NewYour coding agent can read the release notes before it upgrades.Set up the MCP server →
PyPI
PyPI
The Python Runware SDK is used to interact with the Runware API, powered by the Runware inference platform. It supports image generation, video generation, image upscale, video upscale, image caption, video caption, image background removal, video background removal, audio generation, and more. It also allows the use of an existing gallery of models or selecting any model or LoRA from the CivitAI gallery. The API also supports inpainting, outpainting, and a series of other ControlNet models.
Latest stable
0.5.24
Last release 4 months ago
30 Jun 2026
Ships on a steady schedule
a new release about every 1 weeks
Most releases are documented
notes for 38 of the last 60 stable releases
Nothing withdrawn
no release was ever pulled
3 years old
89 releases · first in 2024
Drag across the ruler to compare, or tap a release to open it.
chore: bump version to 0.5.24
chore: bump version to 0.5.24
IAudioInputs now includes:
referenceAudios: Optional[List[str]]referenceImages: Optional[List[str]]audioInference now processes inputs.referenceAudios, inputs.referenceImages, and inputs.referenceVoices[].audio via _process_media before sending the requestThis project publishes no release notes. Between v0.5.23 and v0.5.24 there were 3 commits, 1 of them substantive:
chore: bump version to 0.5.23
chore: bump version to 0.5.23
Your coding agent can read these notes before it upgrades. Set up the MCP server →
This project publishes no release notes. Between v0.5.22 and v0.5.23 there were 3 commits, 1 of them substantive:
chore: bump version to 0.5.22
chore: bump version to 0.5.22
IInputs now includes:
seedImage: Optional[Union[str, File]]imageInference now processes inputs.seedImage via _process_media before sending the requestThis project publishes no release notes. Between v0.5.21 and v0.5.22 there were 3 commits, 1 of them substantive:
chore: bump version to 0.5.21
chore: bump version to 0.5.21
IVideoEditSettingsControls now uses flat fields for edit controls instead of nested per-control dataclasses:
poseStrength: Optional[str]depthBlur: Optional[float]normalsAugmentation: Optional[float]trajectorySparsity: Optional[float]face: Optional[bool]IVideoEditSettingsControls fields:
IVideoEditPoseControlIVideoEditDepthControlIVideoEditNormalsControlIVideoEditTrajectoryControlIVideoEditFaceControlThis project publishes no release notes. Between v0.5.20 and v0.5.21 there were 3 commits, 1 of them substantive:
chore: bump version to 0.5.20
chore: bump version to 0.5.20
ISettings now includes:
preserveInputSize: Optional[bool]This project publishes no release notes. Between v0.5.19 and v0.5.20 there were 5 commits, 2 of them substantive:
chore: bump version to 0.5.19
chore: bump version to 0.5.19
ISourcePositionIVideoEditPoseControlIVideoEditDepthControlIVideoEditNormalsControlIVideoEditTrajectoryControlIVideoEditFaceControlIVideoEditSettingsControlsIVideoEditSettingsITextInferenceMessageTool (toolId, name, toolInput; serialized as id, name, input)ITextInferenceMessage extends SerializableMixin with optional content, toolId, and tools; serialize() maps toolId → id on the wireISettings now supports additional video parameters:
edit: Optional[Union[IVideoEditSettings, Dict[str, Any]]]exrExport: Optional[bool]hdr: Optional[bool]loop: Optional[bool]sourcePosition: Optional[Union[ISourcePosition, Dict[str, Any]]]ISettings.__post_init__ for:
edit → IVideoEditSettingssourcePosition → ISourcePosition_buildTextRequest() now builds messages with ITextInferenceMessage.serialize() instead of dataclasses.asdict()ITextInferenceMessage.content is now Optional[str]instantiateDataclass now skips null values from API payloads instead of passing them through as explicit NoneinstantiateDataclass fills missing required str fields with "" when the API leaves them out, so dataclasses like IModel can keep version: str and architecture: str as required fieldsinstantiateDataclass / instantiateDataclassList (IModel, IImage, IVideo, I3d, etc.) use a compact repr that omits unset, None, blank string, and empty additional_fields valuesThis project publishes no release notes. Between v0.5.18 and v0.5.19 there were 12 commits, 7 of them substantive:
chore: bump version to 0.5.18
chore: bump version to 0.5.18
IDracoCompression dataclass for settings.dracoCompression:
enabled: Optional[bool]level: Optional[int]quantizationPosition: Optional[int]quantizationNormal: Optional[int]quantizationTexCoord: Optional[int]ISettings now includes:
dracoCompression: Optional[Union[bool, IDracoCompression, Dict[str, Any]]]IInputs.referenceImages now accepts Dict[str, Any] entries and coerces them to IInputReference in IInputs.__post_init__This project publishes no release notes. Between v0.5.17 and v0.5.18 there were 9 commits, 6 of them substantive:
Add Riverflow 2.5 types for inputs.fonts, settings.scoringRubric, and promptEnhance rename.
Add Riverflow 2.5 types for inputs.fonts, settings.scoringRubric, and promptEnhance rename.
IInputFont dataclass for inputs.fonts[]:
url: Optional[str]text: Optional[str]IInputs now includes:
fonts: Optional[List[Union[IInputFont, Dict[str, Any]]]]IScoringRubric dataclass for settings.scoringRubric[]:
key: Optional[str]label: Optional[str]description: Optional[str]weight: Optional[float]passingScore: Optional[float]scoreGuidance: Optional[List[Union[IScoringRubricScoreGuidance, Dict[str, Any]]]]IScoringRubricScoreGuidance (score, description)ISettings now includes:
enhancePrompt: Optional[bool]backgroundMode: Optional[str]scoringPrompt: Optional[str]scoringRubric: Optional[List[Union[IScoringRubric, Dict[str, Any]]]]ISettings.promptEnhancement renamed to ISettings.promptEnhanceThis project publishes no release notes. Between v0.5.16 and v0.5.17 there were 7 commits, 3 of them substantive:
chore: bump version to 0.5.16
chore: bump version to 0.5.16
IPromptEnhancement dataclass for nested settings.promptEnhancement:
enabled: Optional[bool]temperature: Optional[float]topP: Optional[float]IInputFrame.timestamp: Optional[float] = NoneISettings now includes:
sourceAudioSync: Optional[bool] = NoneimageInference requests via settings.structuredPrompt:
IStructuredPrompt (high_level_description, style_description, compositional_deconstruction)IStructuredPromptCompositionalDeconstruction (background, elements)IStructuredPromptElement (elementType, desc, text, bbox, color_palette); serializes elementType as API typeIStructuredPromptStyleDescription (aesthetics, lighting, photo, medium, art_style, color_palette)ISettings.structuredPrompt: Optional[Union[IStructuredPrompt, Dict[str, Any]]] with dict coercion in ISettings.__post_init__ISettings.copyrightDetection: Optional[bool]IImage:
status: Optional[str]structuredPrompt: Optional[Dict[str, Any]] (returned as plain dict, matching API shape including type on elements)IImageInferenceOutputs.structuredPrompt: Optional[Dict[str, Any]] when the API nests it under outputs_requestVideo now preprocesses inputs.video the same way as inputs.frameImagescreateImageFromResponseThis project publishes no release notes. Between v0.5.15 and v0.5.16 there were 16 commits, 10 of them substantive:
chore: bump version to 0.5.15
chore: bump version to 0.5.15
IMeshCluster dataclass for settings.meshCluster:
thresholdConeHalfAngleRad: Optional[float]refineIterations: Optional[int]globalIterations: Optional[int]smoothStrength: Optional[int]ISettings now includes:
textureFormat: Optional[str]alphaMode: Optional[str]remeshBand: Optional[float]remeshProject: Optional[float]meshCluster: Optional[Union[IMeshCluster, Dict[str, Any]]]moodboards: Optional[List[Union[IMoodboard, Dict[str, Any]]]]
IUltralytics now includes:
inpaintSize: Optional[int]ISettings now includes:
creativity: Optional[str]moodboards: Optional[List[Union[IMoodboard, Dict[str, Any]]]]preserveAudio: Optional[bool]IMoodboard dataclass for settings.moodboards[] entries:
id: strstrength: Optional[float]IInputReference now includes:
role: Optional[str]normalizeLoudness: Optional[bool]
ISettings now includes:
autoCrop: Optional[bool]chunkLength: Optional[int]minChunkLength: Optional[int]normalize: Optional[bool]normalizeLoudness: Optional[bool]latency: Optional[str]conditionOnPreviousChunks: Optional[bool]earlyStopThreshold: Optional[float]IAudioReferenceVoice (audio, text) for reference-voice input on audioInferenceIAudioInputs now includes:
referenceVoices: Optional[List[Union[IAudioReferenceVoice, Dict[str, Any]]]]IAudioSpeech.voices now accepts voice model ID strings in addition to IAudioVoice objectsIAudioSpeech.volume type changed from Optional[int] to Optional[float]IImageInference.outputType now supports Optional[Union[IOutputType, List[IOutputType]]].
IImageInference.outputType now supports Optional[Union[IOutputType, List[IOutputType]]].outputs on IImage:
IImageInferenceOutputs with textBlocks: Optional[List[IImageInferenceTextBlock]]IImageInferenceTextBlock (alignment, angle, color, fontAlternatives, fontName, fontSize, formatting, height, lineHeight, role, text, width, x, y)ISettings now includes:
fit: Optional[str]caption: Optional[bool]IVideoSpeechSettings now includes:
volume: Optional[float]Added support for taskType training.
training.TextInferenceCacheScope = Literal["system", "system+history"]TextInferenceCacheTtl = Literal["5m", "1h"]ITextInferenceCache dataclass (scope, ttl) with request key cacheITextInputs now includes:
documents: Optional[List[Union[str, File]]]ITextInferenceTool now supports:
schema: Optional[Dict[str, Any]]input_schema alias (serialized as schema when schema is not provided)toolType (serialized as type)ITextInferenceToolChoice request serialization support:
request_key = "toolChoice"toolType -> type mapping in serialize()ISettings now includes:
search: Optional[bool] = NonevoicePrompt: Optional[str]safetyFilter: Optional[bool]cfgIntervalStart: Optional[float]cfgIntervalEnd: Optional[float]repaintMode: Optional[str]repaintStrength: Optional[float]IAudioOutputFormat now supports:
oggtoolChoice moved from ISettings to the root ITextInference payload:
ITextInference.toolChoice: Optional[Union[ITextInferenceToolChoice, Dict[str, Any]]]_buildTextRequest() now includes root-level toolChoicerunware/base.py via _processTextInputs() and applied it to both _requestText() and _requestTextStream(), so images/videos/documents are normalized consistently for sync/async/stream text requests.IVideoInference.speech now accepts either:
IVideoSpeechSettingsDict[str, Any] (auto-coerced to IVideoSpeechSettings in __post_init__)Added support for taskType getTaskDetails.
getTaskDetails.IVideoReferenceImage dataclass with:
tag: Optional[str]refType: Optional[str]images: Optional[List[Union[str, File]]]audio: Optional[Union[str, File]]serialize() override that maps refType -> type in outgoing payloads.IVideoReferenceVideo dataclass with:
tag: Optional[str]refType: Optional[str]video: Optional[Union[str, File]]serialize() override that maps refType -> type in outgoing payloads.IVideoInputs to support structured references:
referenceImages: Optional[List[Union[str, File, IVideoReferenceImage]]]referenceVideos: Optional[List[Union[str, IVideoReferenceVideo]]]process_media() in runware/utils.py as the canonical media normalization helper (supports local-file conversion and pass-through for URL/UUID/base64).IVideoInputs.__post_init__:
type are accepted and internally mapped to refType for both referenceImages and referenceVideos.ISettings now includes Meshy 6 / 3D controls:
meshType: Optional[str]topology: Optional[str]decimation: Optional[int]symmetry: Optional[str]pose: Optional[str]removeLighting: Optional[bool]origin: Optional[str]moderation: Optional[bool]texturePrompt: Optional[str]imageEnhancement: Optional[bool]savePreRemeshedModel: Optional[bool]I3dInputs now includes:
referenceImages: Optional[List[Union[str, File]]]ISettings now includes:
keyframe: Optional[int]ISettings now includes additional 3dInference fields:
faceCount: Optional[int]geometryOnly: Optional[bool]generateType: Optional[str]polygonType: Optional[str]I3dInference now includes:
negativePrompt: Optional[str]I3dOutputFormat now supports:
OBJSerializableMixin.serialize() to handle mixed list/tuple payloads by serializing SerializableMixin items item-by-item (instead of requiring all items to be mixin instances).runware/base.py media normalization flow to canonical process_media() semantics (via _process_media delegation).process_image() as a backward-compatible alias to process_media().type usage in video reference dataclasses with refType to avoid reserved-name ambiguity while preserving API payload compatibility (type is still sent on wire via serialize()).I3dOutputFormat now supports Meshy 6 target formats: GLB, OBJ, FBX, STL, USDZ, 3MF (while keeping PLY for compatibility).activeSpeakerDetection: Optional[Union[IActiveSpeakerDetection, Dict[str, Any]]]
ISettings now includes:
syncMode: Optional[str]mode: Optional[str]emotion: Optional[str]activeSpeakerDetection: Optional[Union[IActiveSpeakerDetection, Dict[str, Any]]]occlusionDetection: Optional[bool]segments: Optional[List[Union[ISegment, Dict[str, Any]]]]tts: Optional[Union[ITTSSettings, Dict[str, Any]]]colorCorrection: Optional[bool]renderingSpeed: Optional[str]magicPrompt: Optional[str]background: Optional[str]IActiveSpeakerDetection (autoDetect, frameNumber, coordinates, boundingBoxes) with validation for boundingBoxes entries (must be null or four numeric values).ISegment for Sync segments (startTime, endTime, audio, ref, audioStartTime, audioEndTime).ITTSSettings for Sync TTS tuning (stability, similarityBoost).ISyncSegment = ISegmentISyncActiveSpeakerDetection = IActiveSpeakerDetectionISyncSettings = ISyncProviderSettingsIAudioSpeech now includes:
voices: Optional[List[Union[IAudioVoice, Dict[str, Any]]]]IAudioVoice for multi-speaker TTS mapping:
speaker: strvoice: strIImageInference now includes:
upscaleFactor: Optional[float]ISyncProviderSettings to include Sync 3 fields:
mode: Optional[str]emotion: Optional[str]activeSpeakerDetection: Optional[Union[IActiveSpeakerDetection, Dict[str, Any], bool]]occlusionDetection: Optional[bool]segments: Optional[List[Union[ISegment, Dict[str, Any]]]]IActiveSpeakerDetection, ISegment, and ITTSSettings objects.useOriginalAlpha: Optional[bool]
ISettings now includes:
useOriginalAlpha: Optional[bool]material: Optional[str]polyCount: Optional[float]taPose: Optional[bool]boundingBox: Optional[List[int]]meshMode: Optional[str]addons: Optional[List[str]]hdTexture: Optional[bool]instrumental: Optional[bool]lyricsOptimizer: Optional[bool]IAudioInference now includes:
outputQuality: Optional[int]ITextInputs now includes:
videos: Optional[List[Union[str, File]]]IVideoInputs now includes:
referenceAudios: Optional[List[str]]httpx version dependency during installation in setup.py.setup.py: find_packages excludes the top-level tests package and its subpackages.audioTemperature: Optional[float]
ISettings now includes:
includePrefix: Optional[bool]audioTemperature: Optional[float]topK: Optional[int]IAudioInputs now includes:
audios: Optional[List[str]]`IUpscaleSettings`: deprecated subclass of `ISettings` (deprecation warning on use).
ISettings now includes:
imageAutoFix: Optional[bool]faceLimit: Optional[int]texture: Optional[bool]pbr: Optional[bool]textureSeed: Optional[int]textureAlignment: Optional[str]textureQuality: Optional[str]autoSize: Optional[bool]orientation: Optional[str]quad: Optional[bool]compress: Optional[str]smartLowPoly: Optional[bool]generateParts: Optional[bool]exportUv: Optional[bool]geometryQuality: Optional[str]editRegions: Optional[List[List[List[int]]]]sequential: Optional[bool]thinking: Optional[bool]colorPalette: Optional[List[Union[IColorPaletteEntry, Dict[str, Any]]]]style: Optional[str]thinking: Optional[str]multiClip: Optional[bool]voiceDescription: Optional[str]enhanceDetails, realism, and existing upscale tuning fields on ISettings (steps, seed, CFGScale, prompts, scheduler, etc.)I3dInputs now includes images: Optional[List[Union[str, File]]].IVideoInputs.audios for video inference.IImageUpscale: targetMegapixels, optional upscaleFactor, deliveryMethod (default "sync"); IImageUpscale.settings typed as ISettings.IUpscaleSettings: deprecated subclass of ISettings (deprecation warning on use)._upscaleGan: async deliveryMethod handling for IAsyncTaskResponse + polling.IColorPaletteEntry: hex: str, ratio: Optional[Union[str, float]].IInputs now includes images: Optional[List[Union[str, File]]]._pollResults: imageUpscale, vectorize, and imageBackgroundRemoval resolve to IImage (fixes getResponse returning IVideo without image URLs)._upscaleGan: conditional upscaleFactor / targetMegapixels; forward deliveryMethod.Relax aiofiles and python-dotenv pins to >= minimums in setup.py and requirements.txt so the package installs cleanly alongside environments that pin
aiofiles and python-dotenv pins to >= minimums in setup.py and requirements.txt so the package installs cleanly alongside environments that pin newer compatible releases.Added runware/version.py as the canonical __version__ string; runware/__init__.py imports it so from runware import __version__ matches the package ve
runware/version.py as the canonical __version__ string; runware/__init__.py imports it so from runware import __version__ matches the package version.X-SDK-Name: python and X-SDK-Version: <__version__> (see runware/server.py).Added width: Optional[int], height: Optional[int], and fps: Optional[int] to IVideoUpscale.
width: Optional[int], height: Optional[int], and fps: Optional[int] to IVideoUpscale.IElements dataclass for video inference inputs.elements[] with:
id: Optional[str]description: Optional[str]frontalImage: Optional[Union[str, File]]images: Optional[List[Union[str, File]]]videos: Optional[List[str]]voice: Optional[List[str]]tags: Optional[List[str]]elements: Optional[List[Union[IElements, Dict[str, Any]]]] to IVideoInputs.IAudioOutputFormat extended with wav, mp3, pcm, opus, aac, flac (existing MP3 retained).ISettings: maxNewTokens, transcript, xVectorOnly for reference-audio / ICL TTS flows.IAudioInputs: audio (reference audio URL or data URI).IAudioSpeech: language (e.g. input language for TTS)._requestVideoUpscale() to forward width, height, and fps when provided.IVideoInputs.__post_init__ to coerce inputs.elements dictionary items into IElements instances.Added photoMaker as a nested object in imageInference using IPhotoMakerSettings; IImageInference.photoMaker: Optional[Union[IPhotoMakerSettings, Dict[
IPhotoMakerSettings; IImageInference.photoMaker: Optional[Union[IPhotoMakerSettings, Dict[str, Any]]]. Nested settings use images/inputImages + style/strength. Standalone taskType="photoMaker" remains a separate operation via IPhotoMaker.IIpAdapter.guideImages, combineMethod, weightType, embedScaling, weightComposition; guideImage optional; base supports guideImages with process_image.IEmbedding.weight: Optional[float].IImageInference.post_init for photoMaker, instantID, acePlusPlus, puLID, ultralytics, outpaint, refiner; list coercion for embeddings (dict -> IEmbedding) and ipAdapters (dict -> IIpAdapter).IInputReference.type and IInputReference.strength (for sketch-style refs).inputs.referenceImages can be IInputReference items; each image is run through process_image before send.IGoogleProviderSettings.safetyTolerance.bpm, keyScale, timeSignature, vocalLanguage, coverConditioningScale, repaintingStart, and repaintingEnd to ISettings.audio: Optional[str] to IAudioInputs.IGoogleProviderSettings.resizeMode.IImageInference: refiner, outpaint, instantID, acePlusPlus, puLID, ultralytics, photoMaker typed as Union[Type, Dict[str, Any]]; embeddings as List[Union[IEmbedding, Dict[str, Any]]]; ipAdapters as List[Union[IIpAdapter, Dict[str, Any]]].IVectorize: width, height, positivePrompt, providerSettings (optional).
IVectorize: width, height, positivePrompt, providerSettings (optional).IVectorize.inputs optional (prompt-only vectorize e.g. recraft:v4@vector).VectorizeProviderSettings type alias (IRecraftProviderSettings) in types._processVectorizeInputs and _buildVectorizeRequest in base (vectorize follows video/3D flow)._addProviderSettings(request_object, payload) for image, image background removal, image upscale, vectorize, video, audio, text (replaces _addImageProviderSettings, _addAudioProviderSettings, _addTextProviderSettings).post_init to IAudioInference dataclass for speech = Optional[Union[IAudioSpeech, Dict[str, Any]]].textNormalization: Optional[bool] = None to ISettings dataclass.IAdvancedFeatures: layerDiffuse, hiresFix, watermark (IWatermark), and regionalPrompting (IRegionalPrompting with regions: List[IRegion]).IVideoAdvancedFeatures: IWatermark (text/image, position, opacity, colors) and watermark field.IFluxKontext, IRegion, IRegionalPrompting, IWatermark, IAdvancedFeatures, and IVideoAdvancedFeatures use SerializableMixin for request serialization.avatar: Optional[str], background: Optional[str] to IVideoInputs.speed: Optional[float], pitch: Optional[float], language: Optional[str] to IVideoSpeechSettings.expressiveness: Optional[str], removeBackground: Optional[bool], backgroundColor: Optional[str] to ISettings.inputs resolved via process_image when present; optional and added with _addOptionalField._addProviderSettings payload type: Union[IImageInference, IImageBackgroundRemoval, IImageUpscale, IVectorize, IVideoInference, IAudioInference, ITextInference]._addOptionalBuiltInDataTypesFields() instead of hardcoded optional field lists, ensuring fields like acceleration and webhookURL are always forwarded when set.Replace polling with asyncio.Future for O(1) message routing via _pending_operations HashMap
_pending_operations HashMapRUNWARE_MAX_CONCURRENT_REQUESTS)PING_TIMEOUT_DURATION from 10s to 30s for stabilitypost_init to IImageInference, IImageCaption, IImageUpscale, IVideoInference, I3dInference, IAudioInference, IVideoBackgroundRemovalpromptExtend: Optional[bool] to ISettingsIBytedanceProviderSettings.optimizePromptMode: Optional[str]settings to IVideoInference dataclassdraft: Optional[bool], audio: Optional[bool], and promptUpsampling: Optional[bool] to ISettings dataclasspost_init to ISettings dataclassAdded IRGB dataclass (rgb: List[int]) with validation (exactly 3 elements, each 0–255)
IRGB dataclass (rgb: List[int]) with validation (exactly 3 elements, each 0–255)IRecraftProviderSettings with styleId: Optional[str], colors: Optional[List[IRGB]], backgroundColor: Optional[IRGB]; added as ImageProviderSettings.ISettings.lyrics: Optional[str] and ISettings.guidanceType: Optional[str].negativePrompt, seed, steps, CFGScale, and settings to IAudioInference dataclass.IAudioSpeech dataclass (text, voice, speed, volume, pitch, emotion, tone: Optional[List[str]]).channels: Optional[int] to IAudioSettings.speech: Optional[IAudioSpeech] and settings: Optional[ISettings] to IAudioInference.languageBoost: Optional[str] and turbo: Optional[bool] to ISettings.edit, color, lightDirection, lightType, season (all Optional[str]) to IBriaProviderSettings.Added IKlingMultiPrompt: new dataclass with prompt: str and duration: float for Kling multiPrompt payloads.
IKlingMultiPrompt: new dataclass with prompt: str and duration: float for Kling multiPrompt payloads.fileToBase64(): allow application/octet-stream as MIME type for .glb and .ply files when MIME cannot be guessed.SerializableMixin.serialize(): replaced flat asdict + None filter with a recursive implementation that:
None and keys starting with _SerializableMixin instancesSerializableMixin as lists of dicts (e.g. for multiPrompt and other list payloads).Added ISparseStructure, IShapeSlat, ITexSlat (guidanceStrength, guidanceRescale, steps, rescaleT).
ISparseStructure, IShapeSlat, ITexSlat (guidanceStrength, guidanceRescale, steps, rescaleT).I3dInference.settings: Optional[ISettings] and outputQuality: Optional[int].I3dInputs.meshFile: Optional[Union[str, File]].ISettings with 3D fields: textureSize, decimationTarget, remesh, resolution, sparseStructure, shapeSlat, texSlat.Added support for textInference task type (TEXT_INFERENCE in ETaskType enum)
ITextInferenceMessage with role and content for chat messagesITextInferenceUsage for token usage (promptTokens, completionTokens, totalTokens, thinkingTokens)IGoogleTextProviderSettings with thinkingLevel for GeminiTextProviderSettings type alias for text inference provider settingsITextInference for requests (model, messages, taskUUID, deliveryMethod, maxTokens, temperature, topP, topK, seed, stopSequences, includeCost, providerSettings)IText for responses (taskType, taskUUID, text, finishReason, usage, cost, status)textInference() to Runware_buildTextRequest, _requestText, _handleInitialTextResponse in base_addTextProviderSettings for provider settings (matching image/video/audio pattern)getResponse() support for List[IText] when polling textInference tasksTEXT_INITIAL_TIMEOUT for async delivery (configurable via RUNWARE_TEXT_INITIAL_TIMEOUT)TEXT_POLLING_DELAY for polling cadence (configurable via RUNWARE_TEXT_POLLING_DELAY)IVideoInference.scheduler: Optional[str] = NoneAdded support for "3dInference" task type
I3dInputs dataclass with image and mask parameters for 3D inference inputsI3dOutputFormat type: Literal["GLB", "PLY"]I3dInference dataclass for 3D inference requests (model, positivePrompt, seed, inputs, outputFormat, etc.)IObject dataclass with uuid and url for 3D output filesI3dOutput dataclass with files: Optional[List[IObject]] = None for 3D output files in responsesI3d dataclass for 3D inference responses (taskType, taskUUID, cost, status, seed, outputs)inference3d() method to Runware for 3D inferencegetResponse() support for List[I3d] when polling 3dInference tasks_handleInitial3dResponse, _process3dInputs, _build3dRequest, _request3d in baseIUltralytics 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]IImageInference.ultralytics: Optional[IUltralytics]quality: Optional[str] = None to ISettingsaudio: Optional[bool] = None to IViduProviderSettingsIOutput into media-specific output types: IOutput (video: draftId, videoId) and I3dOutput (3D: files)Added IOutput dataclass with the following parameters:
IOutput dataclass with the following parameters:
draftId: Optional[str] = NonevideoId: Optional[str] = NoneIVideo.outputs: Optional[IOutput] = None to support draft ID in video responsesIVideoInputs.draftId: Optional[str] = None for passing draft task ID in video inference inputs for bytedance:seedance@1.5-proIVideoInputs.videoId: Optional[str] = None for passing video ID in video inference inputs for openai:3@1IBytedanceProviderSettings.draft: Optional[bool] = None to enable draft mode for Seedance video generationIInputs.mask: Optional[Union[str, File]] = NoneISourcefulProviderSettings with the following parameters:
transparency: Optional[bool] = NoneenhancePrompt: Optional[bool] = NonefontInputs: Optional[List[Dict[str, Any]]] = NoneIInputs.superResolutionReferences: Optional[List[Union[str, File]]] = None for super resolution guidance in image-to-image inferenceinstantiateDataclass() function to automatically handle nested dataclassesUpdated IBriaProviderSettings.preserveAudio default from Optional[bool] = True to None
IBriaProviderSettings.preserveAudio default from Optional[bool] = True to NoneIBriaProviderSettings.autoTrim default from Optional[bool] = False to NonetaskUUID attribute to IUploadModelBaseType to allow passing taskUUID from user in modelUpload() methodAdded _retry_with_reconnect() method that wraps all public API methods for automatic reconnection on authentication errors (up to 10 retries)
_retry_with_reconnect() method that wraps all public API methods for automatic reconnection on authentication errors (up to 10 retries)MAX_RETRY_ATTEMPTS constant in runware/utils.py for retry mechanism configurationIMireloProviderSettings with startOffset: Optional[int] = None parameterISettings.layers: Optional[int] = None parameterISettings.trueCFGScale: Optional[float] = None parameterIAudio.videoUUID: Optional[str] = None fieldIAudio.videoURL: Optional[str] = None fieldIAudio.seed: Optional[int] = None fieldrunware/server.pyconflictTaskUUID error handling to extract taskType and deliveryMethod from error context instead of relying on global message stateasyncRetry calls from 9 methods and them wrapped with _retry_with_reconnectAdded MODEL_UPLOAD_TIMEOUT constant in runware/utils.py with 15-minute default (900000ms) for large model uploads
MODEL_UPLOAD_TIMEOUT constant in runware/utils.py with 15-minute default (900000ms) for large model uploadsRUNWARE_MODEL_UPLOAD_TIMEOUT environment variable support for configuring model upload timeout_modelUpload() in runware/base.py to use MODEL_UPLOAD_TIMEOUT instead of general timeout to prevent timeouts during large LoRA model uploadsAdded IWanAnimate dataclass with the following parameters:
IWanAnimate dataclass with the following parameters:
mode: Optional[str] = NoneretargetPose: Optional[bool] = NoneprevSegCondFrames: Optional[int] = NoneVideoAdvancedFeatureTypes union type: VideoAdvancedFeatureTypes = IWanAnimateIVideoAdvancedFeatures to support advancedFeature: Optional[VideoAdvancedFeatureTypes] = None for incorporating WanAnimate featuresIVideoInputs.referenceVoices: Optional[List[str]] = NoneIKlingAIProviderSettings.characterOrientation: Optional[str] = NoneIAsyncTaskResponse object instead of processing None values_handleInitialVideoResponse, _handleInitialImageResponse, and _handleInitialAudioResponse to handle empty/None initial responsesinstantiateDataclassList() method to raise descriptive error when data_list is None or emptyBaseRequestField class and moved to_request_dict() method to SerializableMixinSerializableMixin instead of BaseRequestField:
IAcceleratorOptionsISafetyISettingsIInputsIVideoInputsIAudioSettingsIVideoSpeechSettingsIAudioInputsIVideoAdvancedFeatures.serialize() method to properly handle advancedFeature serialization using to_request_dict()Added audio parameter to IBytedanceProviderSettings (providerSettings.bytedance.audio)
audio parameter to IBytedanceProviderSettings (providerSettings.bytedance.audio)IBriaProviderSettings:
preserveAudioautoTrimIBriaMaskSettings with the following parameters:
foregroundpromptframeIndexkeyPoints_pollResults(): if number_results is None, then number_results = 1Added IBlackForestLabsProviderSettings with safetyTolerance parameter
IBlackForestLabsProviderSettings with safetyTolerance parameterISettings class with the following parameters:
temperaturesystemPrompttopPsearch parameter to IGoogleProviderSettingsIAlibabaProviderSettings:
promptExtendaudioshotTypeIInputs.references default value from field(default_factory=list) to NoneIVideoInputs.references default value from field(default_factory=list) to NoneAdded provider setting for Alibaba: alibaba.promptEnhancer
alibaba.promptEnhancerIAudioInput and ISpeechInput to IVideoInputs for video inferenceISyncProviderSettings to video inferencesound parameter to IKlingAIProviderSettings (providerSetting.Kling.sound)referenceImages type from Optional[List[IInputReference]] to Optional[List[Union[str, File, IInputReference]]] in IInputsIPixverseSpeechSettings to IVideoSpeechSettings with backward compatibilitynumberResults optional in videoInferenceNothing published for this version
Nothing published for this version
Nothing published for this version
Nothing published for this version
Nothing published for this version
Nothing published for this version
Nothing published for this version
Nothing published for this version
Nothing published for this version
Nothing published for this version
Nothing published for this version
Nothing published for this version
Nothing published for this version
Nothing published for this version
Nothing published for this version
Nothing published for this version
Nothing published for this version
Nothing published for this version
Nothing published for this version
Nothing published for this version
Nothing published for this version
Nothing published for this version
One column per month.