Skip to content

stable_diffusion_3_img_2_img_image_generation_pipeline_driver

StableDiffusion3Img2ImgImageGenerationPipelineDriver

Bases: StableDiffusion3ImageGenerationPipelineDriver

Image generation model driver for Stable Diffusion 3 model image to image pipelines.

For more information, see the HuggingFace documentation for the StableDiffusion3Img2ImgPipeline: https://github.com/huggingface/diffusers/blob/main/src/diffusers/pipelines/stable_diffusion_3/pipeline_stable_diffusion_3_img2img.py

Attributes:

Name Type Description
strength Optional[float]

A value [0.0, 1.0] that determines the strength of the initial image in the output.
Source code in griptape/drivers/image_generation_pipeline/stable_diffusion_3_img_2_img_image_generation_pipeline_driver.py 
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
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
@define
class StableDiffusion3Img2ImgImageGenerationPipelineDriver(StableDiffusion3ImageGenerationPipelineDriver):
    """Image generation model driver for Stable Diffusion 3 model image to image pipelines.

    For more information, see the HuggingFace documentation for the StableDiffusion3Img2ImgPipeline:
        https://github.com/huggingface/diffusers/blob/main/src/diffusers/pipelines/stable_diffusion_3/pipeline_stable_diffusion_3_img2img.py

    Attributes:
        strength: A value [0.0, 1.0] that determines the strength of the initial image in the output.
    """

    strength: Optional[float] = field(default=None, kw_only=True, metadata={"serializable": True})

    def prepare_pipeline(self, model: str, device: Optional[str]) -> Any:
        sd3_img2img_pipeline = import_optional_dependency(
            "diffusers.pipelines.stable_diffusion_3.pipeline_stable_diffusion_3_img2img"
        ).StableDiffusion3Img2ImgPipeline

        pipeline_params = {}
        if self.torch_dtype is not None:
            pipeline_params["torch_dtype"] = self.torch_dtype

        if self.drop_t5_encoder:
            pipeline_params["text_encoder_3"] = None
            pipeline_params["tokenizer_3"] = None

        # A model can be provided either as a path to a local file
        # or as a HuggingFace model repo name.
        if os.path.isfile(model):
            # If the model provided is a local file (not a directory),
            # we load it using the from_single_file method.
            pipeline = sd3_img2img_pipeline.from_single_file(model, **pipeline_params)
        else:
            # If the model is a local directory or hosted on HuggingFace,
            # we load it using the from_pretrained method.
            pipeline = sd3_img2img_pipeline.from_pretrained(model, **pipeline_params)

        if self.enable_model_cpu_offload:
            pipeline.enable_model_cpu_offload()

        # Move inference to particular device if requested.
        if device is not None:
            pipeline.to(device)

        return pipeline

    def make_image_param(self, image: Optional[Image]) -> Optional[dict[str, Image]]:
        if image is None:
            raise ValueError("Input image is required for image to image pipelines.")

        return {"image": image}

    def make_additional_params(self, negative_prompts: Optional[list[str]], device: Optional[str]) -> dict[str, Any]:
        additional_params = super().make_additional_params(negative_prompts, device)

        # Explicit height and width params are not supported, but
        # are instead inferred from input image.
        del additional_params["height"]
        del additional_params["width"]

        if self.strength is not None:
            additional_params["strength"] = self.strength

        return additional_params

strength: Optional[float] = field(default=None, kw_only=True, metadata={'serializable': True}) class-attribute instance-attribute

make_additional_params(negative_prompts, device)

Source code in griptape/drivers/image_generation_pipeline/stable_diffusion_3_img_2_img_image_generation_pipeline_driver.py 
67
68
69
70
71
72
73
74
75
76
77
78
def make_additional_params(self, negative_prompts: Optional[list[str]], device: Optional[str]) -> dict[str, Any]:
    additional_params = super().make_additional_params(negative_prompts, device)

    # Explicit height and width params are not supported, but
    # are instead inferred from input image.
    del additional_params["height"]
    del additional_params["width"]

    if self.strength is not None:
        additional_params["strength"] = self.strength

    return additional_params

make_image_param(image)

Source code in griptape/drivers/image_generation_pipeline/stable_diffusion_3_img_2_img_image_generation_pipeline_driver.py 
61
62
63
64
65
def make_image_param(self, image: Optional[Image]) -> Optional[dict[str, Image]]:
    if image is None:
        raise ValueError("Input image is required for image to image pipelines.")

    return {"image": image}

prepare_pipeline(model, device)

Source code in griptape/drivers/image_generation_pipeline/stable_diffusion_3_img_2_img_image_generation_pipeline_driver.py 
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
54
55
56
57
58
59
def prepare_pipeline(self, model: str, device: Optional[str]) -> Any:
    sd3_img2img_pipeline = import_optional_dependency(
        "diffusers.pipelines.stable_diffusion_3.pipeline_stable_diffusion_3_img2img"
    ).StableDiffusion3Img2ImgPipeline

    pipeline_params = {}
    if self.torch_dtype is not None:
        pipeline_params["torch_dtype"] = self.torch_dtype

    if self.drop_t5_encoder:
        pipeline_params["text_encoder_3"] = None
        pipeline_params["tokenizer_3"] = None

    # A model can be provided either as a path to a local file
    # or as a HuggingFace model repo name.
    if os.path.isfile(model):
        # If the model provided is a local file (not a directory),
        # we load it using the from_single_file method.
        pipeline = sd3_img2img_pipeline.from_single_file(model, **pipeline_params)
    else:
        # If the model is a local directory or hosted on HuggingFace,
        # we load it using the from_pretrained method.
        pipeline = sd3_img2img_pipeline.from_pretrained(model, **pipeline_params)

    if self.enable_model_cpu_offload:
        pipeline.enable_model_cpu_offload()

    # Move inference to particular device if requested.
    if device is not None:
        pipeline.to(device)

    return pipeline