Kurvy for Krea 2
workflow: https://pastebin.com/qNUSnTCS
Trigger word: kurvy
Important: Not compatible with Krea Turbo (at least with the settings in the workflow; you may have better luck using your own workflow if you want to use Turbo)
This model was trained on very descriptive captions, so you can be very descriptive and nuanced with the type of body you want.
Lora strength:
No character lora use 1-1.5
With character lora 0.4-1.2
Try switching the sampler to deis_2mon the second phase if you have any issues.
--------
Kurvy is a body-focused model for Krea 2, built around women with fuller, curvier silhouettes—from slim-thick and athletic-curvy builds to voluptuous and plus-size bodies.
The model focuses on creating balanced, feminine proportions with curves in all the right places—fuller hips and thighs, shapely legs, defined waists, and naturally fuller figures—while maintaining believable anatomy and body diversity.
Kurvy is tuned toward a modern social-media photography aesthetic rather than polished studio photography. Think Instagram posts, mirror selfies, casual outfit shots, lifestyle photos, vacation pics, street photography, bedroom selfies, and everyday creator content.
📱 The Look
Slim-thick, curvy, voluptuous & plus-size body types
Fuller hips, thighs and naturally curvy silhouettes
Realistic, balanced body proportions
Smartphone-style photography
Instagram / TikTok / influencer aesthetic
Casual selfies, mirror shots & candid lifestyle photos
Natural skin texture and everyday lighting
Less “professional photoshoot,” more “someone actually took this on their phone”
Kurvy is especially useful when you want Krea 2 generations to have more shape, more body diversity, and a convincing social-media feel without every image looking like it came from a professional fashion campaign.
-----
🎬 Introducing Creator Studio
The All-in-One AI Influencer Operating System
Creator Studio brings the entire AI influencer workflow into one connected production studio.
Plan content, create consistent characters, generate and edit images, produce videos and audio, write captions, publish posts, and review performance—all from one creator-focused platform.
We tested more than 100 AI models and platforms, selected the strongest options, and put them inside one useful wrapper with the tools creators need to produce high-quality AI content.
Whether you manage one virtual creator or an entire roster, Creator Studio helps turn scattered generations into a repeatable content-production system.
Bring Your Own API or ComfyUI Workflow
Creator Studio is not a closed ecosystem. Add any API or import a custom ComfyUI workflow, give the agent the documentation, and it can figure out the integration, build the controls, and generate the UI for it.
All 45 Creator Studio Tools
Create and Direct
• Copilot • Image Generation • Image Edit • AI Swap • Copy
Characters and Campaigns
• Photoshoot Studio • Character Creation • Character Manager
Video and Sound
• Video Studio • Motion Control • Audio Studio • Lipsync Studio
Finish and Enhance
• Upscale Studio • Skin Enhancement • Post-Process
Workflows and Operations
• ComfyUI Workflows • Provider Builder • Caption Studio • Publish Studio • Social Analytics • LoRA Training
Automated Flows
• Inspiration Dump • InstaClone • InstaReel • UGC Creator • Hybrid Lipsync • Watermark
See It in Action
https://ultramuse.art/creatorstudio/demo/
━━━━━━━━━━━━━━━━━━━━
🧬 Krea 2 & Z-Image LoRA Trainer
Train your own custom character and style LoRAs for Krea 2 and Z-Image.
The trainer helps you prepare your dataset, configure the training, monitor its progress, and create reusable LoRAs for consistent AI influencers, characters, styles, and creator content.
The Krea 2 and Z-Image LoRA Trainer is available through the UltraMuse Discord server:
👉 [Access the LoRA Trainer Here] - https://ultramuse.art/train-lora
Description
FAQ
Comments (21)
is this s checkpoint or a lora?
definitely a lora, just miscategorized
Is this the trigger? wide hips, narrow waist, hourglass body, slim thick body, kurvy woman
The workflow link in the description is broken and the prompts in your images look like something you'd pass to an llm.
The trigger word is Kurvy
It's more like opposite. It's because OP is using a prompt building interface that constructs these monstrosities. Some are quite good, but others just seem like blind luck lol.
@makiaeveli I just realized there's a bug in my interface. Those instructions were supposed to be flattened and applied, not included into the prompt itself lol...but eh I guess that still worked too.
@UltraMuse I mean only a few of the prompts seem broken, and at least its something.
Thanks for updating the workflow link. My problem was I was using Krea2 Turbo and it was having very little effect. The lora is working well with Krea2 Raw.
Your uploaded images are nice but I can't recreate them. Your lora also doesn't really do much of anything compared to turning it off. Something's off.
actually it adds lighting meaningully
Are you using all the exact models. The turbo makes a major difference as well
https://civitai.com/models/2727641/krea-2-turbo-lora-256dim
@UltraMuse Didn't see your workflow before. I was just using raw with and without turbo loras. Your workflow is interesting, I'm still getting used to comfy but this looks very new compared to what I was doing. I see why your images could look so different
@civitdude1342 the biggest thing is the niche sampler via the Clownshark nodes I think. I can't exactly recreate the photos but I also wasn't testing seed. It meaningfully improves adherence though, I actually got a few prompts that never went before using this setup (goth grandma lol). I do still get better results with my own (or Ashen's) sampler step 1 being 10 steps of euler-simple and sampler step 2 being 10 steps of euler-linear-quadratic. Well, the faces are less unique, but it also doesn't need the censor bypass and is like a minute faster lol. I think part of the problem with the workflow is the OP's last sampler step keeps CFG at 5. I think if you just turn it to 1 and then double or triple the Deis steps, it'll be less cursed sometimes. But will still be censored from that niche quin-zhang sampler.
@UltraMuse This LORA looks incredible! Exactly the kind of body shape that I like.
I'm a beginner learning ComfyUI. Do I have to use the same base model that you have in your workflow, or will another work?
@tomberwick1984524 Another base model should have some form of effect, but because of how other base models (eg finetunes and merges) are put together, no one can really say how it will work until you try it. Generally, these type of general loras work really well with other finetuned base models, but I just use raw like the workflow lol
It doesn't seem to be working properly? Trying at 1.0 strength and it changes the image, but doesn't make them any curvier.
Are you using the workflow? Also make sure you're adding trigger kurvy
It doesn't need the trigger word. You just have to pay attention. It's a 1.5 gb model, this thing will change everything about your image if it can. The best option is to turn off almost all loras that aren't necessary and keep Kurvy at 1.0 weight. Eg Snofs + Kurvy at 1.0 works exactly as planned. Because what's changing is very subtle with no extra prompting: notice the edges of clothing how it digs into flesh, or how hands that grab are now grabbing into more skin. It's about being able to construct women with a given description. Do not expect this lora to just make a sexy goddess if you just write "woman". You have to read some of those long prompts, and find what words you think will matter for you, and apply them. If you do not change your prompts, or do not have phrasing deeply describing the shape of their body, do not expect this to work. The data in the training set is just prompted too meticulously to have nothing provided. These framegrid loras, and really any lora of a high rank, is going to need you to listen to its recommendations. Its why there are so many people struggling on these loras.
TLDR is why you have no idea what you're doing lmfao
Loras like this really highlight the food-truck nature of civitai: all the food comes out of a truck, and most people simply seem to let their thought process end there. For some it may be the novelty of buying something from the truck on the street in a convenient location. But sometimes, there are those rare food trucks that are run by legitimate chefs, that deserve a more nuanced appreciation for the food that may just seem served from the back of a truck.
In my opinion, if people can't see the quality of the examples in the gallery and the simplicity of the workflow, I guess that's on them. Unfortunately I don't get the sustenance of the food, but ideally you also don't rely on us purchasing your outputs.
Either way, this lora and sampler combination absolutely takes outputs to a third level. I have control over body parts I didn't before, from angles I didn't before, in settings and emotions and information I couldn't provide before. This is some secret sauce right here. It's still surprises me that most of the comments are just recoiling at the sight, but I guess it shouldn't. This is America, after all.
As per the workflow, I find I can remove the filterbypass by doing three things: Only including one references to the word "photograph" in the prompt, increasing ETA on the first sampler to a number between 0.6 and 0.7, and reducing the denoise and CFG on the first sampler to 0.9 and a number between 4 and 5 respectively. All three together seem to be removing all large empty spaces or censoring without a bypass. If it still happens, could still likely bump the ETA and CFG more, but bumping the denoise might incur more steps.
Not all of us are from the USA
@tomberwick1984524 lmfao that is fair. It's more like a state of being. American websites make us all a little like them