CLI User Guide

What is NadouPro CLI?

✈️ The bridge that brings AI into your creative workflow

From a single idea and a few reference images to storyboards and video nodes, you only need to describe the goal in natural language. The AI assistant calls the available NadouPro creative capabilities through NadouPro CLI and confirms with you at key steps. NadouPro provides Canvas, Workbench, Agent, and other creation entry points; the actual CLI coverage depends on the current version.

Think of a film production as a set collaboration: you provide creative intent and make the key calls, the AI assistant breaks down tasks, invokes tools, and follows up on results, and NadouPro provides models, assets, spaces, and Canvas. NadouPro CLI is the working channel among the three.

Its value is not teaching users commands. It is enabling the AI assistant to actually execute: choose a currently available model, read reference assets, submit image or video generation, wait for results, and deliver them to a personal space, team space, or a specified Canvas.

Who is it for?

• Film, video, and content creators: Want an AI assistant to handle character design, storyboards, shot generation, and asset organization.

• Creative teams: Need results placed in a team space or Canvas for review, collaboration, and iteration.

• Users of AI tools such as Claude Code and Cursor: Want to call NadouPro directly in conversation, instead of switching pages and copying assets back and forth.

1. 📦 Quick install

Option 1: One-click install with an AI assistant

Send the following to your AI assistant:

Please install NadouPro CLI: npm install -g @iqiyi-intl-nadoupro/nadou-cli. After installation, run nadou version --json to verify (requires Node.js 16 or later)

The installer does three things:

1. Installs the Nadou CLI build that matches the current operating system and CPU architecture.

2. Verifies that the CLI starts correctly and prints the version, commit, and build time.

3. Syncs the matching official Nadou Skill to the global agents supported on this machine.

Installation success criteria

nadou version --json returns version information, and the Skill version in nadou skill read --json matches the CLI version.

Option 2: Manual install

Manual install steps

If you prefer to operate in a terminal yourself, run:

npm exec --yes --registry=https://registry.npmjs.org/ --package=@iqiyi-intl-nadoupro/nadou-cli@latest -- nadou-install
nadou version --json
nadou skill install --json
nadou skill read --json

If you are sure this machine has never had an older Nadou CLI installed, you can also install the npm package directly:

npm install -g @iqiyi-intl-nadoupro/nadou-cli
nadou version --json
nadou skill install --json
nadou skill status --json

The current installer supports Windows x64, Linux x64/arm64, and macOS x64/arm64. There is no prebuilt package for Windows arm64 yet.

First-time authorization and sign-in

Before running account-required operations, the Agent should check sign-in status first to avoid opening the browser repeatedly:

nadou auth status --no-interactive --json

If you are not signed in, then run:

nadou auth login --timeout 10m --json

1. The CLI tries to open the system default browser. If the current environment cannot open a browser automatically, the same authorization URL is printed in the output.

2. On the international sign-in page, click “Continue to international sign in”, then complete international-site account sign-in as prompted.

3. After authorization succeeds, the CLI verifies sign-in status automatically. The Agent can continue the original task; you do not need to run another login command.

This guide uses the international public CLI, which connects to the international production environment by default. This distribution supports the production environment only.

Sign in on a server or in a container

If the CLI runs on a server, in a Docker/Claw container, or in an environment without a graphical interface where the system browser or system credential store is unavailable, run:

nadou auth login --server

The CLI displays a sign-in URL and verification code, then waits for authorization. Open the URL in a browser on any device, enter the code, sign in, and confirm the account to use with the CLI. To use a different account, select “Switch account”.

After sign-in succeeds, credentials are saved in a private file in the current CLI state directory. Subsequent CLI commands in the same container or using the same state directory can reuse them directly. If the container is destroyed without persisting this directory, you will need to sign in again.

The default nadou auth login continues to use the existing browser authorization flow and system credential store.

2. 📹 Why it fits film and video production

NadouPro CLI currently focuses on text, image, video, spaces, and Canvas. It can complete a single generation, or organize a multi-asset, multi-stage production pipeline around the same project.

Capability	Typical work
Text	Idea exploration, script outlines, storyboard descriptions, prompts, and copy; text results are returned directly.
Image	Text-to-image, image-to-image with references, inpainting, panoramas, and subject three-views.
Video	Text-to-video, multi-reference video, first/last-frame generation, semantic video editing, and motion control.
Canvas	Create or read a canvas, organize text, image, and video nodes and edges, and generate on specified nodes.
Spaces and delivery	Save generated results to a personal space or a specified team space for continued collaboration and management.
Local video processing	Inspect video information, burn in subtitles, and concatenate compatible clips without re-encoding.

Creative advantages

1. Expand from a single asset to a complete production pipeline

2. Ordinary generation tools are better at completing one request. Film and video production needs to pass information back and forth among ideas, characters, scenes, storyboards, and shots. NadouPro CLI lets the AI assistant keep working toward the same creative goal and put results from each stage back in a clear location.

3. Reference assets have clearer roles

4. Character, costume, scene, style, first-frame, last-frame, and motion references each serve a different purpose. As long as you describe how each asset should be used in natural language, the AI assistant can choose the matching generation method and reduce mistakes such as treating an ordinary reference image as a first frame.

5. The process can be confirmed in stages

6. Image and video generation consume points. The AI assistant can show the model, specs, estimated points, and save location before submitting, so you can judge the direction first and then decide whether to generate. For staged results such as storyboards and key visuals, you can also review first and continue later.

7. Canvas preserves upstream and downstream relationships

8. When a task involves multiple storyboards, multiple shots, or several rounds of revision, Canvas can express dependencies with nodes and edges. The AI assistant works in a specified canvas and node, and results return to the corresponding node, making further adjustment and collaboration easier.

3. 🫧 Get started

Tell the Agent in natural language

After install and sign-in, you do not need to memorize CLI parameters. You can simply say:

• “Generate a cinematic poster of a neon-lit rainy night street.”

• “Use this reference image to generate a 5-second video. Tell me the model, parameters, and estimated cost first.”

• “Upload this local image and place it on the canvas I specify.”

• “Read the current structure of this canvas, add a text node on the right, and connect it to the image node.”

• “The previous video is not finished yet. Keep waiting on the original task ID. Do not resubmit.”

The Agent should first confirm information that is ambiguous and must not be guessed, such as account, space, canvas, target node, first/last-frame roles, or a high-cost batch range; then call NadouPro CLI.

4. ✏️ Typical production scenarios

Scenario 1: From one idea to shot drafts

Early production often starts with a single concept. Ask the AI assistant to organize characters, scenes, and mood first, then generate character sheets, atmosphere frames, or storyboard drafts. After the visual direction is confirmed, move on to video.

You can say:

I want to make a cinematic Latin dance short set in a rainy night. First, I'll organize a brief creative overview, character, and scene settings, and then generate 4 storyboard drafts in 16:9. The overall vibe is high-contrast neon lighting, with character outfits matching the rainy night street. Let me know the model, quantity, and estimated points before generating.

Scenario 2: Keep character and art consistent with reference images

When a production needs a stable character look and visual style, upload multiple reference images and specify the role of each image. The AI assistant chooses available models based on character, costume, scene, and other reference duties. Generated results are not placed on Canvas automatically.

You can say:

Please use the 3 reference images I uploaded with this message to independently generate 3 consecutive cinematic storyboards: image 1 should reference only the heroine’s facial features and hairstyle, image 2 should reference only the cut and fabric of the red fringe dance dress, and image 3 should reference only the neon lighting and color of the rainy night street. All three storyboards should be 16:9 and keep the same character, costume, and scene. Do not add new characters. Before generating, tell me the model, how the references are assigned, the generation count, and the estimated points. Save the results to my personal space. Do not create a Canvas.

Scenario 3: Turn a storyboard into a video shot

When you already have a clear storyboard, specify it as the first frame of the video and add character action, camera movement, duration, and frame specs. The AI assistant chooses the matching video generation method based on the first-frame role, instead of treating the storyboard as an ordinary style reference.

You can say:

Please use the storyboard I uploaded with this message as the first frame, and independently generate a 5-second, 16:9, 720p video. The two dancers in the frame complete one Latin dance spin, and the camera slowly pushes from a wide shot to a medium shot. Keep the characters’ looks, costumes, rainy night street, and neon tones from the first frame. Do not add new people. Before generating, tell me the model, video parameters, and estimated points. Save the results to my personal space. Do not create a Canvas.

Scenario 4: Organize a short film on Canvas

When a production includes a brief, character design, scenes, storyboards, and multiple video shots, use Canvas to organize assets and their upstream/downstream relationships. The AI assistant confirms the save location and canvas structure first, then creates nodes and runs generation.

You can say:

I want to plan a “coastal highway chase” short on Canvas. First read my personal space directories and let me confirm where the canvas will be saved. After confirmation, create a Canvas with the same name, and plan a creative brief, male and female lead designs, a coastal highway scene, 4 storyboard image nodes, and matching video nodes. Also connect the upstream and downstream relationships. Show only the canvas structure and node generation plan first. Do not generate immediately. After I confirm, submit one node at a time, and tell me the model, parameters, and estimated points before each submission.

Scenario 5: Deliver results to the right place

For team projects, specify the team space and target directory at the start. The AI assistant verifies the real space and directory first, then generates and delivers results, so assets are not saved in the wrong place.

You can say:

Please independently generate 3 “autumn outdoor brand ad” storyboards. The frames should show a golden grassland, an outdoor model, and warm cinematic lighting, in 16:9. Deliver the results to the “2026 Autumn Campaign” directory in the “Brand Visuals” team space. Do not create a Canvas. If there are multiple spaces or directories with the same name, list the candidates and let me choose. Do not switch on your own. After confirming the target location, tell me the model, generation count, and estimated points. Submit generation only after I confirm.

5. 💗 Best practices

• State the creative goal first. You do not need to look up commands first. Describe the image, shot, or project you want. The AI assistant discovers currently available capabilities from the current version.

• Give each reference asset a clear job When uploading multiple assets, specify whether each is a character, costume, scene, composition, style, first-frame, last-frame, or motion reference. Do not just say “use these images as references.”

• Specify output specs and save location together Image count and aspect ratio, video duration and resolution, whether audio is needed, and whether the destination is a personal space, team space, or Canvas should all be confirmed before submit.

• Review points and the action plan before generating Before image and video generation, ask the AI assistant to show the model, key parameters, estimated points, current balance, save location, and whether local assets need to be uploaded or results downloaded. An estimate does not lock the price. Final deduction follows the submitted result.

• Keep waiting on long tasks. Do not resubmit. Video generation or node backfill can take time. A wait timeout or a page that has not refreshed does not mean failure. Keep checking the original task. Resubmitting may consume points again and produce duplicate results.

• Work on Canvas with read–modify–reread When changing nodes or edges on an existing canvas, the AI assistant should read the current structure first, complete one confirmed modification, then reread to verify. This reduces overwriting others’ work or generating on the wrong node.

6. 🙋 Important notes and FAQ

What the current CLI does not expose

Independent audio generation, complex professional editing, cross-project copy, and unconfirmed platform timeline render/export are not currently documented as directly usable CLI capabilities. Even if the web product has matching entry points, treat the capabilities currently returned by the CLI as the source of truth.

The browser is blank or a dark page after sign-in

Go back to the original AI assistant conversation and check sign-in status. If you are already signed in, you can close the page and continue. If you are not signed in, check whether the international sign-in page is still waiting for “Continue to international sign in”, or whether it shows an initialization error. Keep the original login command running and complete sign-in as prompted. Do not treat page color as proof of successful authorization, and do not start the login command again.

The canvas already has nodes, but there is no generated result

This usually means only the canvas structure was created. Ask the AI assistant to check each node’s prompt, model, and required parameters, show estimated points, get confirmation, and then submit node generation.

Permission denied, or the target directory cannot be found

Confirm the current account first, then verify whether the target is a personal space or a specific team space, and whether the account has write access to the target directory. Do not let the AI assistant guess directories by name or switch team spaces without confirmation.