Hello everyone,
This collection consists of pose reference images that I've been using for generating pictures.
All of these images are meant to be used with ControlNet.
OpenPose performs very well with SD1.5 models, while its performance with XL, PONY, and Illustrious models is slightly weaker. For front-facing poses, OpenPose tends to work better. Additionally, OpenPose adapts more effectively to different characters and clothing styles compared to other control methods.
Depth is my primary go-to, especially for complex poses, perspective shots, and compositions with strong visual tension. Depth tends to be more accurate and easier to control than OpenPose in these scenarios. However, one drawback of Depth is that it struggles with intricate character clothing. To address this, I currently rely on inpainting or local redrawing to refine the results.
For now, I’ve temporarily set aside using Lineart, as it doesn’t offer the precise control of Depth nor the multi-character and clothing adaptability of OpenPose. Therefore, I currently recommend sticking with these two methods.
Please use the prompts from each image's filename to generate your images. These prompts will help the model understand and produce more accurate results.
Also, all my images are sized at 768x1152. Some poses may actually work better in a 1024x1024 format, so feel free to crop and upscale them as needed.
I’ll be updating this collection regularly and I welcome everyone to use these poses and share your valuable feedback with me!
Have Fun!!!
Description
FAQ
Comments (5)
Nice! I'll be trying these!
WRT lineart. It's not very useful per se, but I usually use it to lock the hands and the weapon in place (with high strength, around 1.1-1.2).
I highly recommend using depth.
Lineart is quite useful for boundary control, but there is almost no sense of hierarchy, especially for the effects of wide-angle and perspective.
@affa1988 Oops sorry, I forgot to mention what I use: 3 at a time (Depth, Lineart, OpenPose). I agree with your observation with lineart: I only use it for details or weapons, I erase the rest of the lineart as it's not useful (I'd wage even counterproductive)
EDIT: If you manage to get a few poses with a hammer, like you did here with an axe, it would be nice (I have a character I'm trying to make, and getting a good hammer is problematic especially for non realistic checkpoints)
@einar_rainhart Sometimes you have to train a LORA specifically for a hammer posture to solve specific posture problems.I will use models or specific game screenshots from multiple angles for training a LORA to solve the problem, which is not easy and requires some time.
@affa1988 Oh, I know the feeling! I use CN along with a weapon LoRA so that whatever is in the hand of the character is as close as possible as what the CN is showing.
Nice idea with the game screenshots!