Skip to content

Latest commit

 

History

History
242 lines (197 loc) · 11.8 KB

CHANGELOG.md

File metadata and controls

242 lines (197 loc) · 11.8 KB

Changelog

All notable changes to this project will be documented in this file.

[0.5.0]

Changed

  • Replace polling with asyncio.Future for O(1) message routing via _pending_operations HashMap
  • Add semaphore to limit concurrent requests (default 15, env RUNWARE_MAX_CONCURRENT_REQUESTS)
  • Add jitter on reconnect to prevent thundering herd
  • Increase PING_TIMEOUT_DURATION from 10s to 30s for stability
  • Graceful cancellation of pending operations on disconnect

Added

  • Added post_init to IImageInference, IImageCaption, IImageUpscale, IVideoInference, I3dInference, IAudioInference, IVideoBackgroundRemoval
  • Added promptExtend: Optional[bool] to ISettings
  • Added IBytedanceProviderSettings.optimizePromptMode: Optional[str]
  • Added settings to IVideoInference dataclass
  • Added draft: Optional[bool], audio: Optional[bool], and promptUpsampling: Optional[bool] to ISettings dataclass
  • Added post_init to ISettings dataclass

[0.4.47]

Added

  • Added IRGB dataclass (rgb: List[int]) with validation (exactly 3 elements, each 0–255)
  • Added IRecraftProviderSettings with styleId: Optional[str], colors: Optional[List[IRGB]], backgroundColor: Optional[IRGB]; added as ImageProviderSettings.
  • Added ISettings.lyrics: Optional[str] and ISettings.guidanceType: Optional[str].
  • Added negativePrompt, seed, steps, CFGScale, and settings to IAudioInference dataclass.
  • Added IAudioSpeech dataclass (text, voice, speed, volume, pitch, emotion, tone: Optional[List[str]]).
  • Added channels: Optional[int] to IAudioSettings.
  • Added speech: Optional[IAudioSpeech] and settings: Optional[ISettings] to IAudioInference.
  • Added languageBoost: Optional[str] and turbo: Optional[bool] to ISettings.
  • Added edit, color, lightDirection, lightType, season (all Optional[str]) to IBriaProviderSettings.

[0.4.46]

Added

  • Added IKlingMultiPrompt: new dataclass with prompt: str and duration: float for Kling multiPrompt payloads.

Changed

  • fileToBase64(): allow application/octet-stream as MIME type for .glb and .ply files when MIME cannot be guessed.
  • Recursive serialization in SerializableMixin.serialize(): replaced flat asdict + None filter with a recursive implementation that:
    • Skips None and keys starting with _
    • Recursively serializes nested SerializableMixin instances
    • Serializes lists/tuples of SerializableMixin as lists of dicts (e.g. for multiPrompt and other list payloads).

[0.4.45]

Added

  • Added ISparseStructure, IShapeSlat, ITexSlat (guidanceStrength, guidanceRescale, steps, rescaleT).
  • Added I3dInference.settings: Optional[ISettings] and outputQuality: Optional[int].
  • Added I3dInputs.meshFile: Optional[Union[str, File]].

Changed

  • Extended ISettings with 3D fields: textureSize, decimationTarget, remesh, resolution, sparseStructure, shapeSlat, texSlat.

[0.4.44]

Added

  • Added support for textInference task type (TEXT_INFERENCE in ETaskType enum)
  • Added ITextInferenceMessage with role and content for chat messages
  • Added ITextInferenceUsage for token usage (promptTokens, completionTokens, totalTokens, thinkingTokens)
  • Added IGoogleTextProviderSettings with thinkingLevel for Gemini
  • Added TextProviderSettings type alias for text inference provider settings
  • Added ITextInference for requests (model, messages, taskUUID, deliveryMethod, maxTokens, temperature, topP, topK, seed, stopSequences, includeCost, providerSettings)
  • Added IText for responses (taskType, taskUUID, text, finishReason, usage, cost, status)
  • Added textInference() to Runware
  • Added _buildTextRequest, _requestText, _handleInitialTextResponse in base
  • Added _addTextProviderSettings for provider settings (matching image/video/audio pattern)
  • Added getResponse() support for List[IText] when polling textInference tasks
  • Added TEXT_INITIAL_TIMEOUT for async delivery (configurable via RUNWARE_TEXT_INITIAL_TIMEOUT)
  • Added TEXT_POLLING_DELAY for polling cadence (configurable via RUNWARE_TEXT_POLLING_DELAY)
  • Added IVideoInference.scheduler: Optional[str] = None

Changed

  • Enabled async delivery for text inference (returns IAsyncTaskResponse and uses getResponse() for polling)

[0.4.43]

Added

  • Added support for "3dInference" task type
  • Added I3dInputs dataclass with image and mask parameters for 3D inference inputs
  • Added I3dOutputFormat type: Literal["GLB", "PLY"]
  • Added I3dInference dataclass for 3D inference requests (model, positivePrompt, seed, inputs, outputFormat, etc.)
  • Added IObject dataclass with uuid and url for 3D output files
  • Added I3dOutput dataclass with files: Optional[List[IObject]] = None for 3D output files in responses
  • Added I3d dataclass for 3D inference responses (taskType, taskUUID, cost, status, seed, outputs)
  • Added inference3d() method to Runware for 3D inference
  • Added getResponse() support for List[I3d] when polling 3dInference tasks
  • Added _handleInitial3dResponse, _process3dInputs, _build3dRequest, _request3d in base
  • Added IUltralytics dataclass with the following parameters:
    • maskBlur: Optional[int]
    • maskPadding: Optional[int]
    • confidence: Optional[float]
    • positivePrompt: Optional[str]
    • negativePrompt: Optional[str]
    • steps: Optional[int]
    • CFGScale: Optional[float]
    • strength: Optional[float]
  • Added IImageInference.ultralytics: Optional[IUltralytics]
  • Added quality: Optional[str] = None to ISettings
  • Added audio: Optional[bool] = None to IViduProviderSettings

Changed

  • Split IOutput into media-specific output types: IOutput (video: draftId, videoId) and I3dOutput (3D: files)

[0.4.42]

Added

  • Added IOutput dataclass with the following parameters:
    • draftId: Optional[str] = None
    • videoId: Optional[str] = None
  • Added IVideo.outputs: Optional[IOutput] = None to support draft ID in video responses
  • Added IVideoInputs.draftId: Optional[str] = None for passing draft task ID in video inference inputs for bytedance:seedance@1.5-pro
  • Added IVideoInputs.videoId: Optional[str] = None for passing video ID in video inference inputs for openai:3@1
  • Added IBytedanceProviderSettings.draft: Optional[bool] = None to enable draft mode for Seedance video generation
  • Added IInputs.mask: Optional[Union[str, File]] = None
  • Added ISourcefulProviderSettings with the following parameters:
    • transparency: Optional[bool] = None
    • enhancePrompt: Optional[bool] = None
    • fontInputs: Optional[List[Dict[str, Any]]] = None
  • Added IInputs.superResolutionReferences: Optional[List[Union[str, File]]] = None for super resolution guidance in image-to-image inference

Changed

  • Enhanced instantiateDataclass() function to automatically handle nested dataclasses

[0.4.41]

Changed

  • Updated IBriaProviderSettings.preserveAudio default from Optional[bool] = True to None
  • Updated IBriaProviderSettings.autoTrim default from Optional[bool] = False to None

Added

  • Added taskUUID attribute to IUploadModelBaseType to allow passing taskUUID from user in modelUpload() method

[0.4.40]

Added

  • Added _retry_with_reconnect() method that wraps all public API methods for automatic reconnection on authentication errors (up to 10 retries)
  • Added MAX_RETRY_ATTEMPTS constant in runware/utils.py for retry mechanism configuration
  • Added IMireloProviderSettings with startOffset: Optional[int] = None parameter
  • Added ISettings.layers: Optional[int] = None parameter
  • Added ISettings.trueCFGScale: Optional[float] = None parameter
  • Added IAudio.videoUUID: Optional[str] = None field
  • Added IAudio.videoURL: Optional[str] = None field
  • Added IAudio.seed: Optional[int] = None field

Fixed

  • Fixed duplicate heartbeat tasks by cancelling existing task before creating new one in runware/server.py
  • Fixed connection loss detection during request waits with session UUID tracking
  • Fixed conflictTaskUUID error handling to extract taskType and deliveryMethod from error context instead of relying on global message state

Changed

  • Refactored retry logic: Removed redundant asyncRetry calls from 9 methods and them wrapped with _retry_with_reconnect

[0.4.39]

Added

  • Added MODEL_UPLOAD_TIMEOUT constant in runware/utils.py with 15-minute default (900000ms) for large model uploads
  • Added RUNWARE_MODEL_UPLOAD_TIMEOUT environment variable support for configuring model upload timeout

Changed

  • Updated _modelUpload() in runware/base.py to use MODEL_UPLOAD_TIMEOUT instead of general timeout to prevent timeouts during large LoRA model uploads

[0.4.38]

Added

  • Added IWanAnimate dataclass with the following parameters:
    • mode: Optional[str] = None
    • retargetPose: Optional[bool] = None
    • prevSegCondFrames: Optional[int] = None
  • Added VideoAdvancedFeatureTypes union type: VideoAdvancedFeatureTypes = IWanAnimate
  • Updated IVideoAdvancedFeatures to support advancedFeature: Optional[VideoAdvancedFeatureTypes] = None for incorporating WanAnimate features
  • Added IVideoInputs.referenceVoices: Optional[List[str]] = None
  • Added IKlingAIProviderSettings.characterOrientation: Optional[str] = None

Fixed

  • Fixed NoneType error when initial requests timeout by returning IAsyncTaskResponse object instead of processing None values
  • Added error handling in _handleInitialVideoResponse, _handleInitialImageResponse, and _handleInitialAudioResponse to handle empty/None initial responses
  • Added validation in instantiateDataclassList() method to raise descriptive error when data_list is None or empty
  • Improved error messages for timeout scenarios with detailed context including TaskUUID and delivery method

Changed

  • Refactored serialization: Removed BaseRequestField class and moved to_request_dict() method to SerializableMixin
  • Updated multiple dataclasses to inherit from SerializableMixin instead of BaseRequestField:
    • IAcceleratorOptions
    • ISafety
    • ISettings
    • IInputs
    • IVideoInputs
    • IAudioSettings
    • IVideoSpeechSettings
    • IAudioInputs
  • Updated IVideoAdvancedFeatures.serialize() method to properly handle advancedFeature serialization using to_request_dict()

[0.4.37]

Added

  • Added audio parameter to IBytedanceProviderSettings (providerSettings.bytedance.audio)
  • Added new parameters to IBriaProviderSettings:
    • preserveAudio
    • autoTrim
  • Added IBriaMaskSettings with the following parameters:
    • foreground
    • prompt
    • frameIndex
    • keyPoints

Changed

  • Added validation in _pollResults(): if number_results is None, then number_results = 1

[0.4.36]

Added

  • Added IBlackForestLabsProviderSettings with safetyTolerance parameter
  • Added new ISettings class with the following parameters:
    • temperature
    • systemPrompt
    • topP
  • Added search parameter to IGoogleProviderSettings
  • Added new parameters to IAlibabaProviderSettings:
    • promptExtend
    • audio
    • shotType

Changed

  • Updated IInputs.references default value from field(default_factory=list) to None
  • Updated IVideoInputs.references default value from field(default_factory=list) to None

[0.4.35]

Added

  • Added provider setting for Alibaba: alibaba.promptEnhancer
  • Added IAudioInput and ISpeechInput to IVideoInputs for video inference
  • Added ISyncProviderSettings to video inference
  • Added sound parameter to IKlingAIProviderSettings (providerSetting.Kling.sound)

Changed

  • Updated referenceImages type from Optional[List[IInputReference]] to Optional[List[Union[str, File, IInputReference]]] in IInputs
  • Renamed IPixverseSpeechSettings to IVideoSpeechSettings with backward compatibility
  • Made numberResults optional in videoInference