- Update: --->
- Only CFG 1.0 is supported. The model has a better understanding of anatomy.
- The inference process is more flexible than before; you can use 12-28 steps.
[GITHUB]
[ComfyUI Custom node]
import torch
from pipeline_walkyrie import pipeline_walkyrie
from diffusers import AutoencoderKLWan
from PIL import Image
device = "cuda" if torch.cuda.is_available() else "cpu"
model_dtype = torch.bfloat16
model_id = "kpsss34/Walkyrie-1.3B-v2.0"
pipe = pipeline_walkyrie.from_pretrained(
model_id,
torch_dtype=model_dtype
)
pipe.to(device)
#pipe.load_lora_weights("lora.safetensors")
prompt = "a portrait of a young woman in a nightclub, cinematic film still, ultra wide aspect ratio, oval bokeh, soft highlight bloom, teal orange grading, film grain, moody lighting"
height = 1024
width = 1024
num_inference_steps = 20
guidance_scale = 1.0
generator = torch.Generator(device=device).manual_seed(0)
output = pipe(
prompt=prompt,
height=height,
width=width,
num_inference_steps=num_inference_steps,
guidance_scale=guidance_scale,
generator=generator,
output_type="pil"
).frames[0]
output.save("output.png")