diff --git a/diffsynth/prompters/omnigen_prompter.py b/diffsynth/prompters/omnigen_prompter.py index 8a6c38b..616efab 100644 --- a/diffsynth/prompters/omnigen_prompter.py +++ b/diffsynth/prompters/omnigen_prompter.py @@ -7,8 +7,37 @@ from PIL import Image from torchvision import transforms from transformers import AutoTokenizer from huggingface_hub import snapshot_download +import numpy as np -from OmniGen.utils import crop_arr + + +def crop_arr(pil_image, max_image_size): + while min(*pil_image.size) >= 2 * max_image_size: + pil_image = pil_image.resize( + tuple(x // 2 for x in pil_image.size), resample=Image.BOX + ) + + if max(*pil_image.size) > max_image_size: + scale = max_image_size / max(*pil_image.size) + pil_image = pil_image.resize( + tuple(round(x * scale) for x in pil_image.size), resample=Image.BICUBIC + ) + + if min(*pil_image.size) < 16: + scale = 16 / min(*pil_image.size) + pil_image = pil_image.resize( + tuple(round(x * scale) for x in pil_image.size), resample=Image.BICUBIC + ) + + arr = np.array(pil_image) + crop_y1 = (arr.shape[0] % 16) // 2 + crop_y2 = arr.shape[0] % 16 - crop_y1 + + crop_x1 = (arr.shape[1] % 16) // 2 + crop_x2 = arr.shape[1] % 16 - crop_x1 + + arr = arr[crop_y1:arr.shape[0]-crop_y2, crop_x1:arr.shape[1]-crop_x2] + return Image.fromarray(arr) diff --git a/examples/image_synthesis/sd35_text_to_image.py b/examples/image_synthesis/sd35_text_to_image.py index cd8c467..37f2da8 100644 --- a/examples/image_synthesis/sd35_text_to_image.py +++ b/examples/image_synthesis/sd35_text_to_image.py @@ -2,18 +2,27 @@ from diffsynth import ModelManager, SD3ImagePipeline import torch - model_manager = ModelManager(torch_dtype=torch.bfloat16, device="cuda", model_id_list=["StableDiffusion3.5-large"]) pipe = SD3ImagePipeline.from_model_manager(model_manager) -prompt = "A capybara holding a sign that reads Hello World" -negative_prompt = "" +prompt = "a full body photo of a beautiful Asian girl. CG, masterpiece, best quality, solo, long hair, wavy hair, silver hair, blue eyes, blue dress, medium breasts, dress, underwater, air bubble, floating hair, refraction, portrait. The girl's flowing silver hair shimmers with every color of the rainbow and cascades down, merging with the floating flora around her." +negative_prompt = "worst quality, low quality, monochrome, zombie, interlocked fingers, Aissist, cleavage, nsfw," +torch.manual_seed(1) image = pipe( - prompt=prompt, + prompt=prompt, negative_prompt=negative_prompt, - cfg_scale=3.5, - num_inference_steps=28, width=1024, height=1024, - seed=0 + cfg_scale=5, + num_inference_steps=100, width=1024, height=1024, ) image.save("image_1024.jpg") + +image = pipe( + prompt=prompt, + negative_prompt=negative_prompt, + cfg_scale=5, + input_image=image.resize((2048, 2048)), denoising_strength=0.5, + num_inference_steps=50, width=2048, height=2048, + tiled=True +) +image.save("image_2048.jpg")