CivArchive
    Mpreg Concept Set - NAI (Illustrious) - LyCORIS - V4
    NSFW
    Preview 128943283
    Preview 128943320
    Preview 128943297
    Preview 128967741
    Preview 128943308
    Preview 128943318
    Preview 128943535
    Preview 128943536
    Preview 128943510
    Preview 128943524
    Preview 128943509
    Preview 128943522
    Preview 128943505
    Preview 128943526
    Preview 128943515
    Preview 128943504
    Preview 128943516
    Preview 128943521
    Preview 128943528
    Preview 128943531

    I present my mpreg model that I made for personal use. It's not perfect but it does a decent job at enhancing existing tags for various prompts related to male pregnancy, inflation, and lactation and such.

    I went with LyCORIS over LoRA so it would mix better when combined with style art style LoRAs. The model I trained on is NoobAI XL 1.1 but it works great with Illustrious LoRAs and checkpoints.


    V4

    This version was also trained on NAI 1.1 but will work great with Illustrious checkpoints/LoRAs.

    • Split the major concepts into 4 separate datasets (and activation tags) instead of one big mess of a folder. It's not perfect but I was trying to better distinguish the concepts from each other even though they do still overlap within the generations.

      mpreg-set (general breast/pregnancy/penetration/belly concepts)
      mp-obesebelly (obese belly and hyper obese concepts)
      mp-fluidtentacle (tentacle concepts)
      mp-com-set (comic concept but has a small dataset to avoid overinfluencing generations)

    • I refined training settings (I had ChatGPT review my config for me).

    • Split the major concepts into 4 separate datasets instead of one big mess of a folder.

    • I refined the translucent tentacles concept and it's now VERY consistent. I tried to train the tentacle in mouth to make more than one tentacle but I'm tired of fighting with it so I just stuck with single tentacle training images.

    • Improved the quality/style of the internal cross-section view (less 2D/better angles).

      Recommended model strength is 0.7-0.8 to avoid having it overinfluence the style you're going for when combined with style LoRAs.

    The concepts that work well include tags like:

    • pectoral expansion / breast expansion / hyper pecs

    • male lactation

    • belly expansion / hyper belly inflation

    • big belly / hyper belly / obese belly / hyper obese

    • cross-section (for an internal view of anal penetration)

    • cum inside (so the cum is more internal compared to "cum in ass") / cum overflow

    • impregnation / male pregnancy

    • tentacle in mouth + drinking from tentacle / tentacle in ass

    • translucent tentacle + liquid-filled tentacle + white liquid in tentacle + visible liquid in tentacle

    • disembodied penis / penis in ass / penis in mouth

      I tried to train it to work for comic / progression panels (to slight success) but it can be temperamental/generate panels in the wrong order so it may take several attempts to get a decent comic/2koma/3koma/etc.

    Specific image examples of various focus tags (at different levels of tag strength) have been included.

    The settings used for my generation testing were 4-5 CFG, 20-24 steps, Euler A - Karras on Forge with an adetailer (face_yolov8s).


    Re-uploaded due to being less consistent on comic prompts without a main activation tag on previous version. This version has the same training set but with an additional activation tag (mpreg-set).

    V3

    This version was also trained on NAI 1.1 but will work great with Illustrious checkpoints/LoRAs.

    • I fixed incorrect training settings so this version can now support a higher strength without being too overtrained.

    • Trained a concept for obese belly that won't give fat face/thighs/arms (only affects the belly size).

    • Tried to add translucent tentacles with fluid transfer (it semi-works but can be hit/miss).

      Recommended model strength is 0.7-0.8 to avoid having it overinfluence the style you're going for when combined with style LoRAs.

    The concepts that work well include tags like:

    • pectoral expansion / breast expansion

    • belly expansion / hyper belly inflation

    • big belly / hyper belly / obese belly

    • male lactation

    • hyper pecs

    • cross-section / cum inside / cum overflow

    • impregnation / male pregnancy

    • tentacle in ass / tentacle penetration

    • tentacle in mouth + drinking from tentacle

    • translucent tentacle + flowing white substance + white substance flowing through tentacle

    • disembodied penis / penis in ass / penis in mouth

      I tried to train it to work for comic / progression panels (to slight success) but it can be temperamental/generate panels in the wrong order so it may take several attempts to get a decent comic/2koma/3koma/etc.

    Specific image examples of various focus tags (at different levels of tag strength) have been included.

    The settings used for my generation testing were 4-5 CFG, 20-24 steps, Euler A - Karras on Forge with an adetailer (face_yolov8s).


    V2

    This version was also trained on NAI 1.1 but will work great with Illustrious checkpoints/LoRAs. I tried to refine the tagging and removed some excess training images that weren't helping much with the consistency of tags. I did try to improve the cross-section so it better resembles cervical-penetration so penetration goes deeper inside.

    The concepts that work well include tags like:

    • pectoral expansion / breast expansion

    • belly expansion / hyper belly inflation

    • big belly / hyper belly

    • male lactation

    • hyper pecs

    • cross-section / cum inside / cum overflow

    • impregnation / male pregnancy

    • tentacle in ass / tentacle penetration

    • disembodied penis / penis in ass / penis in mouth

      I tried to train it to work for comic / progression panels (to slight success) but it can be temperamental/generate panels in the wrong order so it may take several attempts to get a decent comic/2koma/3koma/etc.

    I've included specific image examples of various focus tags at different levels of tag strength.

    Recommended model strength is 0.5-0.6 to avoid having it overinfluence the style you're going for when combined with style LoRAs.

    The settings used for my generation testing were 3-4 CFG, 20 steps, DPM++ 2M SDE - Karras on Forge with an adetailer (face_yolov8s).


    V1

    I focused more specifically on generic tags as opposed to unique tags so it would simply enhance existing concepts/make them stand out more.

    Those concepts include tags like:

    • pectoral expansion / breast expansion

    • belly expansion / hyper belly inflation

    • big belly / hyper belly

    • male lactation

    • hyper pecs

    • anal cross-section / cum in ass / internal cumshot

    • impregnation

    • tentacle in ass / tentacle penetration

      I tried to train it to work for comic / progression panels (to slight success) but it can be temperamental/generate panels in the wrong order so it may take several attempts to get a decent comic/2koma/3koma/etc.

    I've included specific image examples of various focus tags at different levels of tag strength.

    Recommended model strength is 0.5-0.6 to avoid having it overinfluence the style you're going for when combined with style LoRAs.

    The settings used for my generation testing were 3-4 CFG, 20 steps, DPM++ 2M SDE - Karras on Forge with an adetailer (face_yolov8s).

    Description

    This version was also trained on NAI 1.1 but will work great with Illustrious checkpoints/LoRAs.

    • Split the major concepts into 4 separate datasets (and activation tags) instead of one big mess of a folder. It's not perfect but I was trying to better distinguish the concepts from each other even though they do still overlap within the generations.

      mpreg-set (general breast/pregnancy/penetration/belly concepts)
      mp-obesebelly
      (obese belly and hyper obese concepts)
      mp-fluidtentacle
      (tentacle concepts)
      mp-com-set
      (comic concept but has a small dataset to avoid overinfluencing generations)

    • I refined training settings (I had ChatGPT review my config for me).

    • Split the major concepts into 4 separate datasets instead of one big mess of a folder.

    • I refined the translucent tentacles concept and it's now VERY consistent. I tried to train the tentacle in mouth to make more than one tentacle but I'm tired of fighting with it so I just stuck with single tentacle training images.

    • Improved the quality/style of the internal cross-section view (less 2D/better angles).

      Recommended model strength is 0.7-0.8 to avoid having it overinfluence the style you're going for when combined with style LoRAs.

    FAQ

    LoCon
    NoobAI

    Details

    Downloads
    29
    Platform
    CivitAI
    Platform Status
    Available
    Created
    4/28/2026
    Updated
    5/14/2026
    Deleted
    -
    Trigger Words:
    mpreg-set
    mp-obesebelly
    mp-fluidtentacle
    mp-com-set

    Files

    Mpreg Concept LyCORIS - Illustrious 4_1-000006.safetensors