c63f

    Mirrors

    CivitAI (1 mirrors)

    krea2Turbo18For_v1.safetensors

    Size:
    11.95 GB
    SHA256:
    c1024e92447a1834490e791ddd3fc6843ceaa7aa3669609e0bf4bc3c197fe81d

    Mirrors

    CivitAI (1 mirrors)

    krea2Turbo18For_v1.gguf

    Size:
    9.73 GB
    SHA256:
    ca31b167cb8824f1092876899d85475a44da6064b56b762d7ec440c1f4f529b2

    Mirrors

    CivitAI (1 mirrors)

    krea2Turbo18For_v1.gguf

    Size:
    9.02 GB
    SHA256:
    30ed9b66a94f64184b3a579eb4e476792820155a78e00ed2bb969ba8ed2c61f0

    Mirrors

    CivitAI (1 mirrors)

    krea2Turbo18For_v1.safetensors

    Size:
    45.90 GB
    SHA256:
    2b3983d7d458f321a70992fea7c0c960f4a22f4b93dfba6eb7ac4f302e845b61

    Mirrors

    CivitAI (1 mirrors)

    krea2Turbo18For_v1.gguf

    Size:
    10.53 GB
    SHA256:
    82715c5ae558b7aca56b966a9f3dc503b8b8bc944ddc9d4bfd67258984e23722

    Mirrors

    CivitAI (1 mirrors)

    krea2Turbo18For_v1_txt.safetensors

    Size:
    4.88 GB
    SHA256:
    54bd5144df0bbc25dd6ccadfcb826b521445a1b06ae5a42570bdd2974ca87094

    Mirrors

    HuggingFace (35 mirrors)
    CivitAI (7 mirrors)
    ModelScope (1 mirrors)

    krea2Turbo18For_v1.safetensors

    Size:
    11.95 GB
    SHA256:
    071592880055f5a142d11da068252247f8521e9b16b8182c908bb9cb1c8b5ba6

    Mirrors

    CivitAI (1 mirrors)

    krea2Turbo18For_v1.safetensors

    Size:
    11.94 GB
    SHA256:
    81e4390b5a64f57a4ed270e6d6d07e6962199f523b55f463e79884099c6d5dbb

    Mirrors

    HuggingFace (2 mirrors)
    CivitAI (1 mirrors)

    krea2Turbo18For_v1.gguf

    Size:
    13.27 GB
    SHA256:
    3bb15fba2108e310883d92024d11950969d3eced3009c16dfed714b5a0989d37

    Mirrors

    CivitAI (1 mirrors)

    krea2Turbo18For_v1.safetensors

    Size:
    11.94 GB
    SHA256:
    8fb53b448905ce70165648bb7f3553990953c342e08f3c78850db6f82da15322

    Mirrors

    CivitAI (1 mirrors)

    krea2Turbo18For_v1.gguf

    Size:
    7.61 GB
    SHA256:
    a7628b4ab9a768591aee51fd669ac031116585f3c37f8321edce344879f72ccd

    Mirrors

    CivitAI (1 mirrors)

    krea2Turbo18For_v1.safetensors

    Size:
    11.94 GB
    SHA256:
    939790d08ea004095eb2573dd11944ebc3966da0cf2cac05c53e8f8d57f79338

    Mirrors

    CivitAI (1 mirrors)

    Qwen_Image-VAE.safetensors

    Size:
    242.05 MB
    SHA256:
    a70580f0213e67967ee9c95f05bb400e8fb08307e017a924bf3441223e023d1f

    Mirrors

    HuggingFace (229 mirrors)
    CivitAI (175 mirrors)
    TensorFiles (1 mirrors)
    ModelScope (3 mirrors)
    ModelScope CN (5 mirrors)

    krea2Turbo18For_v1.safetensors

    Size:
    12.29 GB
    SHA256:
    5f1177fee554f9ddf2a8a13ba178a65c937d99c08262bd5c2adab73b3a7db12a

    Mirrors

    CivitAI (1 mirrors)

    krea2Turbo18For_v1.safetensors

    Size:
    6.99 GB
    SHA256:
    62da68cb92348831280448b0cd478ce5060c3dec36562b609c37fdcedbbc7567

    Mirrors

    CivitAI (1 mirrors)