octoai.clients package

Submodules

octoai.clients.asset_orch module

Asset Orchestrator class.

class octoai.clients.asset_orch.Asset(id, asset_type, name, description, size_bytes, status, status_details, created_at, data, tenant_uuid, *args, **kwargs)[source]

Bases: object

Asset Orchestrator implementation of an asset.

Hashable with UUID used as key if passed to a dictionary.

Parameters:

  • id – UUID unique to asset.

  • asset_type – AssetType, including “lora”, “vae”, “checkpoint”, or “textual-inversion”.

  • name – Alphanumeric, _, or - allowed.

  • description – Description of asset.

  • size_bytes – Total bytes of the asset file.

  • status – Status of Asset. One of “ready_to_upload”, “ready”, “uploaded”, “deleted”, “rejected”, or “error”.

  • status_details – Description of asset status.

  • created_at – Time created.

  • data – Additional information about asset such as engine, file_format, etc.

  • tenant_uuid – UUID of person who created the asset.

class octoai.clients.asset_orch.AssetOrchestrator(token: str | None = None, config_path: str | None = None, endpoint: str = 'https://api.octoai.cloud/')[source]

Bases: object

Asset Orchestrator class to create, read, delete, and list assets.

Parameters:

  • token – OCTOAI_TOKEN if one is not set as an environment variable, default to None.

  • config_path – Path to config file from CLI, used if token or envvar is not set, default to None and checks default path.

  • endpoint – Defaults to “https://api.octoai.cloud/”.

create(data: AssetData, name: str, file: str | None = None, url: str | None = None, is_public: bool = False, description: str | None = None, transfer_api_type: str | TransferApiType | None = None) → Asset[source]

Create and upload an asset.

Parameters:

  • data – CheckpointData, LoraData, VAEData, or TextualInversionData

  • name – Name of asset, alphanumeric with - and _ characters allowed.

  • file – str to file path, optional, defaults to None.

  • url – Url to copy file data from instead of file, optional, defaults to None. If set file and transfer_api_type must be None.

  • description – Description of asset, optional, defaults to None.

  • transfer_api_type – TransferApiType or str of either “presigned-url”, or “sts”, defaults to “sts” for >= 50mb and “presigned-url” for under 50mb.

Returns:

Asset

delete(asset_id: str | None) → DeleteAssetResponse[source]

Delete an asset.

Parameters:

asset_id – the UUID of the asset to be deleted.

Returns:

DeleteAssetResponse containing an id as a str and deleted_at as a str with the timestamp.

get(name: str | None = None, id: str | None = None) → Asset[source]

Get an asset associated with an asset name or asset id.

Parameters:

  • name – Name of the asset to get.

  • id – ID of the asset to get.

Returns:

Asset

list(name: str | None = None, is_public: bool | None = None, data_type: DataType | None = None, asset_type: List[AssetType] | dict | None = None, engine: List[BaseEngine] | None = None, limit: int | None = None, offset: int | None = None, owner: str | None = None) → list[Asset][source]

Return list of assets filtered on the non-None parameters.

Parameters:

  • name – Asset name, alphanumeric, -, and _ allowed. Defaults to None.

  • is_public – Whether to filter for public assets, such as looking for octoai public assets.

  • data_type – DataType, defaults to None.

  • asset_type – List of AssetType of assets, defaults to None.

  • engine – List of BaseEngine of assets, defaults to None.

  • limit – Max number of assets to return, defaults to None.

  • offset – Where to start including next list of assets, defaults to None.

  • owner – ID of owner, defaults to None.

Returns:

list[Asset]

wait_for_ready(asset: Asset, poll_interval=10, timeout_seconds=900) → Asset[source]

Wait for asset to be ready to use.

This waits until the asset’s status is READY or an error status.

Parameters:

asset – Asset to wait on

class octoai.clients.asset_orch.CheckpointData(engine: BaseEngine | str, file_format: FileFormat | str, data_type: DataType | str = DataType.FP16)[source]

Bases: ModelData

Checkpoint data associated with checkpoint AssetType.

Used for AssetOrchestrator.create().

Parameters:

  • engine – Compatible BaseEngine type for model. Includes “image/stable-diffusion-v1-5”, “image/stable-diffusion-xl-v1-0”, and “image/controlnet-sdxl”.

  • file_format – FileFormat of model, includes safetensors.

  • data_type – DataType or str matching an enum in DataType, default to ‘fp16’.

class octoai.clients.asset_orch.FileData(file_format: FileExtension | str, version: str = '')[source]

Bases: object

class octoai.clients.asset_orch.LoraData(engine: BaseEngine | str, file_format: FileFormat | str, data_type: DataType | str = DataType.FP16)[source]

Bases: ModelData

LoRA data associated with the lora AssetType.

Used for AssetOrchestrator.create().

Parameters:

  • engine – Compatible BaseEngine type for model. Includes “image/stable-diffusion-v1-5”, “image/stable-diffusion-xl-v1-0”, and “image/controlnet-sdxl”.

  • file_format – FileFormat of model, includes safetensors.

  • data_type – DataType or str matching an enum in DataType, default to ‘fp16’.

class octoai.clients.asset_orch.ModelData(engine: BaseEngine | str, file_format: FileFormat | str, data_type: DataType | str)[source]

Bases: ABC

Base class for Checkpoints, LoRAs, and Textual Inversions.

Parameters:

  • engine – Compatible BaseEngine type for model. Includes “image/stable-diffusion-v1-5” or “image/stable-diffusion-xl-v1-0”.

  • file_format – FileFormat of model, includes safetensors.

  • data_type – DataType or str matching an enum in DataType, default to ‘fp16’.

class octoai.clients.asset_orch.TextualInversionData(engine: BaseEngine | str, file_format: FileFormat | str, trigger_words: List[str], data_type: DataType | str = DataType.FP16)[source]

Bases: ModelData

TextualInversionData associated with textual_inversion AssetType.

Used for AssetOrchestrator.create().

Parameters:

  • engine – Compatible BaseEngine type for model. Includes “image/stable-diffusion-v1-5”, “image/stable-diffusion-xl-v1-0”, and “image/controlnet-sdxl”.

  • file_format – FileFormat of model, includes safetensors.

  • data_type – DataType or str matching an enum in DataType, default to ‘fp16’.

class octoai.clients.asset_orch.VAEData(engine: BaseEngine | str, file_format: FileFormat | str, data_type: DataType | str = DataType.FP16)[source]

Bases: ModelData

VAE data associated with the vae AssetType.

Used for AssetOrchestrator.create().

Parameters:

  • engine – Compatible BaseEngine type for model. Includes “image/stable-diffusion-v1-5”, “image/stable-diffusion-xl-v1-0”, and “image/controlnet-sdxl”.

  • file_format – FileFormat of model, includes safetensors.

  • data_type – DataType or str matching an enum in DataType, default to ‘fp16’.

octoai.clients.fine_tuning module

OctoAI Fine Tuning.

class octoai.clients.fine_tuning.FineTuningClient(token: str, endpoint: str = 'https://api.octoai.cloud/')[source]

Bases: object

OctoAI Fine Tuning Client.

This client is used to interact with the OctoAI Fine Tuning API. This class can be used on its own, providing a valid token and endpoint, but it is mainly used from instances of :class: octoai.client.Client as client.tune.

cancel(id: str) → Tune[source]

Cancel a fine tuning job.

Parameters:

id – Required. The id of the fine tuning job to cancel.

Returns:

Tune object representing the fine tuning job.

create(name: str, base_checkpoint: str | Asset, files: List[LoraTuneFile] | List[Asset] | List[str] | Mapping[Asset, str] | Mapping[str, str], trigger_words: str | List[str], steps: int, description: str | None = None, engine: str | BaseEngine | None = None, seed: int | None = None, continue_on_rejection: bool = False) → Tune[source]

Create a new fine tuning job.

Parameters:

  • name – Required. The name of the fine tuning job.

  • base_checkpoint – Required. The base checkpoint to use. Accepts an asset id or an asset object.

  • files – Required. The training files to use. Supports a list of assets or asset ids without captions, or a mapping of assets or asset ids to captions.

  • trigger_words – Required. The trigger words to use.

  • description – Optional. The description of the fine tuning job.

  • engine – Optional. The engine to use. Defaults to the corresponding engine for the base checkpoint.

  • seed – Optional. The seed to use for training. Defaults to a random seed.

  • continue_on_rejection – Optionally continue with the fine-tune job if any of the training images are identified as NSFW. Defaults to False.

Steps:

Required. The number of steps to train for.

Returns:

Tune object representing the newly created fine tuning job.

delete(id: str)[source]

Delete a fine tuning job.

Parameters:

id – Required. The id of the fine tuning job to delete.

get(id: str) → Tune[source]

Get a fine tuning job.

Parameters:

id – Required. The id of the fine tuning job.

Returns:

Tune object representing the fine tuning job.

list(limit: int | None = None, offset: int | None = None, name: str | None = None, tune_type: str | None = None, base_checkpoint: str | Asset | None = None, trigger_words: str | List[str] | None = None) → ListTunesResponse[source]

List available fine tuning jobs.

Parameters:

  • limit – Optional. The maximum number of fine tuning jobs to return.

  • offset – Optional. The offset to start listing fine tuning jobs from.

  • name – Optional. Filter results by job name.

  • tune_type – Optional. Filter results by job type.

  • base_checkpoint – Optional. Filter results by base checkpoint. Accepts an asset id or an asset object.

  • trigger_words – Optional. Filter results by trigger words.

Returns:

ListTunesResponse object representing the list of fine tuning jobs.

octoai.clients.image_gen module

OctoAI Image Generation.

class octoai.clients.image_gen.Engine(value)[source]

Bases: str, Enum

SDXL: Stable Diffusion XL SD: Stable Diffusion

CONTROLNET_SD = 'controlnet-sd'
CONTROLNET_SDXL = 'controlnet-sdxl'
SD = 'sd'
SDXL = 'sdxl'
SSD = 'ssd'
to_name()[source]
class octoai.clients.image_gen.ImageGenerateResponse(images: List[Image], removed_for_safety: int)[source]

Bases: object

Image generation response.

Contains a list of images as well as a counter of those filtered for safety.

property images: List[Image]

Return list of Image generated from request.

property removed_for_safety: int

Return int representing number of images removed for safety.

class octoai.clients.image_gen.ImageGenerator(api_endpoint: str | None = None, *args, **kwargs)[source]

Bases: Client

Client for image generation.

generate(engine: Engine | str, prompt: str, prompt_2: str | None = None, negative_prompt: str | None = None, negative_prompt_2: str | None = None, checkpoint: str | Asset | None = None, vae: str | Asset | None = None, textual_inversions: Dict[str | Asset, str] | None = None, loras: Dict[str | Asset, float] | None = None, sampler: str | Scheduler | None = None, height: int | None = None, width: int | None = None, cfg_scale: float | None = 12.0, steps: int | None = 30, num_images: int | None = 1, seed: int | List[int] | None = None, init_image: str | Image | None = None, controlnet: str | None = None, controlnet_image: str | Image | None = None, controlnet_conditioning_scale: float | None = None, strength: float | None = None, style_preset: str | SDXLStyles | None = None, use_refiner: bool | None = None, high_noise_frac: float | None = None, enable_safety: bool | None = True, image_encoding: ImageEncoding | str = None) → ImageGenerateResponse[source]

Generate a list of images based on request.

Parameters:

  • engine – Required. “sdxl” for Stable Diffusion XL; “sd” for Stable Diffusion 1.5; “ssd” for Stable Diffusion SSD; “controlnet-sdxl” for ControlNet Stable Diffusion XL; “controlnet-sd” for ControlNet Stable Diffusion 1.5.

  • prompt – Required. Describes the image to generate. ex. “An octopus playing chess, masterpiece, photorealistic”

  • prompt_2 – High level description of the image to generate, defaults to None.

  • negative_prompt – Description of image traits to avoid, defaults to None. ex. “Fingers, distortions”

  • negative_prompt_2 – High level description of things to avoid during generation, defaults to None. ex. “Unusual proportions and distorted faces”

  • checkpoint – Which checkpoint to use for inferences, defaults to None.

  • vae – Custom VAE to be used during image generation, defaults to None.

  • textual_inversions – A dictionary of textual inversion updates, defaults to None ex. {‘name’: ‘trigger_word’}

  • loras – A dictionary of LoRAs updates to apply and their weight, can also be used with Assets created in the SDK directly, defaults to None. ex. {‘crayon-style’: 0.3, my_created_asset: 0.1}

  • sampler – Scheduler to use when generating image, defaults to None.

  • height – Height of image to generate, defaults to None.

  • width – Width of image to generate, defaults to None.

  • cfg_scale – How closely to adhere to prompt description, defaults to 12.0. Must be >= 0 and <= 50.

  • steps – How many steps of diffusion to run, defaults to 30. May be > 0 and <= 100.

  • num_images – How many images to generate, defaults to 1. May be > 0 and <= 4.

  • seed – Fixed random seed, useful when attempting to generate a specific image, defaults to None. May be >= 0 < 2**32.

  • init_image – Starting image for img2img mode, defaults to None. Requires a b64 string image or Image.

  • controlnet – String matching id of controlnet to use for controlnet engine inferences, defaults to None. Required for using controlnet engines.

  • controlnet_image – Starting image for controlnet-sdxl mode, defaults to None. Requires a b64 string image or Image.

  • controlnet_conditioning_scale – How strong the effect of the controlnet should be, defaults to 1.0.

  • strength – How much creative to be in img2img mode, defaults to 0.8. May be >= 0 and <= 1. Must have an init_image.

  • style_preset – Used to guide the output image towards a particular style, only usable with SDXL,defaults to None. ex. “low-poly”

  • use_refiner – Whether to apply the sdxl refiner, defaults to True.

  • high_noise_frac – Which fraction of steps to perform with the base model, defaults to 0.8. May be >= 0 and <= 1.

  • enable_safety – Whether to use safety checking on generated outputs or not, defaults to True.

  • image_encoding – Choose returned ImageEncoding type, defaults to ImageEncoding.JPEG.

Returns:

GenerateImagesResponse object including properties for a list of images as well as a counter of total images returned below the num_images value due to being removed for safety.

Module contents

Client module.