Kling 3.0 & O3 — AI Video Generation by Kuaishou

Generate, animate, and edit AI videos using Kuaishou's Kling 3.0 and Kling Video O3 — featuring cinematic motion quality, realistic physics simulation, reference-based generation, and natural-language video editing.

Kling 3.0 excels at creating cinematic short clips with realistic motion, complex camera movements, and faithful prompt adherence. Kling Video O3 adds MVL (Multi-modal Visual Language) technology with reference-based generation and video editing capabilities. All models support optional synchronized sound generation.

Data usage note: This skill sends text prompts, image URLs, and video URLs to the Atlas Cloud API (api.atlascloud.ai) for video generation and editing. No data is stored locally beyond the downloaded output files. API usage incurs charges per second based on the model selected.

---

Key Capabilities

• - Text-to-Video — Generate video clips from text descriptions
• Image-to-Video — Animate still images into dynamic video with first/last frame control
• Reference-to-Video — Generate videos using character, prop, or scene reference images (O3)
• Video Editing — Natural-language video editing: remove/replace objects, change backgrounds, add effects (O3)
• Sound Generation — Optional synchronized sound effects and audio
• Pro & Standard Tiers — Pro for highest quality, Standard for cost-effective production
• Multiple Aspect Ratios — 16:9, 9:16, 1:1
• Flexible Duration — V3: 5 or 10 seconds; O3: 3-15 seconds
• Negative Prompts — Specify what to exclude from generated video (V3)

---

Setup

1. 1. Sign up at https://www.atlascloud.ai
2. Console → API Keys → Create new key
3. Set env: INLINECODE1

---

Script Usage

This skill includes a Python script for video generation. Zero external dependencies required.

List available video models

CODEBLOCK0

Generate a video (text-to-video)

CODEBLOCK1

Generate a video (image-to-video)

CODEBLOCK2

Upload a local file

CODEBLOCK3

Run python scripts/generate_video.py generate --help for all options. Extra model params can be passed as key=value (e.g. duration=10 shot_type=multi_camera).

---

Pricing

All prices are per second of video generated. Atlas Cloud offers 15% off compared to standard API pricing.

Kling V3.0

Model	Tier	Original Price	Atlas Cloud	Best For
INLINECODE4	Standard	~~$0.18/s~~	$0.153/s	Cost-effective text-to-video
INLINECODE5

Standard | ~~$0.18/s~~ | $0.153/s | Cost-effective image animation | | kwaivgi/kling-v3.0-pro/text-to-video | Pro | ~~$0.24/s~~ | $0.204/s | High-quality text-to-video | | kwaivgi/kling-v3.0-pro/image-to-video | Pro | ~~$0.24/s~~ | $0.204/s | High-quality image animation |

Kling Video O3 Pro

Model	Original Price	Atlas Cloud	Best For
INLINECODE8	~~$0.24/s~~	$0.204/s	MVL-enhanced text-to-video
INLINECODE9

~~$0.24/s~~ | $0.204/s | MVL-enhanced image animation | | kwaivgi/kling-video-o3-pro/reference-to-video | ~~$0.24/s~~ | $0.204/s | Reference-based video generation | | kwaivgi/kling-video-o3-pro/video-edit | ~~$0.36/s~~ | $0.306/s | Professional video editing |

Kling Video O3 Standard

Model	Original Price	Atlas Cloud	Best For
INLINECODE12	-	$0.153/s	Cost-effective MVL text-to-video
INLINECODE13

- | $0.153/s | Cost-effective MVL image animation | | kwaivgi/kling-video-o3-std/reference-to-video | - | $0.085/s | Cost-effective reference-based generation | | kwaivgi/kling-video-o3-std/video-edit | - | $0.238/s | Budget video editing |

---

Parameters

Kling V3.0 — Text-to-Video

Parameter	Type	Required	Default	Options
INLINECODE16	string	Yes	-	Video description
INLINECODE17

string | No | - | What to exclude from the video | | duration | integer | No | 5 | 5, 10 seconds | | aspect_ratio | string | No | 16:9 | 16:9, 9:16, 1:1 | | cfg_scale | number | No | 0.5 | 0-1, controls prompt adherence | | sound | boolean | No | false | Generate synchronized audio |

Kling V3.0 — Image-to-Video

Same as V3.0 text-to-video, plus:

Parameter	Type	Required	Description
INLINECODE22	string	Yes	URL of the source image (jpg/jpeg/png, max 10MB, min 300px, aspect ratio 1:2.5 to 2.5:1)
INLINECODE23

string | No | URL of the target end frame (for guided motion) |

Kling Video O3 — Text-to-Video

Parameter	Type	Required	Default	Options
INLINECODE24	string	Yes	-	Video description
INLINECODE25

string | No | 16:9 | 16:9, 9:16, 1:1 | | duration | integer | No | 5 | 3-15 seconds | | sound | boolean | No | false | Generate synchronized audio |

Kling Video O3 — Image-to-Video

Parameter	Type	Required	Default	Description
INLINECODE28	string	Yes	-	Video description
INLINECODE29

string | Yes | - | First frame image URL | | end_image | string | No | - | Last frame image URL | | duration | integer | No | 5 | 3-15 seconds | | generate_audio | boolean | No | false | Auto-add audio to video |

Kling Video O3 — Reference-to-Video

Parameter	Type	Required	Default	Description
INLINECODE33	string	Yes	-	Video description
INLINECODE34

array | No | - | Reference images (up to 7 without video, up to 4 with video) | | video | string | No | - | Reference video URL | | keep_original_sound | boolean | No | true | Keep original sound from reference video | | sound | boolean | No | false | Generate new audio | | aspect_ratio | string | No | 16:9 | 16:9, 9:16, 1:1 | | duration | integer | No | 5 | 3-15 seconds |

Kling Video O3 — Video Editing

Parameter	Type	Required	Default	Description
INLINECODE40	string	Yes	-	Editing instruction in natural language
INLINECODE41

string | Yes | - | Source video URL (max 10s duration) | | images | array | No | - | Reference images for element, scene, or style (max 4) | | keep_original_sound | boolean | No | true | Keep original audio from the video |

---

Workflow: Submit → Poll → Download

Text-to-Video Example (V3.0 Pro)

CODEBLOCK4

Image-to-Video Example (V3.0 Pro)

CODEBLOCK5

Reference-to-Video Example (O3 Pro)

CODEBLOCK6

Video Editing Example (O3 Pro)

CODEBLOCK7

Standard Tier Example (Cost-Effective)

CODEBLOCK8

Polling Logic

• - processing / starting / running → wait 5s, retry (typically takes ~60-120s)
• INLINECODE47 / succeeded → done, get URL from INLINECODE49
• INLINECODE50 → error, read INLINECODE51

Atlas Cloud MCP Tools (if available)

If the Atlas Cloud MCP server is configured, use built-in tools:

CODEBLOCK9

---

Implementation Guide

1. 1. Determine task type:

- Text-to-video: user describes a scene/action in text - Image-to-video: user provides an image to animate - Reference-to-video: user wants to generate video using character/prop/scene references - Video editing: user wants to modify an existing video

1. 2. Choose model family:

- Kling V3.0 for standard text-to-video and image-to-video with negative prompts and cfg_scale control - Kling Video O3 for MVL-enhanced generation, reference-based video, video editing, and longer durations (3-15s)

1. 3. Choose tier:

- Pro for final output, client-facing content, or quality-critical use - Standard for most production use, cost-effective generation

1. 4. Extract parameters:

- Prompt: describe scene, action, camera movement, and visual details - Negative prompt (V3 only): specify undesired elements (e.g., "blurry, distorted faces, watermark") - Aspect ratio: infer from context (social reel→9:16, YouTube→16:9, square→1:1) - Duration: V3 supports 5 or 10s; O3 supports 3-15s - cfg_scale (V3 only): 0.5 default; increase toward 1.0 for stricter prompt adherence - Sound: enable if user wants audio; disabled by default

1. 5. Execute: POST to generateVideo API → poll result → download MP4

1. 6. Present result: show file path, offer to play

Prompt Tips

Kling produces best results with detailed, descriptive prompts:

• - Scene + Action: "A chef flips a pancake in a busy kitchen, steam rising from the pan"
• Camera direction: "Camera slowly pans left to reveal...", "Close-up tracking shot of...", "Aerial view sweeping over..."
• Style: "cinematic", "documentary style", "slow motion", "timelapse", "anime style"
• Negative prompts (V3): Use to avoid common issues — "blurry, low quality, distorted, watermark, text overlay"
• cfg_scale tuning (V3): Lower values (0.3-0.5) give more creative freedom; higher values (0.7-1.0) follow the prompt more strictly
• Reference-to-video (O3): Provide clear character/prop reference images for consistent results

---

Image Requirements for Image-to-Video

When using image-to-video models, the source image must meet these requirements:

• - Format: JPG, JPEG, or PNG
• Size: Maximum 10MB
• Dimensions: Minimum 300px on shortest side
• Aspect ratio: Between 1:2.5 and 2.5:1