text_to_image.py

from transformers.agents.tools import Tool
from transformers.utils import is_accelerate_available
import torch
import spaces

from diffusers import DiffusionPipeline, DPMSolverMultistepScheduler
if is_accelerate_available():
    from accelerate import PartialState

TEXT_TO_IMAGE_DESCRIPTION = (
    "This is a tool that creates an image according to a prompt."
)

@spaces.GPU
class TextToImageTool(Tool):
    default_checkpoint = "runwayml/stable-diffusion-v1-5"
    description = TEXT_TO_IMAGE_DESCRIPTION
    name = "image_generator"
    inputs = {"prompt": {"type": "string", "description": "the image description"}}
    output_type = "image"

    def __init__(self, device=None, **hub_kwargs) -> None:
        if not is_accelerate_available():
            raise ImportError("Accelerate should be installed in order to use tools.")

        super().__init__()

        self.device = device
        self.pipeline = None
        self.hub_kwargs = hub_kwargs

    def setup(self):
        if self.device is None:
            self.device = PartialState().default_device

        self.pipeline = DiffusionPipeline.from_pretrained(self.default_checkpoint)
        self.pipeline.scheduler = DPMSolverMultistepScheduler.from_config(self.pipeline.scheduler.config)
        self.pipeline.to(self.device)

        if self.device.type == "cuda":
            self.pipeline.to(torch_dtype=torch.float16)

        self.is_initialized = True

    def forward(self, prompt: str):
        if not self.is_initialized:
            self.setup()

        negative_prompt = "low quality, bad quality, deformed, low resolution"
        added_prompt = " , highest quality, highly realistic, very high resolution"

        return self.pipeline(prompt + added_prompt, negative_prompt=negative_prompt, num_inference_steps=25).images[0]