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. <hr />V5This version was trained on both, NAI 1.1 and IllustriousXL<ul><li>I realized splitting the concept folders made minimal difference so it's all combined into a single folder and I removed the extra activation tags as they didn't do much in V4.</li><li>Refined the tagging slightly to remove redundant similar concepts and tagging.</li><li>I refined the training settings again (I had Gemini review my config for me) and this version has minimal influence over the generation art style/etc.</li></ul>The NAI version generally seems to require higher tagging strength than the Illustrious version to have more of an effect but it works well enough. Recommended model strength is 0.5-0.7 to avoid having it overinfluence the generation when combined with additional models (LoRA and LyCORIS).The concepts that work well include tags like:<ul><li>pectoral expansion / hyper pecs / male lactation</li><li>belly expansion / hyper inflation</li><li>big belly / hyper belly</li><li>obese belly / hyper obese (made it larger and more consistent than before)</li><li>internal view (for an internal view of anal penetration / replaced the cross-section tag but both tags are interchangeable for the same effect)</li><li>cum inside (so the cum is more internal compared to "cum in ass" or "cum overflow").</li><li>impregnation / insemination (interchangeable but I find impregnation tag sometimes tends to be overpowering when combined with internal view tag in the same prompt)</li><li>tentacle in mouth + tentacle in ass</li><li>translucent tentacle + liquid-filled tentacle + white liquid in tentacle</li><li>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.</li></ul>Specific image examples of various focus tags (at different levels of tag strength) have been included.<span style="background-color:rgb(26, 27, 30);color:rgb(193, 194, 197);font-family:-apple-system, BlinkMacSystemFont, &quot;Segoe UI&quot;, Roboto, Helvetica, Arial, sans-serif, &quot;Apple Color Emoji&quot;, &quot;Segoe UI Emoji&quot;;font-size:16px">The settings used for my generation testing were 4-5 CFG, 20-24 steps, DPM++ 2M SDE - Karras<span style="background-color:rgb(26, 27, 30);color:rgb(193, 194, 197);font-family:-apple-system, BlinkMacSystemFont, &quot;Segoe UI&quot;, Roboto, Helvetica, Arial, sans-serif, &quot;Apple Color Emoji&quot;, &quot;Segoe UI Emoji&quot;;font-size:16px"> on Forge<span style="background-color:rgb(26, 27, 30);color:rgb(193, 194, 197);font-family:-apple-system, BlinkMacSystemFont, &quot;Segoe UI&quot;, Roboto, Helvetica, Arial, sans-serif, &quot;Apple Color Emoji&quot;, &quot;Segoe UI Emoji&quot;;font-size:16px"> with an adetailer<span style="background-color:rgb(26, 27, 30);color:rgb(193, 194, 197);font-family:-apple-system, BlinkMacSystemFont, &quot;Segoe UI&quot;, Roboto, Helvetica, Arial, sans-serif, &quot;Apple Color Emoji&quot;, &quot;Segoe UI Emoji&quot;;font-size:16px"> (face_yolov8s<span style="background-color:rgb(26, 27, 30);color:rgb(193, 194, 197);font-family:-apple-system, BlinkMacSystemFont, &quot;Segoe UI&quot;, Roboto, Helvetica, Arial, sans-serif, &quot;Apple Color Emoji&quot;, &quot;Segoe UI Emoji&quot;;font-size:16px">).<hr />V4This version was also trained on NAI 1.1 but will work great with Illustrious checkpoints/LoRAs.<ul><li>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) </li><li>I refined training settings (I had ChatGPT review my config for me).</li><li>Split the major concepts into 4 separate datasets instead of one big mess of a folder.</li><li>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.</li><li>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.</li></ul>The concepts that work well include tags like:<ul><li>pectoral expansion / breast expansion / hyper pecs</li><li>male lactation</li><li>belly expansion / hyper belly inflation</li><li>big belly / hyper belly / obese belly / hyper obese</li><li>cross-section (for an internal view of anal penetration)</li><li>cum inside (so the cum is more internal compared to "cum in ass") / cum overflow</li><li>impregnation / male pregnancy</li><li>tentacle in mouth + drinking from tentacle / tentacle in ass</li><li>translucent tentacle + liquid-filled tentacle + white liquid in tentacle + visible liquid in tentacle</li><li>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.</li></ul>Specific image examples of various focus tags (at different levels of tag strength) have been included.<span style="background-color:rgb(26, 27, 30);color:rgb(193, 194, 197);font-family:-apple-system, BlinkMacSystemFont, &quot;Segoe UI&quot;, Roboto, Helvetica, Arial, sans-serif, &quot;Apple Color Emoji&quot;, &quot;Segoe UI Emoji&quot;;font-size:16px">The settings used for my generation testing were 4-5 CFG, 20-24 steps, Euler A - Karras<span style="background-color:rgb(26, 27, 30);color:rgb(193, 194, 197);font-family:-apple-system, BlinkMacSystemFont, &quot;Segoe UI&quot;, Roboto, Helvetica, Arial, sans-serif, &quot;Apple Color Emoji&quot;, &quot;Segoe UI Emoji&quot;;font-size:16px"> on Forge<span style="background-color:rgb(26, 27, 30);color:rgb(193, 194, 197);font-family:-apple-system, BlinkMacSystemFont, &quot;Segoe UI&quot;, Roboto, Helvetica, Arial, sans-serif, &quot;Apple Color Emoji&quot;, &quot;Segoe UI Emoji&quot;;font-size:16px"> with an adetailer<span style="background-color:rgb(26, 27, 30);color:rgb(193, 194, 197);font-family:-apple-system, BlinkMacSystemFont, &quot;Segoe UI&quot;, Roboto, Helvetica, Arial, sans-serif, &quot;Apple Color Emoji&quot;, &quot;Segoe UI Emoji&quot;;font-size:16px"> (face_yolov8s<span style="background-color:rgb(26, 27, 30);color:rgb(193, 194, 197);font-family:-apple-system, BlinkMacSystemFont, &quot;Segoe UI&quot;, Roboto, Helvetica, Arial, sans-serif, &quot;Apple Color Emoji&quot;, &quot;Segoe UI Emoji&quot;;font-size:16px">).<hr /><h3 id="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).">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).</h3>V3This version was also trained on NAI 1.1 but will work great with Illustrious checkpoints/LoRAs.<ul><li>I fixed incorrect training settings so this version can now support a higher strength without being too overtrained.</li><li>Trained a concept for obese belly that won't give fat face/thighs/arms (only affects the belly size).</li><li>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.</li></ul>The concepts that work well include tags like:<ul><li>pectoral expansion / breast expansion</li><li>belly expansion / hyper belly inflation</li><li>big belly / hyper belly / obese belly</li><li>male lactation</li><li>hyper pecs</li><li>cross-section / cum inside / cum overflow</li><li>impregnation / male pregnancy</li><li>tentacle in ass / tentacle penetration</li><li>tentacle in mouth + drinking from tentacle</li><li>translucent tentacle + flowing white substance + white substance flowing through tentacle</li><li>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.</li></ul>Specific image examples of various focus tags (at different levels of tag strength) have been included.<span style="background-color:rgb(26, 27, 30);color:rgb(193, 194, 197);font-family:-apple-system, BlinkMacSystemFont, &quot;Segoe UI&quot;, Roboto, Helvetica, Arial, sans-serif, &quot;Apple Color Emoji&quot;, &quot;Segoe UI Emoji&quot;;font-size:16px">The settings used for my generation testing were 4-5 CFG, 20-24 steps, Euler A - Karras<span style="background-color:rgb(26, 27, 30);color:rgb(193, 194, 197);font-family:-apple-system, BlinkMacSystemFont, &quot;Segoe UI&quot;, Roboto, Helvetica, Arial, sans-serif, &quot;Apple Color Emoji&quot;, &quot;Segoe UI Emoji&quot;;font-size:16px"> on Forge<span style="background-color:rgb(26, 27, 30);color:rgb(193, 194, 197);font-family:-apple-system, BlinkMacSystemFont, &quot;Segoe UI&quot;, Roboto, Helvetica, Arial, sans-serif, &quot;Apple Color Emoji&quot;, &quot;Segoe UI Emoji&quot;;font-size:16px"> with an adetailer<span style="background-color:rgb(26, 27, 30);color:rgb(193, 194, 197);font-family:-apple-system, BlinkMacSystemFont, &quot;Segoe UI&quot;, Roboto, Helvetica, Arial, sans-serif, &quot;Apple Color Emoji&quot;, &quot;Segoe UI Emoji&quot;;font-size:16px"> (face_yolov8s<span style="background-color:rgb(26, 27, 30);color:rgb(193, 194, 197);font-family:-apple-system, BlinkMacSystemFont, &quot;Segoe UI&quot;, Roboto, Helvetica, Arial, sans-serif, &quot;Apple Color Emoji&quot;, &quot;Segoe UI Emoji&quot;;font-size:16px">).<hr />V2This 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:<ul><li>pectoral expansion / breast expansion</li><li>belly expansion / hyper belly inflation</li><li>big belly / hyper belly</li><li>male lactation</li><li>hyper pecs</li><li>cross-section / cum inside / cum overflow</li><li>impregnation / male pregnancy</li><li>tentacle in ass / tentacle penetration</li><li>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.</li></ul>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.<span style="background-color:rgb(26, 27, 30);color:rgb(193, 194, 197);font-family:-apple-system, BlinkMacSystemFont, &quot;Segoe UI&quot;, Roboto, Helvetica, Arial, sans-serif, &quot;Apple Color Emoji&quot;, &quot;Segoe UI Emoji&quot;;font-size:16px">The settings used for my generation testing were 3-4 CFG, 20 steps, DPM++ 2M SDE - Karras<span style="background-color:rgb(26, 27, 30);color:rgb(193, 194, 197);font-family:-apple-system, BlinkMacSystemFont, &quot;Segoe UI&quot;, Roboto, Helvetica, Arial, sans-serif, &quot;Apple Color Emoji&quot;, &quot;Segoe UI Emoji&quot;;font-size:16px"> on Forge<span style="background-color:rgb(26, 27, 30);color:rgb(193, 194, 197);font-family:-apple-system, BlinkMacSystemFont, &quot;Segoe UI&quot;, Roboto, Helvetica, Arial, sans-serif, &quot;Apple Color Emoji&quot;, &quot;Segoe UI Emoji&quot;;font-size:16px"> with an adetailer<span style="background-color:rgb(26, 27, 30);color:rgb(193, 194, 197);font-family:-apple-system, BlinkMacSystemFont, &quot;Segoe UI&quot;, Roboto, Helvetica, Arial, sans-serif, &quot;Apple Color Emoji&quot;, &quot;Segoe UI Emoji&quot;;font-size:16px"> (face_yolov8s<span style="background-color:rgb(26, 27, 30);color:rgb(193, 194, 197);font-family:-apple-system, BlinkMacSystemFont, &quot;Segoe UI&quot;, Roboto, Helvetica, Arial, sans-serif, &quot;Apple Color Emoji&quot;, &quot;Segoe UI Emoji&quot;;font-size:16px">).<hr />V1I 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:<ul><li>pectoral expansion / breast expansion</li><li>belly expansion / hyper belly inflation</li><li>big belly / hyper belly</li><li>male lactation</li><li>hyper pecs</li><li>anal cross-section / cum in ass / internal cumshot</li><li>impregnation</li><li>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.</li></ul>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. <span style="background-color:rgb(26, 27, 30);color:rgb(193, 194, 197);font-family:-apple-system, BlinkMacSystemFont, &quot;Segoe UI&quot;, Roboto, Helvetica, Arial, sans-serif, &quot;Apple Color Emoji&quot;, &quot;Segoe UI Emoji&quot;;font-size:16px">The settings used for my generation testing were 3-4 CFG, 20 steps, DPM++ 2M SDE - Karras<span style="background-color:rgb(26, 27, 30);color:rgb(193, 194, 197);font-family:-apple-system, BlinkMacSystemFont, &quot;Segoe UI&quot;, Roboto, Helvetica, Arial, sans-serif, &quot;Apple Color Emoji&quot;, &quot;Segoe UI Emoji&quot;;font-size:16px"> on Forge<span style="background-color:rgb(26, 27, 30);color:rgb(193, 194, 197);font-family:-apple-system, BlinkMacSystemFont, &quot;Segoe UI&quot;, Roboto, Helvetica, Arial, sans-serif, &quot;Apple Color Emoji&quot;, &quot;Segoe UI Emoji&quot;;font-size:16px"> with an adetailer<span style="background-color:rgb(26, 27, 30);color:rgb(193, 194, 197);font-family:-apple-system, BlinkMacSystemFont, &quot;Segoe UI&quot;, Roboto, Helvetica, Arial, sans-serif, &quot;Apple Color Emoji&quot;, &quot;Segoe UI Emoji&quot;;font-size:16px"> (face_yolov8s<span style="background-color:rgb(26, 27, 30);color:rgb(193, 194, 197);font-family:-apple-system, BlinkMacSystemFont, &quot;Segoe UI&quot;, Roboto, Helvetica, Arial, sans-serif, &quot;Apple Color Emoji&quot;, &quot;Segoe UI Emoji&quot;;font-size:16px">).

Mpreg Concept Set - NAI/Illustrious - LyCORIS

V5This version was trained on NAI 1.1<ul><li>I realized splitting the concept folders made minimal difference so it's all combined into a single folder and I removed the extra activation tags as they didn't do much in V4.</li><li>Refined the tagging slightly to remove redundant similar concepts and tagging.</li><li>I refined the training settings again (I had Gemini review my config for me) and this version has minimal influence over the generation art style/etc.</li></ul>This version will likely need higher tag strength to see more dramatic size increases but it's all the same dataset as the Illustrious version.Recommended model strength is 0.5-0.7 to avoid having it overinfluence the generation when combined with additional models (LoRA and LyCORIS).

Files with SHA256: f771b592c91f301785de46ae7eea78cb406ff3ddca5da2f399cfc8fcfb5948aa