A slider lora to help you make your poses more dynamic and in more extreme angles
Description
Works with both Illustrious and NoobAI models, tested.
Strength is totally optional and will depend on your prompt, but you can use 1, 2, 3 and even 4 in some cases (or more tbh). Most of the images I posted will have that info in the metadata.
IF you push the strength too hard it will flatten the colors unfortunately. this could be fixed with stopping the lora at 60% or 50% steps. Or just img2img/high res. fix
No trigger words needed but the suggested ones might help.
These sliders are trained like this:
https://youtu.be/GaVuQEWqEoM?si=HSVZBaZacxHSaS6W
FAQ
Comments (27)
fantastic!
Would you please tell me how I can stop my lora at 50% or 60% steps? I really can not find any apporach to do that
extensions like "composable lora" allow you to do that. If you are using comfyUI I'm sure there is a way tho I'm not sure what it could be.
Are you having any issues with the lora? I've tried it with some models on civitAI and I feel like the resulting images are very flat, tho I'm not sure if it's the loras fault or that I dont know how to prompt on these models.
For the models I use it works pretty well as shown in the examples tho (without stopping the lora at any point)
@N4Talent It is a very good lora, but high strength can result in some impact on art style, thus i think maybe i can try to control it's steps to reduce such problems.
@LFAA agree with you there. If you are using a1111 regional prompter also has a "lora ending step" setting that works well. (no need to use the regions to use it)
I'm liking this! Very good results so far. Thanks!
thank you! Glad you like it!
Amazing but please tell us about the license
you can do wtv you want with it
it's great awesome work
tysm!
Hey, can you train a version for Qwen Image or share the dataset so I try to train one?
Using this LoRA might alter the original art style.
Hey @N4Talent ! Thank you for making this.
I'm struggling to install it though, the file shows up as "DynamicPoseIL2att_alpha1.0_rank4_noxattn_900steps" which I don't think is normal? It doesn't show up on SD after adding it to my LORA folder either. Am I doing something wrong?
Thanks!
Hey! I'm not quite sure why that could be not gonna lie. Does the name end with .safetensors?
Maybe the extension of the file was lost and it's not recognizing it as a lora.
The name of the file should be "DynamicPoseIL2att_alpha1.0_rank4_noxattn_900steps.safetensors"
(I know it's a weird name but I didnt rename it. I just got the file straight from the training and posted it).
Also, what version or UI of stable diffusion are you using?
@N4Talent Thank you for helping out! I'm using sd-webui-forge-classic, latest version, and for some reason it hides your LoRA by default along with other "sliders" LoRAs. I can access it by toggling 'Always show all LoRA cards (otherwise, those detected as for incompatible version of Stable Diffusion will be hidden)' but it doesn't seem to have any effect on the images generated.
@p4ntalaimon wtf, that's so weird. Are you using an Illustrious/noobAI based model?
@N4Talent Yep, tried with a few, doesn't show for any of them, namely waiIllustrious and Illustrij.
@p4ntalaimon can you link the exact model? I'll try to download them and test on my pc. Maybe it's a forge thingy
@N4Talent Yeah it must be Forge, surprising that no one else seems to be complaining though.
To clarify, I have no issues with other LoRAs.
Here is illustrij (v18): https://civitai.com/models/1025051/illustrij
@p4ntalaimon No idea bro. I tested your same checkpoint in both A1111 and Forge UI with my lora (directly downloaded from this post to make sure it's the same). And it worked in both.
Idk if there is a place to post the images, but I tested with and without the lora and same seed with no other changes. There is quite a difference. Have tried using 4 or 5 or even 8 as a lora weight? result will looked hella cooked but at least you know if it works or not.
It does for me, even in forge.
No idea what it could be ngl... ( I do have an older version of forge tho. Maybe they messed something up in an update )
@N4Talent Oh that's huge! You just put it in models\Lora right, and you don't have a special extension that would let you use it? Do you have the 'Always show all LoRA cards' setting toggled on?
@p4ntalaimon Yep, just dragged the lora onto the lora's folder. No extra extensions. And I might have the toggle active, didnt check if I did or not ngl
Better use this LoRA in prompts that not so specific or may lose control to the pose you want as my using experience.(By the way, it do alter the art style but acceptable for me)
使用这个lora时会改变模型原有的画面风格，需要额外添加一个Image-to-Image节点修复画风，
Details
Files
DynamicPoseIL2att_alpha1.0_rank4_noxattn_900steps.safetensors
Mirrors
DynamicPoseIL2att_alpha1.0_rank4_noxattn_900steps.safetensors
DynamicPoseIL2att_alpha1.0_rank4_noxattn_900steps.safetensors
DynamicPoseIL2att_alpha1.0_rank4_noxattn_900steps.safetensors
DynamicPoseIL2att_alpha1.0_rank4_noxattn_900steps.safetensors
DynamicPoseIL2att_alpha1.0_rank4_noxattn_900steps.safetensors
Available On (1 platform)
Same model published on other platforms. May have additional downloads or version variants.