File size: 1,788 Bytes
e147359
912d9da
936baa4
393b5eb
927b9ac
33debb6
 
393b5eb
 
5109976
393b5eb
 
 
 
 
 
12e8eec
e8e18da
5109976
393b5eb
 
33debb6
393b5eb
 
 
 
 
 
 
 
 
 
02b1331
393b5eb
 
936baa4
393b5eb
 
936baa4
 
 
393b5eb
 
1b6be6f
393b5eb
 
 
625c254
 
 
 
393b5eb
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
from transformers.agents.tools import Tool
from transformers.utils import is_accelerate_available
import torch

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."
)


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]