Creates pose dummy mannequins from reference images.
The intended use case is to use the resulting images as pose references for image generation with e.g. Nano Banana Pro or for LoRA training.
Sample prompt:
Generate a picture of abstract CGI mannequins in the exact same pose as the people in the image. The mannequins should have fingers, toes, and an abstract representation of facial expression. The mannequins do not have any hair. They have perfectly clean gray skin with a black wireframe overlaid on top without any discolorations. The background of the image is a plain featureless middle gray void. Do not include any objects the people in the image may or may not be interacting with.You can get somewhat similar results out of base Qwen-Image-Edit with appropriate prompting, but it is a lot less consistent in my experience.
It seems to work okay with cfg = 1.0 and steps = 20. Higher cfg values can produce better results, but may affect how faithful the result is.
The 8-step Lightning LoRA can also be very viable and often produces cleaner results.
Description
Default prompt:
Generate a picture of an abstract CGI mannequin in the exact same pose as the person in the image.
The mannequin should have fingers, toes, and an abstract representation of facial expression.
The mannequin does not have any hair. It has perfectly clean gray skin with a black wireframe overlaid on top without any discolorations.
The background of the image is a plain featureless middle gray photo studio.
Do not include any objects the person in the image may or may not be interacting with.FAQ
Comments (11)
What a great tool this is! In three quick tests I was inspired to do many complex things I would never have imagined without it! ! will post some images as it goes!
i refuse to download 720.13 MB lora.
Why don't you train a Lora model that can generate images based on CGI models, so that you can generate images based on blender modeling
My current approach is more about using local AI for narrow pre- and post-processing tasks, letting Nano Banana Pro do the heavy lifting. What you're suggesting sounds like a lot of work to accomplish something that (I assume) Nano Banana does much better out of the box.
fwiw, I have experimented with "pose reference to photo" LoRAs and they didn't turn out great. I also have zero experience with 3D modeling software, so I couldn't test a LoRA like this properly and I wouldn't have much personal use for it.
@VV24 In yesterday's test, I found that Nano Banana can indeed generate images of CGI models. In my opinion, your lora model is highly forward-looking and performs much better than the traditional method of generating images using pose images and depth images
@ryeyesuo712 Yeah that's the main reason I made this. I wanted something more detailed than conventional OpenPose stick figures or depth images that would force Nano Banana to follow the reference more accurately.
Mind you Nano Banana will also create Mannequin pose references like these itself (Mannequin type nudes don't seem to trigger its content filters).
But having it available in Qwen adds some flexibility (as well as being cheaper) and I found that my LoRA can produce more faithful results under some circumstances.
@VV24 There is no doubt that it is so
Hello! Do you think this could be trained for Flux2Klein or ZimageBase? Thank you 🙌
Z-Image-Edit isn't out yet, is it? As far as I am aware, you can't do instruction-based edits with Z-Image-Base.
I just did a training run with the exact same dataset on Flux 2 Klein 9B Base and it seems to work okay.
@VV24 Thank you, I can't run Qwen on my potato-PC so this is gold. Thank you very much, I'll give it a try! 🙌
@GlowingGuardianGirl Yea I get it. It's kind of frustrating to make Qwen LoRAs, too, because I know a lot of people can't really use them.
btw I removed the download link from my previous comment now that I published the Flux.2 Klein version of the LoRA to Civitai. You can download it from there if you haven't already.