Leg Behind
A focused Krea 2 pose LoRA trained for 5,000 steps. It works very well on its own and also combines especially well with Double Pen for complementary pose control and composition.
How to use
- Trigger word:
leg_behind - Base model: Krea 2
- Training: 5,000 steps / 33 epochs
- Recommended strength: 1.0 to 1.4
- Start at 1.0, then increase gradually toward 1.4 for stronger pose adherence.
- Works standalone; for the paired setup, also load Double Pen and add
Double_pento the prompt. - Use clear anatomy, camera-angle and composition prompting for the strongest adherence.
Works with my Krea 2 character LoRAs
This pose LoRA works very well with my face-centered Krea 2 character LoRAs:
Created with LoRA Dataset Studio
This LoRA was created with LoRA Dataset Studio (LDS), the open-source dataset curation, captioning, training and checkpoint-testing workbench. Need help, want to share results or follow development? Join the LDS Discord.
Description
Trigger: leg_behind. Works standalone at a recommended strength of 1.0 to 1.4, and can also be paired with Double Pen.
FAQ
Comments (4)
Krea already can do this on it's own.
Legs raised with feet behind head.
Legs raised with feet crossed behind head.
Or legs up.
please post here your generation with the lora and without the lora tell me witch one do you prefer with visible metadata
@perfectgf it wasn't hate on your lora. I was just letting you know it could do it already incase you didn't find the correct prompting. Sorry if it came off that way.
@Plaguekind No worries, I didn’t take it badly at all! Your approach sounds interesting, so feel free to share your prompting methods and results here. Prompt-only methods can sometimes be inconsistent or unpredictable, which is exactly why I preferred creating a LoRA for more reliable results. It would be interesting to compare both approaches with visible metadata.