Claude Fable 5.2 + NadouPro CLI Tutorial: AI 3D Modeling and Cinematic Video Generation
The hard part of AI video is no longer “can it be generated,” but “can it be controlled.” A character’s face changes from another angle, a scene breaks continuity when the camera position changes, and the visual tone is inconsistent from shot to shot. These problems cannot be solved by repeatedly drawing random outputs. NadouPro is iQIYI’s professional AI filmmaking platform. It brings the control methods of the film industry into AI creation: 3D World, 3D Director, camera parameters, performance driving, and film-grade dubbing. NadouPro CLI brings these creative capabilities into your terminal. When you let Claude Fable 5.2 drive the CLI, you only need to describe your goal in natural language. This article follows the sequence “installation → 3D modeling → cinematic video → refinement and final cut” and provides a workflow you can follow directly.
Step 2: Verify the version and capabilities
How NadouPro CLI Works with Claude
NadouPro CLI is a command-line tool that connects your local terminal with the NadouPro web creation platform. Its official positioning is “a transfer station that brings AI into the creative workflow.” The official documentation explicitly supports AI tools such as Claude Code and Cursor. In Claude Code, choose the Claude Fable 5.2 model, and you can drive all commands through conversation. During installation, an official Nadou Skill is synced, so Claude knows how to use each command. A typical collaboration has three steps: 1. You describe the goal in natural language, for example, “Use this reference image to make a 5-second video, and first tell me the model, parameters, and estimated cost.” 2. Claude confirms the available capabilities and models, uses nadou task estimate to estimate the cost, and submits after your confirmation. 3. Claude calls nadou image, nadou video, or Canvas commands to generate content, and the results are automatically archived to a personal or team space.Get Started in Three Steps: Install, Verify, and Log In
Installation requires Node.js 16 or later, and supports Windows x64, Linux x64/arm64, and macOS x64/arm64. Step 1: Ask Claude to install with one command. Switch to Claude Fable 5.2 in Claude Code, then send:Please help me install NadouPro CLI: npm install -g @iqiyi-intl-nadoupro/nadou-cli.
After installation, run nadou version --json to verify it, and run nadou skill install --json to sync the official Skill.
Step 2: Verify the version and capabilities
nadou version --jsonIf the version information returns normally and the Skill version matches the CLI version, the installation is successful. The output of capability list is the only basis for determining “what the current account can do.” Step 3: First authorization login
nadou skill read --json
nadou capability list --json
nadou auth login --timeout 10m --jsonThe CLI will open the default browser, or provide an authorization URL in the output. After login is complete, Claude will continue the original task. For full instructions, see the official NadouPro documentation.
Hands-On 1: AI 3D Modeling, from Three-View Images to 3D World
NadouPro’s 3D modeling has two layers: the CLI batch-produces the character and scene assets needed for modeling, while 3D World and 3D Director on the web turn those assets into schedulable space. Build a stable space first, then discuss shots. 1. Use the CLI to generate character three-view images. Three-view images (front, side, and back) are the foundational assets for character modeling and maintaining character consistency. Say to Claude:Use the three-view scenario in NadouPro CLI to generate character three-view images for the protagonist:Claude will first use nadou model scenarios to query which models support this scenario, then estimate the cost, submit, and archive. 2. Use the CLI to generate a scene panorama
a 30-year-old female bounty hunter, short hair, dark gray tactical trench coat, realistic cinematic texture.
First list the models that support three_view and the estimated cost. Submit only after I confirm,
and archive the result to the “Characters” directory in my personal space.
A panorama provides surrounding scene information and is the starting point for building scene space:3. Generate 3D World in NadouPro. Open the NadouPro web app and use the scene asset from the previous step to generate 3D World. Compared with generating images from different angles one by one, 3D World builds a complete scene space in one pass. When switching to any angle, the spatial structure of streets, buildings, and props remains stable. This is the prerequisite for multi-camera shooting without continuity errors. 4. Use 3D Director to complete complex shot scheduling in advance. Rehearse in 3D Director before generating: use the multi-camera timeline to control character blocking, camera positions, and camera movement. 3D Director is currently a web workflow and is not yet open in the CLI. You can ask Claude to first write a scheduling sheet with camera positions, blocking, and duration, then execute it in 3D Director according to the sheet.
Use the panorama scenario to generate a rainy-night neon street: wet reflective pavement, an elevated road in the distance,
dense signboards on both sides, in a cyberpunk art style. After generation, download it to ./assets/scene and perform a technical check.
Hands-On 2: Cinematic AI Video Generation
A cinematic look comes from control: controlled shots, controlled light, and controlled visual tone. Below, a 30-second rainy-night chase short film is used as an example and completed in four steps. 1. Ask Claude to break down the shots and build the canvasBreak this plot into a 6-shot storyboard table (shot size, camera position, camera movement, duration, dialogue),Claude will use nadou canvas create, canvas node create, and canvas edge connect to automatically build the production pipeline. “Read first, then modify, then read again” is the official recommended way to operate the canvas. 2. Write camera parameters into the image. NadouPro supports professional parameters such as camera model and Color Science directly participating in image generation. When generating keyframes, describe the shot in a cinematographer’s language instead of piling up adjectives:
then use NadouPro CLI to create a new canvas named “Rainy Night Chase.” Create one Image node and one Video node for each shot and connect them,
with the character consistently referencing the protagonist three-view images in my personal space. Read the canvas before writing, and read it again after writing to confirm.
Shot 3: Medium Shot, the protagonist turns back under a neon sign. Cinematic camera texture, 35mm Focal Length, shallow Depth of Field,3. Generate video: First/Last Frame, Multi-reference, and Motion Control
side Rim Light, teal-and-orange tone, visible rain streaks. Please fill in the camera model and Color Science according to the available platform camera parameters.
Estimate the cost for the Video nodes of shots 1–6 one by one, and list the model, parameters, and total cost. After I confirm, generate them in batches,4. Unify the visual tone and export the final cut. When lighting is inconsistent between shots, readjust the light in NadouPro and use a swatch to keep the visual tone consistent across the full film. Finally, ask Claude to use canvas media timeline to write to the timeline, and use canvas media export to submit the final cut export.
wait for the results using the original task IDs, and do not submit repeatedly. After completion, download them in batches to ./output.
Estimating before submission (nadou task estimate) can avoid unexpected cost. According to the official manual, failed tasks are fully refunded.
Refinement and Post-Production: From “Watchable” to “Broadcast-Ready”
The final 20% of finished-film quality is determined by refinement. The following steps are recommended in order in NadouPro: 1. Skin texture enhancement: restore texture and details lost in direct AI output for Close-Up and near shots. 2. Facial refinement: make targeted edits to local features such as eye shape, nose bridge, and facial contour, giving the character recognizability and avoiding faces becoming too similar. 3. Performance driving: record a live-action performance and use a performance-driving algorithm to drive expressions and actions onto the generated character, solving performance details that are hard to describe clearly in text. 4. Voice Cloning and dubbing: use Voice Cloning to replicate an existing voice, then use iQIYI’s self-developed dubbing model to perform dialogue, achieving film-grade dubbing quality. 5. Intelligent Ultra HD and Frame Interpolation: rely on self-developed image algorithms to improve resolution and frame rate, enhancing image detail and motion smoothness. These refinement capabilities are mainly used on the web. On the CLI side, you can ask Claude to do supporting work: use nadou media to complete frame extraction checks, stitching, subtitle processing, and audio-video muxing, and use nadou space to archive each version of the final cut to the Team Space.Contains AI-generated information that may be incorrect. Please verify independently.