import torch
from diffusers import DiffusionPipeline
# switch to "mps" for apple devices
pipe = DiffusionPipeline.from_pretrained("SDXL", dtype=torch.bfloat16, device_map="cuda")
pipe.load_lora_weights("jarod2212/Unity_Spectrum_XL")
prompt = "-"
image = pipe(prompt).images[0]Unity Spectrum LoRA XL
- Prompt
- -
- Prompt
- -
- Prompt
- -
- Prompt
- -
- Prompt
- -
- Prompt
- -
- Prompt
- -
- Prompt
- -
- Prompt
- -
- Prompt
- -
- Prompt
- -
- Prompt
- -
Model description
UnitySpectrum_XL is a LoRA designed to generate natural interactions between two people, with stable poses, believable physical contact and no limb fusion.
It was trained on 12 main poses, each with multiple variations, to ensure consistency and generalization:
friendly_hug
side_hug
selfie_together
handshake
shoulder_touch
arm_hold
close_pose
leaning_pose
couple_pose
supportive_pose
soft_embrace
together_stance
The dataset was built with mixed ethnicities, but LoRA does NOT learn ethnicities:
👉 generalizes perfectly to any combination (white/Black/Asian/Latina etc.).
⭐ Recommended checkpoint
LoRA has been optimized for:
👉 Brave Studio XL — Simple / Normal scheduler
It also works on SDXL Base, but Brave Studio guarantees:
more stable geometry
more natural eyes
more consistent skin
less noise
⭐ Recommended weight
LoRA works best between:
👉 0.7 and 1.0
0.7 → softer, more natural
1.0 → more precise on the pose
Not recommended to exceed 1.1.
Example Prompt: 1_usxl, white man and Black woman, friendly_hug, realistic photo, 50mm lens, natural light, detailed skin, soft shadows, clean iris texture, sharp eyes
2_usxl, Asian woman and Latina woman, selfie_together, candid moment, smartphone angle, natural lighting, realistic skin texture, face in focus
3_usxl, Black man and white man, handshake, business casual, indoor lighting, realistic photo, 50mm, detailed hands, clean geometry
4_usxl, white woman and Black woman, side_hug, warm moment, soft light, realistic skin, balanced lighting, sharp focus
Negative prompt: blurry eyes, glossy eyes, distorted pupils, deformed hands, fused fingers, extra limbs, overexposed skin, low detail, bad anatomy
The cover images were created on mage.space for convenience; however, MageSpace does not preserve metadata, so I include example prompts here to guide proper usage
Trigger words
You should use usxl to trigger the image generation.
Download model
Download them in the Files & versions tab.
- Downloads last month
- 22