A bipolar concept slider for muscle mass and definition, no trigger word, the LoRA weight itself is the control. Weight 0 = no effect (base model). Negative weights move toward a slender body with no visible muscle; positive weights move toward a heavily muscled, powerfully built physique.
V2 Changelog:
No ethnicity drift at any tested weight, either gender
Generalizes correctly across ages it was never trained on
More robust past the trained range
sharper muscles Definition compared to V1
V1:
Recommended range: -3 to +3. ±4 is usable but pushes further into an extreme the training data covers more thinly.
What makes this one different: the same single axis was trained and verified to work across three structurally different subject types, human men and women, fully mechanical androids (build reads as light chassis ↔ heavy armour plating), and anthropomorphic/furry characters. Sample images show all three.
Known limitation: on some human female subjects, the extreme positive end (+3/+4) can show a subtle shift in apparent facial features alongside the muscle growth. A fully "locked" prompt (explicit hair, skin tone, pose, and clothing described in detail) noticeably reduces this.
Usage tip: for best results describe hairstyle, skin tone, pose and clothing explicitly and hold them fixed across your renders, a sparse prompt gives the model room to change things you didn't intend along with the trait.
Description
initial release
FAQ
Comments (10)
thanks for your slider jobs
You’re welcome! ❤️ I’m really glad you’re enjoying the sliders. I’ll keep experimenting with more ideas!
aint no way you're reading my mind like that
Feels a bit more like a weight slider than muscles, at least without specific prompting for the build? Not meant as criticism, just based on my quick tests. This is maybe another case of me pushing the LoRA outside your suggested values but it feels like a smooth progression. Still great for what it does for my generations, and I just may not be properly prompting. Thanks again for the slider work, filling a badly-missed gap since loraholic stepped back and no8d doesn't seem to be doing krea sliders so far. You're hitting a couple of concepts that I hadn't had sliders for since SDXL and I hope you do more.
You’re absolutely right, and I noticed the same thing during my testing, especially with female subjects. At higher weights, it can sometimes feel more like increasing overall body weight rather than specifically adding muscle.
I’m planning to experiment with this as soon as I can and see if I can improve it in a second version through the training process, or if this effect is simply something inherent to Krea 2 itself.
Thanks a lot for the detailed feedback and for testing it beyond the suggested range! I really appreciate it, and I’m glad the sliders are filling that gap for you ✨
Update: Just following up to let you know that v2 has been uploaded
Thanks again for that initial feedback, it was super helpful. I've tuned the model so v2 now properly focuses on sharp, clear muscle definition instead of the weight-slider behavior you experienced in v1.
Give it a try when you get a chance, I'd love to hear your thoughts on the improvement!
@shakir_abdo Nice - retested with v2, I'd say it definitely works better. Still came out more weight than muscle for the woman when using the original prompt, until I changed the clothing - seems to want arms & legs bare to add definition. Also has a more pronounced pale-to-tan skin when I changed the outfits, but mostly when I'm outside the -3/+3 range that you suggest.
Edit: https://civitai.red/posts/31006823 and https://civitai.red/posts/31006905 for v2 in case they don't show up in the gallery.
@boomhilde Thanks for the detailed retest this makes sense. Every training prompt only ever describes the trait through bare skin (sports bra/shorts for the women, shirtless for the men), never through fitted clothing. So it's likely never learned that a clothed woman can read as muscular rather than heavier, probably why you need bare arms/legs to see it.
The skin-tone shift outside your -3/+3 is a known limitation I flagged when shipping this identity isn't fully separated from the trait at the extreme weights, worst on that direction for the female subjects. -3 to +3 is the genuinely verified-clean range past that you're extrapolating beyond what the training data demonstrated.
No guaranteed fix queued, but adding a few prompts that describe definition through fitted athletic wear instead of bare skin is a real, testable next step for a future version. I'll look into it and update this thread if a v3 addresses it. Appreciate you pushing on this.
@shakir_abdo Oh yeah that tracks. It only dawned on me that the clothes might be the cause once I saw v2 vs v1, and with no muscles visible to augment adding bulk does make some sense.
I only spotted the same skin tone shifts in your showcase images after I posted my sample - thanks for having accurate previews btw, it isn't always a given. I like to specifically test outside of the suggested ranges just to see what LoRAs do and to see if or where they break, and I probably need to be clearer on that upfront for future posts. I haven't tested really extreme ranges for your sliders but they hold up well from what I have tried - I've seen a few from others that just go nuts beyond the suggested ranges.
A v3 would be cool at some point but I'd say this does what it needs to do. Really do appreciate your work on this and your other sliders - the reduced rank on this v2 was a nice bonus! Looking forward to seeing what else you have in mind - eyebrow slider looks like a nice addition!
@boomhilde Thanks for the encouraging words and the detailed feedback!
You’re totally right about some sliders holding up surprisingly well even at extreme ranges. For instance, I tested the eyebrow slider all the way from -10 to 10, and it still worked cleanly! That’s mainly because "eyebrows" are a relatively isolated concept for the model to learn. The trickier part is with broader concepts, where cranking up the weight starts pulling in unintended traits and bleeding into other features.
Coincidentally, I’ve actually been experimenting with a new methodology for the past couple of days. If it pans out, it might solve that muscle slider issue and allow me to retrain some of the trickier sliders that didn’t yield great results previously (like nose profile and jaw width).
Fingers crossed! I'll definitely keep you updated once I have something solid to test.