Paper: Calibri: Enhancing Diffusion Transformers via Parameter-Efficient Calibration
Calibri Flux with gates calibration
Guide to run:
import torch
from diffusers import DiffusionPipeline
device = "cuda" if torch.cuda.is_available() else "cpu"
dtype = torch.bfloat16
model_path = "v-gen-ai/flux-calibri-gates"
pipeline = DiffusionPipeline.from_pretrained(
model_path,
torch_dtype=dtype
).to(device)
prompts = [
"a futuristic city at sunset",
"a cute robot playing guitar"
]
images = pipeline(
prompts,
num_inference_steps=15,
guidance_scale=3.5,
height=512,
width=512,
).images
Model was trained with resolution 512 x 512, but it is possible to run with other resolutions.