Assets
Duplicate
List Snapshots
Retrieve
Unlock
Update
Update Tags
ModelsExpand Collapse
class AssetDuplicateResponse: …
asset: Asset
edit_capabilities: List[Literal["DETECTION", "GENERATIVE_FILL", "PIXELATE", 8 more]]List of edit capabilities
List of edit capabilities
metadata: AssetMetadataMetadata of the asset with some additional information
Metadata of the asset with some additional information
type: Literal["3d-texture", "3d-texture-albedo", "3d-texture-metallic", 77 more]The type of the asset. Ex: ‘inference-txt2img’ will represent an asset generated from a text to image model
The type of the asset. Ex: ‘inference-txt2img’ will represent an asset generated from a text to image model
How angular is the surface? 0 is like a sphere, 1 is like a mechanical object
The optional aspect ratio given for the generation, only applicable for some models
A bounding box around the object of interest, in the format [x1, y1, x2, y2].
The control image already processed by canny detector. Must reference an existing AssetId.
Ensure upscaled tile have the same color histogram as original tile.
concepts: Optional[List[AssetMetadataConcept]]Flux Kontext LoRA to style the image.
For Flux Kontext Prompt Editing.
Flux Kontext LoRA to style the image. For Flux Kontext Prompt Editing.
Allow the generation of “hallucinations” during the upscale process, which adds additional details and deviates from the original image. Default: optimized for your preset and style.
Amount of decay in creativity over the upscale process. The lowest the value, the less the creativity will be preserved over the upscale process.
The control image processed by depth estimator. Must reference an existing AssetId.
The name its parent’s files reference it by (an OBJ’s MTL, an MTL’s texture), kept as delivered.
Determine the scale at which the upscale process works.
- With a small value, the upscale works at the largest scale, resulting in fewer added details and more coherent images. Ideal for portraits, for example.
- With a large value, the upscale works at the smallest scale, resulting in more added details and more hallucinations. Ideal for landscapes, for example.
(info): A small value is slower and more expensive to run.
Apply extra control to the Skybox 360 geometry. The higher the value, the more the 360 geometry will influence the generated skybox image.
Use with caution. Default is adapted to the other parameters.
The input image to process. Must reference an existing AssetId or be a data URL.
Strengthen the similarity to the original image during the upscale. Default: optimized for your preset and style.
image_type: Optional[Literal["seamfull", "skybox", "texture"]]Preserve the seamless properties of skybox or texture images. Input has to be of same type (seamless).
Preserve the seamless properties of skybox or texture images. Input has to be of same type (seamless).
input_fidelity: Optional[Literal["high", "low"]]When set to high, allows to better preserve details from the input images in the output.
This is especially useful when using images that contain elements like faces or logos that
require accurate preservation in the generated image.
You can provide multiple input images that will all be preserved with high fidelity, but keep
in mind that the first image will be preserved with richer textures and finer details, so if
you include elements such as faces, consider placing them in the first image.
Only available for the gpt-image-1 model.
When set to high, allows to better preserve details from the input images in the output.
This is especially useful when using images that contain elements like faces or logos that
require accurate preservation in the generated image.
You can provide multiple input images that will all be preserved with high fidelity, but keep in mind that the first image will be preserved with richer textures and finer details, so if you include elements such as faces, consider placing them in the first image.
Only available for the gpt-image-1 model.
input_location: Optional[Literal["bottom", "left", "middle", 2 more]]Location of the input image in the output.
Location of the input image in the output.
Its own references (an OBJ’s mtllib, an MTL’s maps) were kept as delivered.
How polished is the surface? 0 is like a rough surface, 1 is like a mirror
model_type: Optional[Literal["custom", "elevenlabs-voice", "flux.1", 28 more]]The type of the generator used
The type of the generator used
Controls the influence of the negative prompt. Default 0 means the negative prompt has no effect. Higher values increase negative prompt influence. Must be > 0 if negativePrompt is provided.
The number of denoising steps for each image generation.
Override the embeddings of the model. Only your prompt and negativePrompt will be used. Use with caution.
List of points (label, x, y) in the image where label = 0 for background and 1 for object.
How polished is the surface? 0 is like a rough surface, 1 is like a mirror
Increase the fidelity to the prompt during upscale. Default: optimized for your preset and style.
How raised is the surface? 0 is flat like water, 1 is like a very rough rock
Additional refinement steps before scaling.
If scalingFactor == 1, the refinement process will be applied (1 + refinementSteps) times. If scalingFactor > 1, the refinement process will be applied refinementSteps times.
Boolean to return the source image with the mask applied as alpha channel
(RGBA PNG, transparent background) instead of the binary mask.
Mutually exclusive with resultMask — passing both as true returns a 400 error.
Note: backgroundOpacity is no longer honored — alpha is binary (0/255) only.
Boolean to return the binary masks in the response.
Mutually exclusive with resultImage — passing both as true returns a 400 error.
Save a flipbook of the texture. Deactivated when the input texture is larger than 2048x2048px
The seed used to generate this asset. <!> Can be a string or a number in some cases <!>.
How shiny is the surface? 0 is like a matte surface, 1 is like a diamond
Strength for the input image structure preservation
The control image for structure. A canny detector will be applied to this image. Must reference an existing AssetId.
The higher the value the more it will look like the style image(s)
List of style images. Most of the time, only one image is enough. It must be existing AssetIds.
Condition the influence of the style image(s). The higher the value, the more the style images will influence the upscaled image.
Target width for the upscaled image, take priority over scaling factor
The owner (project) ID (example: “proj_23tlk332lkht3kl2” or “team_dlkhgs23tlk3hlkth32lkht3kl2” for old teams)
properties: AssetPropertiesThe properties of the asset, content may depend on the kind of asset returned
The properties of the asset, content may depend on the kind of asset returned
camera_mode: Optional[Literal["aerial", "interior", "turntable"]]Recommended preview camera mode emitted by the generator (gsplat route);
passed through to mesh-rendering verbatim.
Recommended preview camera mode emitted by the generator (gsplat route); passed through to mesh-rendering verbatim.
camera_trajectory: Optional[List[AssetPropertiesCameraTrajectory]]Recommended preview camera path emitted by the generator (gsplat route);
passed through to mesh-rendering verbatim.
Recommended preview camera path emitted by the generator (gsplat route); passed through to mesh-rendering verbatim.
Number of Unicode code points in the text. Code-point-aware (so a non-BMP emoji counts as 1) but not full grapheme-cluster aware (a ZWJ sequence still counts as several).
classification: Optional[Literal["effect", "interview", "music", 5 more]]Classification of the audio
Classification of the audio
True when preview holds the entire content unmodified — consumers can
use it directly without fetching asset.url. False or undefined means
the content exceeds the preview budget and consumers must fetch the
full body from S3 to read past the preview.
Leading slice of the content used for inline UI display and as a search shortcut. Capped at TEXT_PREVIEW_MAX_BYTES (UTF-8) and always cut on a code-point boundary so no character is split. Number of characters in the preview varies by script (around 1024 for ASCII, ~340 for CJK, ~256 for emoji-heavy text at the default 1 KB budget).
source: Literal["3d23d", "3d23d:texture", "3d:texture", 77 more]source of the asset
source of the asset
The description, it will contain in priority:
- the manual description
- the advanced captioning when the asset is used in training flow
- the automatic captioning
The embedding of the asset when requested.
Only available when an asset can be embedded (ie: not Detection maps)
first_frame: Optional[AssetFirstFrame]The video asset’s first frame.
Contains the assetId and the url of the first frame.
The video asset’s first frame.
Contains the assetId and the url of the first frame.
last_frame: Optional[AssetLastFrame]The video asset’s last frame.
Contains the assetId and the url of the last frame.
The video asset’s last frame.
Contains the assetId and the url of the last frame.
The original file url.
Contains the url of the original file. without any conversion. Only available for some specific video, audio and threeD assets. Is only specified if the given asset data has been replaced with a new file during the creation of the asset.
original_mime_type: Optional[Literal["application/gzip", "application/json", "application/octet-stream", 47 more]]The mime type of the original file url.
The original file is stored under a key that carries no extension, so this
is the only way to know which format originalFileUrl serves (e.g. an HDRi
skybox exposes a JPEG preview as url and its .exr or .hdr file as the
original). Omitted when the stored format could not be determined.
The mime type of the original file url.
The original file is stored under a key that carries no extension, so this
is the only way to know which format originalFileUrl serves (e.g. an HDRi
skybox exposes a JPEG preview as url and its .exr or .hdr file as the
original). Omitted when the stored format could not be determined.
The output index of the asset within a job This index is an positive integer that starts at 0 It is used to differentiate between multiple outputs of the same job If the job has only one output, this index is 0
preview: Optional[AssetPreview]The asset’s preview.
Contains the assetId and the url of the preview.
The asset’s preview.
Contains the assetId and the url of the preview.
quality_gate: Optional[AssetQualityGate]Quality Gate result, computed asynchronously after generation.
details (the per-dimension reasons and suggestions) is returned only when
the asset is read with withQualityGateDetails=true.
Quality Gate result, computed asynchronously after generation.
details (the per-dimension reasons and suggestions) is returned only when
the asset is read with withQualityGateDetails=true.
details: Optional[AssetQualityGateDetails]Per-dimension reasons and suggestions, as produced by the detector.
Returned only when the asset is read with withQualityGateDetails=true by
the project that owns it — the brief-compliance text paraphrases that
project’s brief, so a cross-project read of a public asset gets the summary
alone. Also omitted when the archived verdict is unavailable (asset scored
by an older version, or an unreadable archive).
Per-dimension reasons and suggestions, as produced by the detector.
Returned only when the asset is read with withQualityGateDetails=true by
the project that owns it — the brief-compliance text paraphrases that
project’s brief, so a cross-project read of a public asset gets the summary
alone. Also omitted when the archived verdict is unavailable (asset scored
by an older version, or an unreadable archive).
class AssetGetBulkResponse: …
assets: List[Asset]
edit_capabilities: List[Literal["DETECTION", "GENERATIVE_FILL", "PIXELATE", 8 more]]List of edit capabilities
List of edit capabilities
metadata: AssetMetadataMetadata of the asset with some additional information
Metadata of the asset with some additional information
type: Literal["3d-texture", "3d-texture-albedo", "3d-texture-metallic", 77 more]The type of the asset. Ex: ‘inference-txt2img’ will represent an asset generated from a text to image model
The type of the asset. Ex: ‘inference-txt2img’ will represent an asset generated from a text to image model
How angular is the surface? 0 is like a sphere, 1 is like a mechanical object
The optional aspect ratio given for the generation, only applicable for some models
A bounding box around the object of interest, in the format [x1, y1, x2, y2].
The control image already processed by canny detector. Must reference an existing AssetId.
Ensure upscaled tile have the same color histogram as original tile.
concepts: Optional[List[AssetMetadataConcept]]Flux Kontext LoRA to style the image.
For Flux Kontext Prompt Editing.
Flux Kontext LoRA to style the image. For Flux Kontext Prompt Editing.
Allow the generation of “hallucinations” during the upscale process, which adds additional details and deviates from the original image. Default: optimized for your preset and style.
Amount of decay in creativity over the upscale process. The lowest the value, the less the creativity will be preserved over the upscale process.
The control image processed by depth estimator. Must reference an existing AssetId.
The name its parent’s files reference it by (an OBJ’s MTL, an MTL’s texture), kept as delivered.
Determine the scale at which the upscale process works.
- With a small value, the upscale works at the largest scale, resulting in fewer added details and more coherent images. Ideal for portraits, for example.
- With a large value, the upscale works at the smallest scale, resulting in more added details and more hallucinations. Ideal for landscapes, for example.
(info): A small value is slower and more expensive to run.
Apply extra control to the Skybox 360 geometry. The higher the value, the more the 360 geometry will influence the generated skybox image.
Use with caution. Default is adapted to the other parameters.
The input image to process. Must reference an existing AssetId or be a data URL.
Strengthen the similarity to the original image during the upscale. Default: optimized for your preset and style.
image_type: Optional[Literal["seamfull", "skybox", "texture"]]Preserve the seamless properties of skybox or texture images. Input has to be of same type (seamless).
Preserve the seamless properties of skybox or texture images. Input has to be of same type (seamless).
input_fidelity: Optional[Literal["high", "low"]]When set to high, allows to better preserve details from the input images in the output.
This is especially useful when using images that contain elements like faces or logos that
require accurate preservation in the generated image.
You can provide multiple input images that will all be preserved with high fidelity, but keep
in mind that the first image will be preserved with richer textures and finer details, so if
you include elements such as faces, consider placing them in the first image.
Only available for the gpt-image-1 model.
When set to high, allows to better preserve details from the input images in the output.
This is especially useful when using images that contain elements like faces or logos that
require accurate preservation in the generated image.
You can provide multiple input images that will all be preserved with high fidelity, but keep in mind that the first image will be preserved with richer textures and finer details, so if you include elements such as faces, consider placing them in the first image.
Only available for the gpt-image-1 model.
input_location: Optional[Literal["bottom", "left", "middle", 2 more]]Location of the input image in the output.
Location of the input image in the output.
Its own references (an OBJ’s mtllib, an MTL’s maps) were kept as delivered.
How polished is the surface? 0 is like a rough surface, 1 is like a mirror
model_type: Optional[Literal["custom", "elevenlabs-voice", "flux.1", 28 more]]The type of the generator used
The type of the generator used
Controls the influence of the negative prompt. Default 0 means the negative prompt has no effect. Higher values increase negative prompt influence. Must be > 0 if negativePrompt is provided.
The number of denoising steps for each image generation.
Override the embeddings of the model. Only your prompt and negativePrompt will be used. Use with caution.
List of points (label, x, y) in the image where label = 0 for background and 1 for object.
How polished is the surface? 0 is like a rough surface, 1 is like a mirror
Increase the fidelity to the prompt during upscale. Default: optimized for your preset and style.
How raised is the surface? 0 is flat like water, 1 is like a very rough rock
Additional refinement steps before scaling.
If scalingFactor == 1, the refinement process will be applied (1 + refinementSteps) times. If scalingFactor > 1, the refinement process will be applied refinementSteps times.
Boolean to return the source image with the mask applied as alpha channel
(RGBA PNG, transparent background) instead of the binary mask.
Mutually exclusive with resultMask — passing both as true returns a 400 error.
Note: backgroundOpacity is no longer honored — alpha is binary (0/255) only.
Boolean to return the binary masks in the response.
Mutually exclusive with resultImage — passing both as true returns a 400 error.
Save a flipbook of the texture. Deactivated when the input texture is larger than 2048x2048px
The seed used to generate this asset. <!> Can be a string or a number in some cases <!>.
How shiny is the surface? 0 is like a matte surface, 1 is like a diamond
Strength for the input image structure preservation
The control image for structure. A canny detector will be applied to this image. Must reference an existing AssetId.
The higher the value the more it will look like the style image(s)
List of style images. Most of the time, only one image is enough. It must be existing AssetIds.
Condition the influence of the style image(s). The higher the value, the more the style images will influence the upscaled image.
Target width for the upscaled image, take priority over scaling factor
The owner (project) ID (example: “proj_23tlk332lkht3kl2” or “team_dlkhgs23tlk3hlkth32lkht3kl2” for old teams)
properties: AssetPropertiesThe properties of the asset, content may depend on the kind of asset returned
The properties of the asset, content may depend on the kind of asset returned
camera_mode: Optional[Literal["aerial", "interior", "turntable"]]Recommended preview camera mode emitted by the generator (gsplat route);
passed through to mesh-rendering verbatim.
Recommended preview camera mode emitted by the generator (gsplat route); passed through to mesh-rendering verbatim.
camera_trajectory: Optional[List[AssetPropertiesCameraTrajectory]]Recommended preview camera path emitted by the generator (gsplat route);
passed through to mesh-rendering verbatim.
Recommended preview camera path emitted by the generator (gsplat route); passed through to mesh-rendering verbatim.
Number of Unicode code points in the text. Code-point-aware (so a non-BMP emoji counts as 1) but not full grapheme-cluster aware (a ZWJ sequence still counts as several).
classification: Optional[Literal["effect", "interview", "music", 5 more]]Classification of the audio
Classification of the audio
True when preview holds the entire content unmodified — consumers can
use it directly without fetching asset.url. False or undefined means
the content exceeds the preview budget and consumers must fetch the
full body from S3 to read past the preview.
Leading slice of the content used for inline UI display and as a search shortcut. Capped at TEXT_PREVIEW_MAX_BYTES (UTF-8) and always cut on a code-point boundary so no character is split. Number of characters in the preview varies by script (around 1024 for ASCII, ~340 for CJK, ~256 for emoji-heavy text at the default 1 KB budget).
source: Literal["3d23d", "3d23d:texture", "3d:texture", 77 more]source of the asset
source of the asset
The description, it will contain in priority:
- the manual description
- the advanced captioning when the asset is used in training flow
- the automatic captioning
The embedding of the asset when requested.
Only available when an asset can be embedded (ie: not Detection maps)
first_frame: Optional[AssetFirstFrame]The video asset’s first frame.
Contains the assetId and the url of the first frame.
The video asset’s first frame.
Contains the assetId and the url of the first frame.
last_frame: Optional[AssetLastFrame]The video asset’s last frame.
Contains the assetId and the url of the last frame.
The video asset’s last frame.
Contains the assetId and the url of the last frame.
The original file url.
Contains the url of the original file. without any conversion. Only available for some specific video, audio and threeD assets. Is only specified if the given asset data has been replaced with a new file during the creation of the asset.
original_mime_type: Optional[Literal["application/gzip", "application/json", "application/octet-stream", 47 more]]The mime type of the original file url.
The original file is stored under a key that carries no extension, so this
is the only way to know which format originalFileUrl serves (e.g. an HDRi
skybox exposes a JPEG preview as url and its .exr or .hdr file as the
original). Omitted when the stored format could not be determined.
The mime type of the original file url.
The original file is stored under a key that carries no extension, so this
is the only way to know which format originalFileUrl serves (e.g. an HDRi
skybox exposes a JPEG preview as url and its .exr or .hdr file as the
original). Omitted when the stored format could not be determined.
The output index of the asset within a job This index is an positive integer that starts at 0 It is used to differentiate between multiple outputs of the same job If the job has only one output, this index is 0
preview: Optional[AssetPreview]The asset’s preview.
Contains the assetId and the url of the preview.
The asset’s preview.
Contains the assetId and the url of the preview.
quality_gate: Optional[AssetQualityGate]Quality Gate result, computed asynchronously after generation.
details (the per-dimension reasons and suggestions) is returned only when
the asset is read with withQualityGateDetails=true.
Quality Gate result, computed asynchronously after generation.
details (the per-dimension reasons and suggestions) is returned only when
the asset is read with withQualityGateDetails=true.
details: Optional[AssetQualityGateDetails]Per-dimension reasons and suggestions, as produced by the detector.
Returned only when the asset is read with withQualityGateDetails=true by
the project that owns it — the brief-compliance text paraphrases that
project’s brief, so a cross-project read of a public asset gets the summary
alone. Also omitted when the archived verdict is unavailable (asset scored
by an older version, or an unreadable archive).
Per-dimension reasons and suggestions, as produced by the detector.
Returned only when the asset is read with withQualityGateDetails=true by
the project that owns it — the brief-compliance text paraphrases that
project’s brief, so a cross-project read of a public asset gets the summary
alone. Also omitted when the archived verdict is unavailable (asset scored
by an older version, or an unreadable archive).
class AssetListResponse: …
edit_capabilities: List[Literal["DETECTION", "GENERATIVE_FILL", "PIXELATE", 8 more]]List of edit capabilities
List of edit capabilities
metadata: MetadataMetadata of the asset with some additional information
Metadata of the asset with some additional information
type: Literal["3d-texture", "3d-texture-albedo", "3d-texture-metallic", 77 more]The type of the asset. Ex: ‘inference-txt2img’ will represent an asset generated from a text to image model
The type of the asset. Ex: ‘inference-txt2img’ will represent an asset generated from a text to image model
How angular is the surface? 0 is like a sphere, 1 is like a mechanical object
The optional aspect ratio given for the generation, only applicable for some models
A bounding box around the object of interest, in the format [x1, y1, x2, y2].
The control image already processed by canny detector. Must reference an existing AssetId.
Ensure upscaled tile have the same color histogram as original tile.
concepts: Optional[List[MetadataConcept]]Flux Kontext LoRA to style the image.
For Flux Kontext Prompt Editing.
Flux Kontext LoRA to style the image. For Flux Kontext Prompt Editing.
Allow the generation of “hallucinations” during the upscale process, which adds additional details and deviates from the original image. Default: optimized for your preset and style.
Amount of decay in creativity over the upscale process. The lowest the value, the less the creativity will be preserved over the upscale process.
The control image processed by depth estimator. Must reference an existing AssetId.
The name its parent’s files reference it by (an OBJ’s MTL, an MTL’s texture), kept as delivered.
Determine the scale at which the upscale process works.
- With a small value, the upscale works at the largest scale, resulting in fewer added details and more coherent images. Ideal for portraits, for example.
- With a large value, the upscale works at the smallest scale, resulting in more added details and more hallucinations. Ideal for landscapes, for example.
(info): A small value is slower and more expensive to run.
Apply extra control to the Skybox 360 geometry. The higher the value, the more the 360 geometry will influence the generated skybox image.
Use with caution. Default is adapted to the other parameters.
The input image to process. Must reference an existing AssetId or be a data URL.
Strengthen the similarity to the original image during the upscale. Default: optimized for your preset and style.
image_type: Optional[Literal["seamfull", "skybox", "texture"]]Preserve the seamless properties of skybox or texture images. Input has to be of same type (seamless).
Preserve the seamless properties of skybox or texture images. Input has to be of same type (seamless).
input_fidelity: Optional[Literal["high", "low"]]When set to high, allows to better preserve details from the input images in the output.
This is especially useful when using images that contain elements like faces or logos that
require accurate preservation in the generated image.
You can provide multiple input images that will all be preserved with high fidelity, but keep
in mind that the first image will be preserved with richer textures and finer details, so if
you include elements such as faces, consider placing them in the first image.
Only available for the gpt-image-1 model.
When set to high, allows to better preserve details from the input images in the output.
This is especially useful when using images that contain elements like faces or logos that
require accurate preservation in the generated image.
You can provide multiple input images that will all be preserved with high fidelity, but keep in mind that the first image will be preserved with richer textures and finer details, so if you include elements such as faces, consider placing them in the first image.
Only available for the gpt-image-1 model.
input_location: Optional[Literal["bottom", "left", "middle", 2 more]]Location of the input image in the output.
Location of the input image in the output.
Its own references (an OBJ’s mtllib, an MTL’s maps) were kept as delivered.
How polished is the surface? 0 is like a rough surface, 1 is like a mirror
model_type: Optional[Literal["custom", "elevenlabs-voice", "flux.1", 28 more]]The type of the generator used
The type of the generator used
Controls the influence of the negative prompt. Default 0 means the negative prompt has no effect. Higher values increase negative prompt influence. Must be > 0 if negativePrompt is provided.
The number of denoising steps for each image generation.
Override the embeddings of the model. Only your prompt and negativePrompt will be used. Use with caution.
List of points (label, x, y) in the image where label = 0 for background and 1 for object.
How polished is the surface? 0 is like a rough surface, 1 is like a mirror
Increase the fidelity to the prompt during upscale. Default: optimized for your preset and style.
How raised is the surface? 0 is flat like water, 1 is like a very rough rock
Additional refinement steps before scaling.
If scalingFactor == 1, the refinement process will be applied (1 + refinementSteps) times. If scalingFactor > 1, the refinement process will be applied refinementSteps times.
Boolean to return the source image with the mask applied as alpha channel
(RGBA PNG, transparent background) instead of the binary mask.
Mutually exclusive with resultMask — passing both as true returns a 400 error.
Note: backgroundOpacity is no longer honored — alpha is binary (0/255) only.
Boolean to return the binary masks in the response.
Mutually exclusive with resultImage — passing both as true returns a 400 error.
Save a flipbook of the texture. Deactivated when the input texture is larger than 2048x2048px
The seed used to generate this asset. <!> Can be a string or a number in some cases <!>.
How shiny is the surface? 0 is like a matte surface, 1 is like a diamond
Strength for the input image structure preservation
The control image for structure. A canny detector will be applied to this image. Must reference an existing AssetId.
The higher the value the more it will look like the style image(s)
List of style images. Most of the time, only one image is enough. It must be existing AssetIds.
Condition the influence of the style image(s). The higher the value, the more the style images will influence the upscaled image.
Target width for the upscaled image, take priority over scaling factor
The owner (project) ID (example: “proj_23tlk332lkht3kl2” or “team_dlkhgs23tlk3hlkth32lkht3kl2” for old teams)
properties: PropertiesThe properties of the asset, content may depend on the kind of asset returned
The properties of the asset, content may depend on the kind of asset returned
camera_mode: Optional[Literal["aerial", "interior", "turntable"]]Recommended preview camera mode emitted by the generator (gsplat route);
passed through to mesh-rendering verbatim.
Recommended preview camera mode emitted by the generator (gsplat route); passed through to mesh-rendering verbatim.
camera_trajectory: Optional[List[PropertiesCameraTrajectory]]Recommended preview camera path emitted by the generator (gsplat route);
passed through to mesh-rendering verbatim.
Recommended preview camera path emitted by the generator (gsplat route); passed through to mesh-rendering verbatim.
Number of Unicode code points in the text. Code-point-aware (so a non-BMP emoji counts as 1) but not full grapheme-cluster aware (a ZWJ sequence still counts as several).
classification: Optional[Literal["effect", "interview", "music", 5 more]]Classification of the audio
Classification of the audio
True when preview holds the entire content unmodified — consumers can
use it directly without fetching asset.url. False or undefined means
the content exceeds the preview budget and consumers must fetch the
full body from S3 to read past the preview.
Leading slice of the content used for inline UI display and as a search shortcut. Capped at TEXT_PREVIEW_MAX_BYTES (UTF-8) and always cut on a code-point boundary so no character is split. Number of characters in the preview varies by script (around 1024 for ASCII, ~340 for CJK, ~256 for emoji-heavy text at the default 1 KB budget).
source: Literal["3d23d", "3d23d:texture", "3d:texture", 77 more]source of the asset
source of the asset
The description, it will contain in priority:
- the manual description
- the advanced captioning when the asset is used in training flow
- the automatic captioning
The embedding of the asset when requested.
Only available when an asset can be embedded (ie: not Detection maps)
first_frame: Optional[FirstFrame]The video asset’s first frame.
Contains the assetId and the url of the first frame.
The video asset’s first frame.
Contains the assetId and the url of the first frame.
last_frame: Optional[LastFrame]The video asset’s last frame.
Contains the assetId and the url of the last frame.
The video asset’s last frame.
Contains the assetId and the url of the last frame.
The original file url.
Contains the url of the original file. without any conversion. Only available for some specific video, audio and threeD assets. Is only specified if the given asset data has been replaced with a new file during the creation of the asset.
original_mime_type: Optional[Literal["application/gzip", "application/json", "application/octet-stream", 47 more]]The mime type of the original file url.
The original file is stored under a key that carries no extension, so this
is the only way to know which format originalFileUrl serves (e.g. an HDRi
skybox exposes a JPEG preview as url and its .exr or .hdr file as the
original). Omitted when the stored format could not be determined.
The mime type of the original file url.
The original file is stored under a key that carries no extension, so this
is the only way to know which format originalFileUrl serves (e.g. an HDRi
skybox exposes a JPEG preview as url and its .exr or .hdr file as the
original). Omitted when the stored format could not be determined.
The output index of the asset within a job This index is an positive integer that starts at 0 It is used to differentiate between multiple outputs of the same job If the job has only one output, this index is 0
quality_gate: Optional[QualityGate]Quality Gate result, computed asynchronously after generation.
details (the per-dimension reasons and suggestions) is returned only when
the asset is read with withQualityGateDetails=true.
Quality Gate result, computed asynchronously after generation.
details (the per-dimension reasons and suggestions) is returned only when
the asset is read with withQualityGateDetails=true.
details: Optional[QualityGateDetails]Per-dimension reasons and suggestions, as produced by the detector.
Returned only when the asset is read with withQualityGateDetails=true by
the project that owns it — the brief-compliance text paraphrases that
project’s brief, so a cross-project read of a public asset gets the summary
alone. Also omitted when the archived verdict is unavailable (asset scored
by an older version, or an unreadable archive).
Per-dimension reasons and suggestions, as produced by the detector.
Returned only when the asset is read with withQualityGateDetails=true by
the project that owns it — the brief-compliance text paraphrases that
project’s brief, so a cross-project read of a public asset gets the summary
alone. Also omitted when the archived verdict is unavailable (asset scored
by an older version, or an unreadable archive).
class AssetLockResponse: …
asset: Asset
edit_capabilities: List[Literal["DETECTION", "GENERATIVE_FILL", "PIXELATE", 8 more]]List of edit capabilities
List of edit capabilities
metadata: AssetMetadataMetadata of the asset with some additional information
Metadata of the asset with some additional information
type: Literal["3d-texture", "3d-texture-albedo", "3d-texture-metallic", 77 more]The type of the asset. Ex: ‘inference-txt2img’ will represent an asset generated from a text to image model
The type of the asset. Ex: ‘inference-txt2img’ will represent an asset generated from a text to image model
How angular is the surface? 0 is like a sphere, 1 is like a mechanical object
The optional aspect ratio given for the generation, only applicable for some models
A bounding box around the object of interest, in the format [x1, y1, x2, y2].
The control image already processed by canny detector. Must reference an existing AssetId.
Ensure upscaled tile have the same color histogram as original tile.
concepts: Optional[List[AssetMetadataConcept]]Flux Kontext LoRA to style the image.
For Flux Kontext Prompt Editing.
Flux Kontext LoRA to style the image. For Flux Kontext Prompt Editing.
Allow the generation of “hallucinations” during the upscale process, which adds additional details and deviates from the original image. Default: optimized for your preset and style.
Amount of decay in creativity over the upscale process. The lowest the value, the less the creativity will be preserved over the upscale process.
The control image processed by depth estimator. Must reference an existing AssetId.
The name its parent’s files reference it by (an OBJ’s MTL, an MTL’s texture), kept as delivered.
Determine the scale at which the upscale process works.
- With a small value, the upscale works at the largest scale, resulting in fewer added details and more coherent images. Ideal for portraits, for example.
- With a large value, the upscale works at the smallest scale, resulting in more added details and more hallucinations. Ideal for landscapes, for example.
(info): A small value is slower and more expensive to run.
Apply extra control to the Skybox 360 geometry. The higher the value, the more the 360 geometry will influence the generated skybox image.
Use with caution. Default is adapted to the other parameters.
The input image to process. Must reference an existing AssetId or be a data URL.
Strengthen the similarity to the original image during the upscale. Default: optimized for your preset and style.
image_type: Optional[Literal["seamfull", "skybox", "texture"]]Preserve the seamless properties of skybox or texture images. Input has to be of same type (seamless).
Preserve the seamless properties of skybox or texture images. Input has to be of same type (seamless).
input_fidelity: Optional[Literal["high", "low"]]When set to high, allows to better preserve details from the input images in the output.
This is especially useful when using images that contain elements like faces or logos that
require accurate preservation in the generated image.
You can provide multiple input images that will all be preserved with high fidelity, but keep
in mind that the first image will be preserved with richer textures and finer details, so if
you include elements such as faces, consider placing them in the first image.
Only available for the gpt-image-1 model.
When set to high, allows to better preserve details from the input images in the output.
This is especially useful when using images that contain elements like faces or logos that
require accurate preservation in the generated image.
You can provide multiple input images that will all be preserved with high fidelity, but keep in mind that the first image will be preserved with richer textures and finer details, so if you include elements such as faces, consider placing them in the first image.
Only available for the gpt-image-1 model.
input_location: Optional[Literal["bottom", "left", "middle", 2 more]]Location of the input image in the output.
Location of the input image in the output.
Its own references (an OBJ’s mtllib, an MTL’s maps) were kept as delivered.
How polished is the surface? 0 is like a rough surface, 1 is like a mirror
model_type: Optional[Literal["custom", "elevenlabs-voice", "flux.1", 28 more]]The type of the generator used
The type of the generator used
Controls the influence of the negative prompt. Default 0 means the negative prompt has no effect. Higher values increase negative prompt influence. Must be > 0 if negativePrompt is provided.
The number of denoising steps for each image generation.
Override the embeddings of the model. Only your prompt and negativePrompt will be used. Use with caution.
List of points (label, x, y) in the image where label = 0 for background and 1 for object.
How polished is the surface? 0 is like a rough surface, 1 is like a mirror
Increase the fidelity to the prompt during upscale. Default: optimized for your preset and style.
How raised is the surface? 0 is flat like water, 1 is like a very rough rock
Additional refinement steps before scaling.
If scalingFactor == 1, the refinement process will be applied (1 + refinementSteps) times. If scalingFactor > 1, the refinement process will be applied refinementSteps times.
Boolean to return the source image with the mask applied as alpha channel
(RGBA PNG, transparent background) instead of the binary mask.
Mutually exclusive with resultMask — passing both as true returns a 400 error.
Note: backgroundOpacity is no longer honored — alpha is binary (0/255) only.
Boolean to return the binary masks in the response.
Mutually exclusive with resultImage — passing both as true returns a 400 error.
Save a flipbook of the texture. Deactivated when the input texture is larger than 2048x2048px
The seed used to generate this asset. <!> Can be a string or a number in some cases <!>.
How shiny is the surface? 0 is like a matte surface, 1 is like a diamond
Strength for the input image structure preservation
The control image for structure. A canny detector will be applied to this image. Must reference an existing AssetId.
The higher the value the more it will look like the style image(s)
List of style images. Most of the time, only one image is enough. It must be existing AssetIds.
Condition the influence of the style image(s). The higher the value, the more the style images will influence the upscaled image.
Target width for the upscaled image, take priority over scaling factor
The owner (project) ID (example: “proj_23tlk332lkht3kl2” or “team_dlkhgs23tlk3hlkth32lkht3kl2” for old teams)
properties: AssetPropertiesThe properties of the asset, content may depend on the kind of asset returned
The properties of the asset, content may depend on the kind of asset returned
camera_mode: Optional[Literal["aerial", "interior", "turntable"]]Recommended preview camera mode emitted by the generator (gsplat route);
passed through to mesh-rendering verbatim.
Recommended preview camera mode emitted by the generator (gsplat route); passed through to mesh-rendering verbatim.
camera_trajectory: Optional[List[AssetPropertiesCameraTrajectory]]Recommended preview camera path emitted by the generator (gsplat route);
passed through to mesh-rendering verbatim.
Recommended preview camera path emitted by the generator (gsplat route); passed through to mesh-rendering verbatim.
Number of Unicode code points in the text. Code-point-aware (so a non-BMP emoji counts as 1) but not full grapheme-cluster aware (a ZWJ sequence still counts as several).
classification: Optional[Literal["effect", "interview", "music", 5 more]]Classification of the audio
Classification of the audio
True when preview holds the entire content unmodified — consumers can
use it directly without fetching asset.url. False or undefined means
the content exceeds the preview budget and consumers must fetch the
full body from S3 to read past the preview.
Leading slice of the content used for inline UI display and as a search shortcut. Capped at TEXT_PREVIEW_MAX_BYTES (UTF-8) and always cut on a code-point boundary so no character is split. Number of characters in the preview varies by script (around 1024 for ASCII, ~340 for CJK, ~256 for emoji-heavy text at the default 1 KB budget).
source: Literal["3d23d", "3d23d:texture", "3d:texture", 77 more]source of the asset
source of the asset
The description, it will contain in priority:
- the manual description
- the advanced captioning when the asset is used in training flow
- the automatic captioning
The embedding of the asset when requested.
Only available when an asset can be embedded (ie: not Detection maps)
first_frame: Optional[AssetFirstFrame]The video asset’s first frame.
Contains the assetId and the url of the first frame.
The video asset’s first frame.
Contains the assetId and the url of the first frame.
last_frame: Optional[AssetLastFrame]The video asset’s last frame.
Contains the assetId and the url of the last frame.
The video asset’s last frame.
Contains the assetId and the url of the last frame.
The original file url.
Contains the url of the original file. without any conversion. Only available for some specific video, audio and threeD assets. Is only specified if the given asset data has been replaced with a new file during the creation of the asset.
original_mime_type: Optional[Literal["application/gzip", "application/json", "application/octet-stream", 47 more]]The mime type of the original file url.
The original file is stored under a key that carries no extension, so this
is the only way to know which format originalFileUrl serves (e.g. an HDRi
skybox exposes a JPEG preview as url and its .exr or .hdr file as the
original). Omitted when the stored format could not be determined.
The mime type of the original file url.
The original file is stored under a key that carries no extension, so this
is the only way to know which format originalFileUrl serves (e.g. an HDRi
skybox exposes a JPEG preview as url and its .exr or .hdr file as the
original). Omitted when the stored format could not be determined.
The output index of the asset within a job This index is an positive integer that starts at 0 It is used to differentiate between multiple outputs of the same job If the job has only one output, this index is 0
preview: Optional[AssetPreview]The asset’s preview.
Contains the assetId and the url of the preview.
The asset’s preview.
Contains the assetId and the url of the preview.
quality_gate: Optional[AssetQualityGate]Quality Gate result, computed asynchronously after generation.
details (the per-dimension reasons and suggestions) is returned only when
the asset is read with withQualityGateDetails=true.
Quality Gate result, computed asynchronously after generation.
details (the per-dimension reasons and suggestions) is returned only when
the asset is read with withQualityGateDetails=true.
details: Optional[AssetQualityGateDetails]Per-dimension reasons and suggestions, as produced by the detector.
Returned only when the asset is read with withQualityGateDetails=true by
the project that owns it — the brief-compliance text paraphrases that
project’s brief, so a cross-project read of a public asset gets the summary
alone. Also omitted when the archived verdict is unavailable (asset scored
by an older version, or an unreadable archive).
Per-dimension reasons and suggestions, as produced by the detector.
Returned only when the asset is read with withQualityGateDetails=true by
the project that owns it — the brief-compliance text paraphrases that
project’s brief, so a cross-project read of a public asset gets the summary
alone. Also omitted when the archived verdict is unavailable (asset scored
by an older version, or an unreadable archive).
class AssetRetrieveResponse: …
asset: Asset
edit_capabilities: List[Literal["DETECTION", "GENERATIVE_FILL", "PIXELATE", 8 more]]List of edit capabilities
List of edit capabilities
metadata: AssetMetadataMetadata of the asset with some additional information
Metadata of the asset with some additional information
type: Literal["3d-texture", "3d-texture-albedo", "3d-texture-metallic", 77 more]The type of the asset. Ex: ‘inference-txt2img’ will represent an asset generated from a text to image model
The type of the asset. Ex: ‘inference-txt2img’ will represent an asset generated from a text to image model
How angular is the surface? 0 is like a sphere, 1 is like a mechanical object
The optional aspect ratio given for the generation, only applicable for some models
A bounding box around the object of interest, in the format [x1, y1, x2, y2].
The control image already processed by canny detector. Must reference an existing AssetId.
Ensure upscaled tile have the same color histogram as original tile.
concepts: Optional[List[AssetMetadataConcept]]Flux Kontext LoRA to style the image.
For Flux Kontext Prompt Editing.
Flux Kontext LoRA to style the image. For Flux Kontext Prompt Editing.
Allow the generation of “hallucinations” during the upscale process, which adds additional details and deviates from the original image. Default: optimized for your preset and style.
Amount of decay in creativity over the upscale process. The lowest the value, the less the creativity will be preserved over the upscale process.
The control image processed by depth estimator. Must reference an existing AssetId.
The name its parent’s files reference it by (an OBJ’s MTL, an MTL’s texture), kept as delivered.
Determine the scale at which the upscale process works.
- With a small value, the upscale works at the largest scale, resulting in fewer added details and more coherent images. Ideal for portraits, for example.
- With a large value, the upscale works at the smallest scale, resulting in more added details and more hallucinations. Ideal for landscapes, for example.
(info): A small value is slower and more expensive to run.
Apply extra control to the Skybox 360 geometry. The higher the value, the more the 360 geometry will influence the generated skybox image.
Use with caution. Default is adapted to the other parameters.
The input image to process. Must reference an existing AssetId or be a data URL.
Strengthen the similarity to the original image during the upscale. Default: optimized for your preset and style.
image_type: Optional[Literal["seamfull", "skybox", "texture"]]Preserve the seamless properties of skybox or texture images. Input has to be of same type (seamless).
Preserve the seamless properties of skybox or texture images. Input has to be of same type (seamless).
input_fidelity: Optional[Literal["high", "low"]]When set to high, allows to better preserve details from the input images in the output.
This is especially useful when using images that contain elements like faces or logos that
require accurate preservation in the generated image.
You can provide multiple input images that will all be preserved with high fidelity, but keep
in mind that the first image will be preserved with richer textures and finer details, so if
you include elements such as faces, consider placing them in the first image.
Only available for the gpt-image-1 model.
When set to high, allows to better preserve details from the input images in the output.
This is especially useful when using images that contain elements like faces or logos that
require accurate preservation in the generated image.
You can provide multiple input images that will all be preserved with high fidelity, but keep in mind that the first image will be preserved with richer textures and finer details, so if you include elements such as faces, consider placing them in the first image.
Only available for the gpt-image-1 model.
input_location: Optional[Literal["bottom", "left", "middle", 2 more]]Location of the input image in the output.
Location of the input image in the output.
Its own references (an OBJ’s mtllib, an MTL’s maps) were kept as delivered.
How polished is the surface? 0 is like a rough surface, 1 is like a mirror
model_type: Optional[Literal["custom", "elevenlabs-voice", "flux.1", 28 more]]The type of the generator used
The type of the generator used
Controls the influence of the negative prompt. Default 0 means the negative prompt has no effect. Higher values increase negative prompt influence. Must be > 0 if negativePrompt is provided.
The number of denoising steps for each image generation.
Override the embeddings of the model. Only your prompt and negativePrompt will be used. Use with caution.
List of points (label, x, y) in the image where label = 0 for background and 1 for object.
How polished is the surface? 0 is like a rough surface, 1 is like a mirror
Increase the fidelity to the prompt during upscale. Default: optimized for your preset and style.
How raised is the surface? 0 is flat like water, 1 is like a very rough rock
Additional refinement steps before scaling.
If scalingFactor == 1, the refinement process will be applied (1 + refinementSteps) times. If scalingFactor > 1, the refinement process will be applied refinementSteps times.
Boolean to return the source image with the mask applied as alpha channel
(RGBA PNG, transparent background) instead of the binary mask.
Mutually exclusive with resultMask — passing both as true returns a 400 error.
Note: backgroundOpacity is no longer honored — alpha is binary (0/255) only.
Boolean to return the binary masks in the response.
Mutually exclusive with resultImage — passing both as true returns a 400 error.
Save a flipbook of the texture. Deactivated when the input texture is larger than 2048x2048px
The seed used to generate this asset. <!> Can be a string or a number in some cases <!>.
How shiny is the surface? 0 is like a matte surface, 1 is like a diamond
Strength for the input image structure preservation
The control image for structure. A canny detector will be applied to this image. Must reference an existing AssetId.
The higher the value the more it will look like the style image(s)
List of style images. Most of the time, only one image is enough. It must be existing AssetIds.
Condition the influence of the style image(s). The higher the value, the more the style images will influence the upscaled image.
Target width for the upscaled image, take priority over scaling factor
The owner (project) ID (example: “proj_23tlk332lkht3kl2” or “team_dlkhgs23tlk3hlkth32lkht3kl2” for old teams)
properties: AssetPropertiesThe properties of the asset, content may depend on the kind of asset returned
The properties of the asset, content may depend on the kind of asset returned
camera_mode: Optional[Literal["aerial", "interior", "turntable"]]Recommended preview camera mode emitted by the generator (gsplat route);
passed through to mesh-rendering verbatim.
Recommended preview camera mode emitted by the generator (gsplat route); passed through to mesh-rendering verbatim.
camera_trajectory: Optional[List[AssetPropertiesCameraTrajectory]]Recommended preview camera path emitted by the generator (gsplat route);
passed through to mesh-rendering verbatim.
Recommended preview camera path emitted by the generator (gsplat route); passed through to mesh-rendering verbatim.
Number of Unicode code points in the text. Code-point-aware (so a non-BMP emoji counts as 1) but not full grapheme-cluster aware (a ZWJ sequence still counts as several).
classification: Optional[Literal["effect", "interview", "music", 5 more]]Classification of the audio
Classification of the audio
True when preview holds the entire content unmodified — consumers can
use it directly without fetching asset.url. False or undefined means
the content exceeds the preview budget and consumers must fetch the
full body from S3 to read past the preview.
Leading slice of the content used for inline UI display and as a search shortcut. Capped at TEXT_PREVIEW_MAX_BYTES (UTF-8) and always cut on a code-point boundary so no character is split. Number of characters in the preview varies by script (around 1024 for ASCII, ~340 for CJK, ~256 for emoji-heavy text at the default 1 KB budget).
source: Literal["3d23d", "3d23d:texture", "3d:texture", 77 more]source of the asset
source of the asset
The description, it will contain in priority:
- the manual description
- the advanced captioning when the asset is used in training flow
- the automatic captioning
The embedding of the asset when requested.
Only available when an asset can be embedded (ie: not Detection maps)
first_frame: Optional[AssetFirstFrame]The video asset’s first frame.
Contains the assetId and the url of the first frame.
The video asset’s first frame.
Contains the assetId and the url of the first frame.
last_frame: Optional[AssetLastFrame]The video asset’s last frame.
Contains the assetId and the url of the last frame.
The video asset’s last frame.
Contains the assetId and the url of the last frame.
The original file url.
Contains the url of the original file. without any conversion. Only available for some specific video, audio and threeD assets. Is only specified if the given asset data has been replaced with a new file during the creation of the asset.
original_mime_type: Optional[Literal["application/gzip", "application/json", "application/octet-stream", 47 more]]The mime type of the original file url.
The original file is stored under a key that carries no extension, so this
is the only way to know which format originalFileUrl serves (e.g. an HDRi
skybox exposes a JPEG preview as url and its .exr or .hdr file as the
original). Omitted when the stored format could not be determined.
The mime type of the original file url.
The original file is stored under a key that carries no extension, so this
is the only way to know which format originalFileUrl serves (e.g. an HDRi
skybox exposes a JPEG preview as url and its .exr or .hdr file as the
original). Omitted when the stored format could not be determined.
The output index of the asset within a job This index is an positive integer that starts at 0 It is used to differentiate between multiple outputs of the same job If the job has only one output, this index is 0
preview: Optional[AssetPreview]The asset’s preview.
Contains the assetId and the url of the preview.
The asset’s preview.
Contains the assetId and the url of the preview.
quality_gate: Optional[AssetQualityGate]Quality Gate result, computed asynchronously after generation.
details (the per-dimension reasons and suggestions) is returned only when
the asset is read with withQualityGateDetails=true.
Quality Gate result, computed asynchronously after generation.
details (the per-dimension reasons and suggestions) is returned only when
the asset is read with withQualityGateDetails=true.
details: Optional[AssetQualityGateDetails]Per-dimension reasons and suggestions, as produced by the detector.
Returned only when the asset is read with withQualityGateDetails=true by
the project that owns it — the brief-compliance text paraphrases that
project’s brief, so a cross-project read of a public asset gets the summary
alone. Also omitted when the archived verdict is unavailable (asset scored
by an older version, or an unreadable archive).
Per-dimension reasons and suggestions, as produced by the detector.
Returned only when the asset is read with withQualityGateDetails=true by
the project that owns it — the brief-compliance text paraphrases that
project’s brief, so a cross-project read of a public asset gets the summary
alone. Also omitted when the archived verdict is unavailable (asset scored
by an older version, or an unreadable archive).
class AssetUnlockResponse: …
asset: Asset
edit_capabilities: List[Literal["DETECTION", "GENERATIVE_FILL", "PIXELATE", 8 more]]List of edit capabilities
List of edit capabilities
metadata: AssetMetadataMetadata of the asset with some additional information
Metadata of the asset with some additional information
type: Literal["3d-texture", "3d-texture-albedo", "3d-texture-metallic", 77 more]The type of the asset. Ex: ‘inference-txt2img’ will represent an asset generated from a text to image model
The type of the asset. Ex: ‘inference-txt2img’ will represent an asset generated from a text to image model
How angular is the surface? 0 is like a sphere, 1 is like a mechanical object
The optional aspect ratio given for the generation, only applicable for some models
A bounding box around the object of interest, in the format [x1, y1, x2, y2].
The control image already processed by canny detector. Must reference an existing AssetId.
Ensure upscaled tile have the same color histogram as original tile.
concepts: Optional[List[AssetMetadataConcept]]Flux Kontext LoRA to style the image.
For Flux Kontext Prompt Editing.
Flux Kontext LoRA to style the image. For Flux Kontext Prompt Editing.
Allow the generation of “hallucinations” during the upscale process, which adds additional details and deviates from the original image. Default: optimized for your preset and style.
Amount of decay in creativity over the upscale process. The lowest the value, the less the creativity will be preserved over the upscale process.
The control image processed by depth estimator. Must reference an existing AssetId.
The name its parent’s files reference it by (an OBJ’s MTL, an MTL’s texture), kept as delivered.
Determine the scale at which the upscale process works.
- With a small value, the upscale works at the largest scale, resulting in fewer added details and more coherent images. Ideal for portraits, for example.
- With a large value, the upscale works at the smallest scale, resulting in more added details and more hallucinations. Ideal for landscapes, for example.
(info): A small value is slower and more expensive to run.
Apply extra control to the Skybox 360 geometry. The higher the value, the more the 360 geometry will influence the generated skybox image.
Use with caution. Default is adapted to the other parameters.
The input image to process. Must reference an existing AssetId or be a data URL.
Strengthen the similarity to the original image during the upscale. Default: optimized for your preset and style.
image_type: Optional[Literal["seamfull", "skybox", "texture"]]Preserve the seamless properties of skybox or texture images. Input has to be of same type (seamless).
Preserve the seamless properties of skybox or texture images. Input has to be of same type (seamless).
input_fidelity: Optional[Literal["high", "low"]]When set to high, allows to better preserve details from the input images in the output.
This is especially useful when using images that contain elements like faces or logos that
require accurate preservation in the generated image.
You can provide multiple input images that will all be preserved with high fidelity, but keep
in mind that the first image will be preserved with richer textures and finer details, so if
you include elements such as faces, consider placing them in the first image.
Only available for the gpt-image-1 model.
When set to high, allows to better preserve details from the input images in the output.
This is especially useful when using images that contain elements like faces or logos that
require accurate preservation in the generated image.
You can provide multiple input images that will all be preserved with high fidelity, but keep in mind that the first image will be preserved with richer textures and finer details, so if you include elements such as faces, consider placing them in the first image.
Only available for the gpt-image-1 model.
input_location: Optional[Literal["bottom", "left", "middle", 2 more]]Location of the input image in the output.
Location of the input image in the output.
Its own references (an OBJ’s mtllib, an MTL’s maps) were kept as delivered.
How polished is the surface? 0 is like a rough surface, 1 is like a mirror
model_type: Optional[Literal["custom", "elevenlabs-voice", "flux.1", 28 more]]The type of the generator used
The type of the generator used
Controls the influence of the negative prompt. Default 0 means the negative prompt has no effect. Higher values increase negative prompt influence. Must be > 0 if negativePrompt is provided.
The number of denoising steps for each image generation.
Override the embeddings of the model. Only your prompt and negativePrompt will be used. Use with caution.
List of points (label, x, y) in the image where label = 0 for background and 1 for object.
How polished is the surface? 0 is like a rough surface, 1 is like a mirror
Increase the fidelity to the prompt during upscale. Default: optimized for your preset and style.
How raised is the surface? 0 is flat like water, 1 is like a very rough rock
Additional refinement steps before scaling.
If scalingFactor == 1, the refinement process will be applied (1 + refinementSteps) times. If scalingFactor > 1, the refinement process will be applied refinementSteps times.
Boolean to return the source image with the mask applied as alpha channel
(RGBA PNG, transparent background) instead of the binary mask.
Mutually exclusive with resultMask — passing both as true returns a 400 error.
Note: backgroundOpacity is no longer honored — alpha is binary (0/255) only.
Boolean to return the binary masks in the response.
Mutually exclusive with resultImage — passing both as true returns a 400 error.
Save a flipbook of the texture. Deactivated when the input texture is larger than 2048x2048px
The seed used to generate this asset. <!> Can be a string or a number in some cases <!>.
How shiny is the surface? 0 is like a matte surface, 1 is like a diamond
Strength for the input image structure preservation
The control image for structure. A canny detector will be applied to this image. Must reference an existing AssetId.
The higher the value the more it will look like the style image(s)
List of style images. Most of the time, only one image is enough. It must be existing AssetIds.
Condition the influence of the style image(s). The higher the value, the more the style images will influence the upscaled image.
Target width for the upscaled image, take priority over scaling factor
The owner (project) ID (example: “proj_23tlk332lkht3kl2” or “team_dlkhgs23tlk3hlkth32lkht3kl2” for old teams)
properties: AssetPropertiesThe properties of the asset, content may depend on the kind of asset returned
The properties of the asset, content may depend on the kind of asset returned
camera_mode: Optional[Literal["aerial", "interior", "turntable"]]Recommended preview camera mode emitted by the generator (gsplat route);
passed through to mesh-rendering verbatim.
Recommended preview camera mode emitted by the generator (gsplat route); passed through to mesh-rendering verbatim.
camera_trajectory: Optional[List[AssetPropertiesCameraTrajectory]]Recommended preview camera path emitted by the generator (gsplat route);
passed through to mesh-rendering verbatim.
Recommended preview camera path emitted by the generator (gsplat route); passed through to mesh-rendering verbatim.
Number of Unicode code points in the text. Code-point-aware (so a non-BMP emoji counts as 1) but not full grapheme-cluster aware (a ZWJ sequence still counts as several).
classification: Optional[Literal["effect", "interview", "music", 5 more]]Classification of the audio
Classification of the audio
True when preview holds the entire content unmodified — consumers can
use it directly without fetching asset.url. False or undefined means
the content exceeds the preview budget and consumers must fetch the
full body from S3 to read past the preview.
Leading slice of the content used for inline UI display and as a search shortcut. Capped at TEXT_PREVIEW_MAX_BYTES (UTF-8) and always cut on a code-point boundary so no character is split. Number of characters in the preview varies by script (around 1024 for ASCII, ~340 for CJK, ~256 for emoji-heavy text at the default 1 KB budget).
source: Literal["3d23d", "3d23d:texture", "3d:texture", 77 more]source of the asset
source of the asset
The description, it will contain in priority:
- the manual description
- the advanced captioning when the asset is used in training flow
- the automatic captioning
The embedding of the asset when requested.
Only available when an asset can be embedded (ie: not Detection maps)
first_frame: Optional[AssetFirstFrame]The video asset’s first frame.
Contains the assetId and the url of the first frame.
The video asset’s first frame.
Contains the assetId and the url of the first frame.
last_frame: Optional[AssetLastFrame]The video asset’s last frame.
Contains the assetId and the url of the last frame.
The video asset’s last frame.
Contains the assetId and the url of the last frame.
The original file url.
Contains the url of the original file. without any conversion. Only available for some specific video, audio and threeD assets. Is only specified if the given asset data has been replaced with a new file during the creation of the asset.
original_mime_type: Optional[Literal["application/gzip", "application/json", "application/octet-stream", 47 more]]The mime type of the original file url.
The original file is stored under a key that carries no extension, so this
is the only way to know which format originalFileUrl serves (e.g. an HDRi
skybox exposes a JPEG preview as url and its .exr or .hdr file as the
original). Omitted when the stored format could not be determined.
The mime type of the original file url.
The original file is stored under a key that carries no extension, so this
is the only way to know which format originalFileUrl serves (e.g. an HDRi
skybox exposes a JPEG preview as url and its .exr or .hdr file as the
original). Omitted when the stored format could not be determined.
The output index of the asset within a job This index is an positive integer that starts at 0 It is used to differentiate between multiple outputs of the same job If the job has only one output, this index is 0
preview: Optional[AssetPreview]The asset’s preview.
Contains the assetId and the url of the preview.
The asset’s preview.
Contains the assetId and the url of the preview.
quality_gate: Optional[AssetQualityGate]Quality Gate result, computed asynchronously after generation.
details (the per-dimension reasons and suggestions) is returned only when
the asset is read with withQualityGateDetails=true.
Quality Gate result, computed asynchronously after generation.
details (the per-dimension reasons and suggestions) is returned only when
the asset is read with withQualityGateDetails=true.
details: Optional[AssetQualityGateDetails]Per-dimension reasons and suggestions, as produced by the detector.
Returned only when the asset is read with withQualityGateDetails=true by
the project that owns it — the brief-compliance text paraphrases that
project’s brief, so a cross-project read of a public asset gets the summary
alone. Also omitted when the archived verdict is unavailable (asset scored
by an older version, or an unreadable archive).
Per-dimension reasons and suggestions, as produced by the detector.
Returned only when the asset is read with withQualityGateDetails=true by
the project that owns it — the brief-compliance text paraphrases that
project’s brief, so a cross-project read of a public asset gets the summary
alone. Also omitted when the archived verdict is unavailable (asset scored
by an older version, or an unreadable archive).
class AssetUpdateResponse: …
asset: Asset
edit_capabilities: List[Literal["DETECTION", "GENERATIVE_FILL", "PIXELATE", 8 more]]List of edit capabilities
List of edit capabilities
metadata: AssetMetadataMetadata of the asset with some additional information
Metadata of the asset with some additional information
type: Literal["3d-texture", "3d-texture-albedo", "3d-texture-metallic", 77 more]The type of the asset. Ex: ‘inference-txt2img’ will represent an asset generated from a text to image model
The type of the asset. Ex: ‘inference-txt2img’ will represent an asset generated from a text to image model
How angular is the surface? 0 is like a sphere, 1 is like a mechanical object
The optional aspect ratio given for the generation, only applicable for some models
A bounding box around the object of interest, in the format [x1, y1, x2, y2].
The control image already processed by canny detector. Must reference an existing AssetId.
Ensure upscaled tile have the same color histogram as original tile.
concepts: Optional[List[AssetMetadataConcept]]Flux Kontext LoRA to style the image.
For Flux Kontext Prompt Editing.
Flux Kontext LoRA to style the image. For Flux Kontext Prompt Editing.
Allow the generation of “hallucinations” during the upscale process, which adds additional details and deviates from the original image. Default: optimized for your preset and style.
Amount of decay in creativity over the upscale process. The lowest the value, the less the creativity will be preserved over the upscale process.
The control image processed by depth estimator. Must reference an existing AssetId.
The name its parent’s files reference it by (an OBJ’s MTL, an MTL’s texture), kept as delivered.
Determine the scale at which the upscale process works.
- With a small value, the upscale works at the largest scale, resulting in fewer added details and more coherent images. Ideal for portraits, for example.
- With a large value, the upscale works at the smallest scale, resulting in more added details and more hallucinations. Ideal for landscapes, for example.
(info): A small value is slower and more expensive to run.
Apply extra control to the Skybox 360 geometry. The higher the value, the more the 360 geometry will influence the generated skybox image.
Use with caution. Default is adapted to the other parameters.
The input image to process. Must reference an existing AssetId or be a data URL.
Strengthen the similarity to the original image during the upscale. Default: optimized for your preset and style.
image_type: Optional[Literal["seamfull", "skybox", "texture"]]Preserve the seamless properties of skybox or texture images. Input has to be of same type (seamless).
Preserve the seamless properties of skybox or texture images. Input has to be of same type (seamless).
input_fidelity: Optional[Literal["high", "low"]]When set to high, allows to better preserve details from the input images in the output.
This is especially useful when using images that contain elements like faces or logos that
require accurate preservation in the generated image.
You can provide multiple input images that will all be preserved with high fidelity, but keep
in mind that the first image will be preserved with richer textures and finer details, so if
you include elements such as faces, consider placing them in the first image.
Only available for the gpt-image-1 model.
When set to high, allows to better preserve details from the input images in the output.
This is especially useful when using images that contain elements like faces or logos that
require accurate preservation in the generated image.
You can provide multiple input images that will all be preserved with high fidelity, but keep in mind that the first image will be preserved with richer textures and finer details, so if you include elements such as faces, consider placing them in the first image.
Only available for the gpt-image-1 model.