Skip to content

Bedrock stable diffusion image generation model driver

BedrockStableDiffusionImageGenerationModelDriver

Bases: BaseImageGenerationModelDriver

Image generation model driver for Stable Diffusion models on Amazon Bedrock.

For more information on all supported parameters, see the Stable Diffusion documentation: https://platform.stability.ai/docs/api-reference#tag/v1generation

Attributes:

Name Type Description
cfg_scale int

Specifies how strictly image generation follows the provided prompt. Defaults to 7.
mask_source int

Specifies mask image configuration for image-to-image generations. Defaults to "MASK_IMAGE_BLACK".
style_preset Optional[str]

If provided, specifies a specific image generation style preset.
clip_guidance_preset Optional[str]

If provided, requests a specific clip guidance preset to be used in the diffusion process.
sampler Optional[str]

If provided, requests a specific sampler to be used in the diffusion process.
steps Optional[int]

If provided, specifies the number of diffusion steps to use in the image generation.
start_schedule Optional[float]

If provided, specifies the start_schedule parameter used to determine the influence of the input image in image-to-image generation.
Source code in griptape/drivers/image_generation_model/bedrock_stable_diffusion_image_generation_model_driver.py 
 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
 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
 79
 80
 81
 82
 83
 84
 85
 86
 87
 88
 89
 90
 91
 92
 93
 94
 95
 96
 97
 98
 99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
@define
class BedrockStableDiffusionImageGenerationModelDriver(BaseImageGenerationModelDriver):
    """Image generation model driver for Stable Diffusion models on Amazon Bedrock.

    For more information on all supported parameters, see the Stable Diffusion documentation:
        https://platform.stability.ai/docs/api-reference#tag/v1generation

    Attributes:
        cfg_scale: Specifies how strictly image generation follows the provided prompt. Defaults to 7.
        mask_source: Specifies mask image configuration for image-to-image generations. Defaults to "MASK_IMAGE_BLACK".
        style_preset: If provided, specifies a specific image generation style preset.
        clip_guidance_preset: If provided, requests a specific clip guidance preset to be used in the diffusion process.
        sampler: If provided, requests a specific sampler to be used in the diffusion process.
        steps: If provided, specifies the number of diffusion steps to use in the image generation.
        start_schedule: If provided, specifies the start_schedule parameter used to determine the influence of the input
            image in image-to-image generation.
    """

    cfg_scale: int = field(default=7, kw_only=True, metadata={"serializable": True})
    style_preset: Optional[str] = field(default=None, kw_only=True, metadata={"serializable": True})
    clip_guidance_preset: Optional[str] = field(default=None, kw_only=True, metadata={"serializable": True})
    sampler: Optional[str] = field(default=None, kw_only=True, metadata={"serializable": True})
    steps: Optional[int] = field(default=None, kw_only=True, metadata={"serializable": True})
    start_schedule: Optional[float] = field(default=None, kw_only=True, metadata={"serializable": True})

    def text_to_image_request_parameters(
        self,
        prompts: list[str],
        image_width: int,
        image_height: int,
        negative_prompts: Optional[list[str]] = None,
        seed: Optional[int] = None,
    ) -> dict:
        return self._request_parameters(
            prompts, width=image_width, height=image_height, negative_prompts=negative_prompts, seed=seed
        )

    def image_variation_request_parameters(
        self,
        prompts: list[str],
        image: ImageArtifact,
        negative_prompts: Optional[list[str]] = None,
        seed: Optional[int] = None,
    ) -> dict:
        return self._request_parameters(prompts, image=image, negative_prompts=negative_prompts, seed=seed)

    def image_inpainting_request_parameters(
        self,
        prompts: list[str],
        image: ImageArtifact,
        mask: ImageArtifact,
        negative_prompts: Optional[list[str]] = None,
        seed: Optional[int] = None,
    ) -> dict:
        return self._request_parameters(
            prompts,
            image=image,
            mask=mask,
            mask_source="MASK_IMAGE_BLACK",
            negative_prompts=negative_prompts,
            seed=seed,
        )

    def image_outpainting_request_parameters(
        self,
        prompts: list[str],
        image: ImageArtifact,
        mask: ImageArtifact,
        negative_prompts: Optional[list[str]] = None,
        seed: Optional[int] = None,
    ) -> dict:
        return self._request_parameters(
            prompts,
            image=image,
            mask=mask,
            mask_source="MASK_IMAGE_WHITE",
            negative_prompts=negative_prompts,
            seed=seed,
        )

    def _request_parameters(
        self,
        prompts: list[str],
        width: Optional[int] = None,
        height: Optional[int] = None,
        image: Optional[ImageArtifact] = None,
        mask: Optional[ImageArtifact] = None,
        negative_prompts: Optional[list[str]] = None,
        seed: Optional[int] = None,
        mask_source: Optional[str] = None,
    ) -> dict:
        if negative_prompts is None:
            negative_prompts = []

        text_prompts = [{"text": prompt, "weight": 1.0} for prompt in prompts]
        text_prompts += [{"text": negative_prompt, "weight": -1.0} for negative_prompt in negative_prompts]

        request = {"text_prompts": text_prompts, "cfg_scale": self.cfg_scale}

        if self.style_preset is not None:
            request["style_preset"] = self.style_preset

        if self.clip_guidance_preset is not None:
            request["clip_guidance_preset"] = self.clip_guidance_preset

        if self.sampler is not None:
            request["sampler"] = self.sampler

        if image is not None:
            request["init_image"] = image.base64
            request["width"] = image.width
            request["height"] = image.height
        else:
            request["width"] = width
            request["height"] = height

        if self.steps is not None:
            request["steps"] = self.steps

        if seed is not None:
            request["seed"] = seed

        if mask is not None:
            if not mask_source:
                raise ValueError("mask_source must be provided when mask is provided")

            request["mask_source"] = mask_source
            request["mask_image"] = mask.base64

        if self.start_schedule is not None:
            request["start_schedule"] = self.start_schedule

        return request

    def get_generated_image(self, response: dict) -> bytes:
        image_response = response["artifacts"][0]

        # finishReason may be SUCCESS, CONTENT_FILTERED, or ERROR.
        if image_response.get("finishReason") == "ERROR":
            raise Exception(f"Image generation failed: {image_response.get('finishReason')}")
        elif image_response.get("finishReason") == "CONTENT_FILTERED":
            logging.warning(f"Image generation triggered content filter and may be blurred")

        return base64.decodebytes(bytes(image_response.get("base64"), "utf-8"))

cfg_scale: int = field(default=7, kw_only=True, metadata={'serializable': True}) class-attribute instance-attribute

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

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

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

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

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

get_generated_image(response)

Source code in griptape/drivers/image_generation_model/bedrock_stable_diffusion_image_generation_model_driver.py 
147
148
149
150
151
152
153
154
155
156
def get_generated_image(self, response: dict) -> bytes:
    image_response = response["artifacts"][0]

    # finishReason may be SUCCESS, CONTENT_FILTERED, or ERROR.
    if image_response.get("finishReason") == "ERROR":
        raise Exception(f"Image generation failed: {image_response.get('finishReason')}")
    elif image_response.get("finishReason") == "CONTENT_FILTERED":
        logging.warning(f"Image generation triggered content filter and may be blurred")

    return base64.decodebytes(bytes(image_response.get("base64"), "utf-8"))

image_inpainting_request_parameters(prompts, image, mask, negative_prompts=None, seed=None)

Source code in griptape/drivers/image_generation_model/bedrock_stable_diffusion_image_generation_model_driver.py 
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
def image_inpainting_request_parameters(
    self,
    prompts: list[str],
    image: ImageArtifact,
    mask: ImageArtifact,
    negative_prompts: Optional[list[str]] = None,
    seed: Optional[int] = None,
) -> dict:
    return self._request_parameters(
        prompts,
        image=image,
        mask=mask,
        mask_source="MASK_IMAGE_BLACK",
        negative_prompts=negative_prompts,
        seed=seed,
    )

image_outpainting_request_parameters(prompts, image, mask, negative_prompts=None, seed=None)

Source code in griptape/drivers/image_generation_model/bedrock_stable_diffusion_image_generation_model_driver.py 
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
def image_outpainting_request_parameters(
    self,
    prompts: list[str],
    image: ImageArtifact,
    mask: ImageArtifact,
    negative_prompts: Optional[list[str]] = None,
    seed: Optional[int] = None,
) -> dict:
    return self._request_parameters(
        prompts,
        image=image,
        mask=mask,
        mask_source="MASK_IMAGE_WHITE",
        negative_prompts=negative_prompts,
        seed=seed,
    )

image_variation_request_parameters(prompts, image, negative_prompts=None, seed=None)

Source code in griptape/drivers/image_generation_model/bedrock_stable_diffusion_image_generation_model_driver.py 
50
51
52
53
54
55
56
57
def image_variation_request_parameters(
    self,
    prompts: list[str],
    image: ImageArtifact,
    negative_prompts: Optional[list[str]] = None,
    seed: Optional[int] = None,
) -> dict:
    return self._request_parameters(prompts, image=image, negative_prompts=negative_prompts, seed=seed)

text_to_image_request_parameters(prompts, image_width, image_height, negative_prompts=None, seed=None)

Source code in griptape/drivers/image_generation_model/bedrock_stable_diffusion_image_generation_model_driver.py 
38
39
40
41
42
43
44
45
46
47
48
def text_to_image_request_parameters(
    self,
    prompts: list[str],
    image_width: int,
    image_height: int,
    negative_prompts: Optional[list[str]] = None,
    seed: Optional[int] = None,
) -> dict:
    return self._request_parameters(
        prompts, width=image_width, height=image_height, negative_prompts=negative_prompts, seed=seed
    )