{
  "Notes": "Notes\nThis only applies to the task creation interface; query interfaces do not consume concurrency.\nThis limitation concerns the number of concurrent tasks and is unrelated to Queries Per Second(QPS)— the system imposes no QPS limit.\nCore Rules\nDimension\tRule Description\nApplication Scope\tApplied at the account level. Calculated independently per resource pack type (video/image/virtual try-on). All API keys under the same account share the same concurrency quota.\nOccupancy Logic\tA task occupies concurrency from entering submitted status until completion (including failures). Released immediately after task ends.\nQuota Calculation\tDetermined by the highest concurrency value among all active resource packages of the same type. Example: If a 5-concurrency + 10-concurrency video package are both active → video concurrency capacity = 10\nSpecial Notes\nVideo / Virtual Try-on tasks: Each task occupies 1 concurrency.\nImage generation tasks: Concurrency used = the n value in the API request parameter. (Example: n = 9 → occupies 9 concurrency)\nOver-limit Error Mechanism\nWhen the number of running tasks reaches the concurrency limit, submitting a request will return an error.\nJSON\nCopy\nCollapse\n{\n\"code\": 1303,\n\"message\": \"parallel task over resource pack limit\",\n\"request_id\": \"9984d27b-a408-4073-ae28-17ca6a13622d\" //uuid\n}\nRecommended Approach\nSince this error is triggered by system load (not by parameter issues), it is recommended to:\nBackoff Retry Strategy: Use an exponential backoff algorithm to delay retries (recommended initial delay ≥ 1 second).\nQueue Management: Control the submission rate through a task queue and dynamically adapt to available concurrency.\nPrevious chapter：General Info\nNext chapter：Callback Schema\nWhat is Kling API concurrency?\nCore Rules\nOver-limit Error Mechanism\nRecommended Approach\nThe Kling 3.0 Series Models API is Now Fully Available\n– All in One, One for All！\nModels Available in This Release\nKling 3.0 Motion Control, Kling Video 3.0, Kling Video 3.0 Omni, Kling Image 3.0, Kling Image 3.0 Omni\nRefer to <Kling AI Series 3.0 Model API Specification>\nKey Highlights of the Models\n3.0 All-in-One: A unified model for multi-modal input and output.\nMost powerful consistency across the universe: Subject consistency (supports cameo, subject with voice control, i2v + subject) and text consistency.\nNarrative control at your fingertips: More freedom, precision, and control—up to 15 seconds long, video scene cuts, ultra-high-definition storyboards/images, custom seconds.\nUpgraded native audio-visual output: Supports multiple speakers and languages (with accents).\nKling 3.0 Motion Control\nConsistent Facial Identity from any angle\nComplex Emotions faithfully reproduced\nHigh fidelity Restoration, Even with Face Occlusions\nConsistent Facial Clarity Across Dynamic Framing\nUser Guide ->\nKling Video 3.0\nCompared to 2.6, expected improvements:\nSupports subject upload in I2V scenarios for enhanced consistency\nSignificant improvement in multi-character referencing, especially for three-person scenarios\nSupports Japanese, Korean, and Spanish in addition to Chinese and English\nCapable of generating certain dialects and accents\nBetter distinction and control over different types of audio (speech, sound effects, BGM)\nImproved text retention in I2V scenarios\nSupports scene transitions, with up to 6 shots and customizable storyboarding\nUser Guide ->\nKling Video 3.0 Omni\nCompared to O1, expected improvements:\nNative audio-visual synchronization\nSupports video subject creation\nFurther improved consistency in reference-based tasks, especially for characters and products\nCombined capabilities of reference + storyboarding + audio-visual sync significantly enhance usability\nSupports scene transitions, with up to 6 shots\nExtended generation duration up to 15 seconds\nUser Guide ->\nKling Image 3.0\nHighly consistent feature retention\nPrecise response to detail modifications\nAccurate control over style and tone\nRich imaginative capabilities\nUser Guide ->\nKling Image 3.0 Omni\nEnhanced narrative sense\nNew storyboard image set generation, retaining reference image features with scene relevance\nDirect output of 2K/4K ultra-high-definition images\nFurther improved detail consistency\nUser Guide ->\nThank you for your support and understanding!\nI Got It",
  "Concurrency Rules": "Concurrency Rules\nWhat is Kling API concurrency?\nKling API concurrency refers to the maximum number of generation tasks that an account can process in parallel at any given time. This capability is determined by the resource package. A higher concurrency level allows you to submit more API generation requests simultaneously (each call to the task creation interface initiates a new generation task).\n💡\nNotes\nThis only applies to the task creation interface; query interfaces do not consume concurrency.\nThis limitation concerns the number of concurrent tasks and is unrelated to Queries Per Second(QPS)— the system imposes no QPS limit.\nCore Rules\nDimension\tRule Description\nApplication Scope\tApplied at the account level. Calculated independently per resource pack type (video/image/virtual try-on). All API keys under the same account share the same concurrency quota.\nOccupancy Logic\tA task occupies concurrency from entering submitted status until completion (including failures). Released immediately after task ends.\nQuota Calculation\tDetermined by the highest concurrency value among all active resource packages of the same type. Example: If a 5-concurrency + 10-concurrency video package are both active → video concurrency capacity = 10\nSpecial Notes\nVideo / Virtual Try-on tasks: Each task occupies 1 concurrency.\nImage generation tasks: Concurrency used = the n value in the API request parameter. (Example: n = 9 → occupies 9 concurrency)\nOver-limit Error Mechanism\nWhen the number of running tasks reaches the concurrency limit, submitting a request will return an error.\nJSON\nCopy\nCollapse\n{\n\"code\": 1303,\n\"message\": \"parallel task over resource pack limit\",\n\"request_id\": \"9984d27b-a408-4073-ae28-17ca6a13622d\" //uuid\n}\nRecommended Approach\nSince this error is triggered by system load (not by parameter issues), it is recommended to:\nBackoff Retry Strategy: Use an exponential backoff algorithm to delay retries (recommended initial delay ≥ 1 second).\nQueue Management: Control the submission rate through a task queue and dynamically adapt to available concurrency.\nPrevious chapter：General Info\nNext chapter：Callback Schema\nWhat is Kling API concurrency?\nCore Rules\nOver-limit Error Mechanism\nRecommended Approach\nThe Kling 3.0 Series Models API is Now Fully Available\n– All in One, One for All！\nModels Available in This Release\nKling 3.0 Motion Control, Kling Video 3.0, Kling Video 3.0 Omni, Kling Image 3.0, Kling Image 3.0 Omni\nRefer to <Kling AI Series 3.0 Model API Specification>\nKey Highlights of the Models\n3.0 All-in-One: A unified model for multi-modal input and output.\nMost powerful consistency across the universe: Subject consistency (supports cameo, subject with voice control, i2v + subject) and text consistency.\nNarrative control at your fingertips: More freedom, precision, and control—up to 15 seconds long, video scene cuts, ultra-high-definition storyboards/images, custom seconds.\nUpgraded native audio-visual output: Supports multiple speakers and languages (with accents).\nKling 3.0 Motion Control\nConsistent Facial Identity from any angle\nComplex Emotions faithfully reproduced\nHigh fidelity Restoration, Even with Face Occlusions\nConsistent Facial Clarity Across Dynamic Framing\nUser Guide ->\nKling Video 3.0\nCompared to 2.6, expected improvements:\nSupports subject upload in I2V scenarios for enhanced consistency\nSignificant improvement in multi-character referencing, especially for three-person scenarios\nSupports Japanese, Korean, and Spanish in addition to Chinese and English\nCapable of generating certain dialects and accents\nBetter distinction and control over different types of audio (speech, sound effects, BGM)\nImproved text retention in I2V scenarios\nSupports scene transitions, with up to 6 shots and customizable storyboarding\nUser Guide ->\nKling Video 3.0 Omni\nCompared to O1, expected improvements:\nNative audio-visual synchronization\nSupports video subject creation\nFurther improved consistency in reference-based tasks, especially for characters and products\nCombined capabilities of reference + storyboarding + audio-visual sync significantly enhance usability\nSupports scene transitions, with up to 6 shots\nExtended generation duration up to 15 seconds\nUser Guide ->\nKling Image 3.0\nHighly consistent feature retention\nPrecise response to detail modifications\nAccurate control over style and tone\nRich imaginative capabilities\nUser Guide ->\nKling Image 3.0 Omni\nEnhanced narrative sense\nNew storyboard image set generation, retaining reference image features with scene relevance\nDirect output of 2K/4K ultra-high-definition images"
}