import torch
from diffusers import DiffusionPipeline
# switch to "mps" for apple devices
pipe = DiffusionPipeline.from_pretrained("ByteDance/Bernini-R", dtype=torch.bfloat16, device_map="cuda")
pipe.load_lora_weights("rzgar/Wan2.2-Bernini-R-Motion-Enhancer-n4w-i2v")
prompt = "Astronaut in a jungle, cold color palette, muted colors, detailed, 8k"
image = pipe(prompt).images[0]Bernini-R Motion Enhancer I2V n4w
LTX-2.3 Add Custom Audio to WAN2.2 / Bernini-R Wordkflow demo
Model description
LoRA pack for Bernini-R and vanilla Wan 2.2 I2V tuned for image-to-video workflows. Includes main motion enhancers (strong focus on motion & rhythm via targeted FFN/attention zeroing) + appearance-only enhancers for anatomical accuracy. Respond well to explicit prompts describing different acts.
Main Motion Enhancers
| File | Variant |
|---|---|
| `bernini_n4w_motion_enhancer_high.safetensors` | High noise |
| `bernini_n4w_motion_enhancer_low.safetensors` | Low noise |
Appearance-Only Enhancers
| File | Variant | Purpose |
|---|---|---|
| `male_genitalia_enhancer_high.safetensors` | High noise | Male anatomical accuracy (appearance) |
| `male_genitalia_enhancer_low.safetensors` | Low noise | Male anatomical accuracy (appearance) |
| `female_genitalia_enhancer_high.safetensors` | High noise | Female anatomical accuracy (appearance) |
| `female_genitalia_enhancer_low.safetensors` | Low noise | Female anatomical accuracy (appearance) |
Usage
Text encoders (recommended)
Bernini-R (recommended)
- Main motion LoRA strength: start at 1.0
- If motion is broken/laggy (version of your LightX2V can influence the output): try 1.5–2.0
- Male enhancer: start at 0.55 when additional male(s) or anatomy needs needs to enters the scene.
- Female enhancer: start at 0.75, adjust based on input image /prompt complexity.
Wan 2.2 (vanilla)
- Main motion LoRA strength 0.9–1.2 works well with vanilla Wan 2.2 (Comfy-Org / KJ FP8)
- Merged models: 0.5 is usually enough
- Strength 2.2 on vanilla reduces prompt reliance. Explicit input images trigger expected animation in ~8/10 cases (tested)
tuned ONLY for I2V usage based on video outputs from Bernini-R FP8, NVFP4, and Bernini-R N4W merge.
Download model
Download them in the Files & versions tab.
- Downloads last month
- 1,652
Model tree for rzgar/Wan2.2-Bernini-R-Motion-Enhancer-n4w-i2v
Base model
ByteDance/Bernini-R