Qwen

/

Qwen-Image-Layered

like 771

Follow

Qwen 62.7k

Tested with an RTX 5000 24GB.
It's slow though.

from diffusers import QwenImageTransformer2DModel
import torch
from PIL import Image
from transformers import BitsAndBytesConfig as TransformersBitsAndBytesConfig
from transformers import Qwen2_5_VLForConditionalGeneration

from diffusers import BitsAndBytesConfig as DiffusersBitsAndBytesConfig
from diffusers import QwenImageLayeredPipeline


model_id = "Qwen/Qwen-Image-Layered"
torch_dtype = torch.bfloat16

quantization_config = DiffusersBitsAndBytesConfig(
    load_in_4bit=True,
    bnb_4bit_quant_type="nf4",
    bnb_4bit_compute_dtype=torch.bfloat16,
    llm_int8_skip_modules=["transformer_blocks.0.img_mod"],
)
transformer = QwenImageTransformer2DModel.from_pretrained(
    model_id,
    subfolder="transformer",
    quantization_config=quantization_config,
    torch_dtype=torch_dtype,
)
transformer = transformer.to("cpu")

quantization_config = TransformersBitsAndBytesConfig(
    load_in_4bit=True,
    bnb_4bit_quant_type="nf4",
    bnb_4bit_compute_dtype=torch.bfloat16,
)

text_encoder = Qwen2_5_VLForConditionalGeneration.from_pretrained(
    model_id,
    subfolder="text_encoder",
    quantization_config=quantization_config,
    torch_dtype=torch_dtype,
)
text_encoder = text_encoder.to("cpu")

pipeline = QwenImageLayeredPipeline.from_pretrained(
    model_id,
    transformer=transformer,
    text_encoder=text_encoder,
    torch_dtype=torch_dtype,
)
pipeline.enable_model_cpu_offload()
pipeline.set_progress_bar_config(disable=None)

image = Image.open("workdir/bc1e03f40776b8bee006ea2b2b0d8103.webp").convert("RGBA")
inputs = {
    "image": image,
    "generator": torch.Generator(device="cuda").manual_seed(777),
    "true_cfg_scale": 4.0,
    "negative_prompt": " ",
    "num_inference_steps": 50,
    "num_images_per_prompt": 1,
    "layers": 4,
    "resolution": 640,  # Using different bucket (640, 1024) to determine the resolution. For this version, 640 is recommended
    "cfg_normalize": True,  # Whether enable cfg normalization.
    "use_en_prompt": True,  # Automatic caption language if user does not provide caption
}

with torch.inference_mode():
    output = pipeline(**inputs)
    output_image = output.images[0]

for i, image in enumerate(output_image):
    image.save(f"{i}.png")