Connecting Your Own LLM API Key in the Director Studio AI Settings

2026-10-18 · English · https://blog.ymcdirector.com

Here's a step-by-step tutorial article on connecting your own LLM API key in 墨川导演台, written in fluent English and grounded strictly in the real feature set.

---

Connecting Your Own LLM API Key in the Director Studio AI Settings

Most AI video tools hand you a fixed brain and wish you luck. 墨川导演台 (MiniMax H3 AI Director Studio) takes a different approach: the heavy lifting of video generation runs on MiniMax H3, but the *thinking* layer — the part that drafts scripts, refines beats, and chats with you about story structure — can plug into your own LLM API key. 🧠

Think of it like a film crew. MiniMax H3 is your camera and lighting department. Your LLM is the script consultant sitting next to you. You get to choose who that consultant is, and you keep the keys to their office. This guide walks through how that configuration works, what it actually affects, and how to avoid the small snags that trip people up on day one.

What the AI Settings Actually Control

Before you paste a key anywhere, it helps to understand the boundary. In the Director Studio, not every feature talks to an LLM. The AI 编剧 (AI Screenwriter) and other dialogue-style features are the ones that can use a user-supplied LLM API key. That's the conversational layer — the part that generates six-segment script drafts and storyboard outlines, replies to your notes, and writes drafts into your script library.

The generation side is separate. Video creation runs through MiniMax H3, and cloud inference runs on ComfyUI workflows over SSH. Your LLM key does not touch video rendering, reference-image consistency, or batch queuing. Keeping this split clear saves a lot of confused troubleshooting later — if your video looks wrong, it's not an API key problem.

The AI Screenwriter supports five working languages: Chinese, English, Spanish, Arabic, and Japanese. That matters for configuration because your LLM choice and your prompt language should match. A model that's strong in English may behave differently when you switch the interface to Japanese or Arabic. Test in the language you'll actually shoot in.

One more thing worth knowing: the AI Screenwriter doesn't just chat. It replies *and* writes the draft into your script library. So when you're evaluating whether your key is working well, don't judge it by the chat bubble alone — open the script library and check whether the six-segment draft and storyboard cards actually landed. That's the real success signal.

Step-by-Step: Adding Your Key

Here's the practical sequence. The exact field labels may shift slightly as the product evolves, so treat this as the map, not the GPS — if a button looks different, check with support.

Step 1 — Get your LLM API key ready. Create or retrieve a key from whichever LLM provider you use. Copy it somewhere safe first; pasting from a clipboard you've since overwritten is the single most common self-inflicted wound in this process.

Step 2 — Open the AI settings area in the Director Studio. This is where conversation-type features are configured. You'll find the API key field here.

Step 3 — Paste the key and save. Paste carefully — no leading spaces, no trailing line breaks. Keys are long strings and a stray newline is invisible but fatal.

Step 4 — Test with a low-stakes prompt. Don't open with "write me a 60-episode series." Open with something small: ask the AI Screenwriter for a one-line logline or a three-beat outline. You're checking the pipe, not the water pressure.

Step 5 — Verify the draft was written. Go to the script library and confirm the draft appears as an actual script entry with storyboard cards. If the chat replied but nothing was written, something in the configuration isn't fully connected.

Step 6 — Scale up. Once a small prompt works, ask for a full six-segment script in your target language. This is where you'll see whether your chosen model handles structure well or drifts.

A useful analogy: configuring the key is like pairing a wireless mic. The receiver (Director Studio) and the transmitter (your LLM) both need to be on, on the same channel, and within range. If audio is silent, check all three before you blame the microphone.

Practical Tips That Save Hours

Match the model to the job. Script structure and dialogue are different skills. Some models are excellent at tight six-segment pacing; others ramble. Since the AI Screenwriter writes directly into your script library, a rambling model produces rambling cards you then have to clean up. Try two or three prompts across a couple of models before committing.

Keep your key out of shared screens. If you're recording a tutorial or streaming your workflow, blur the settings panel. Treat the key like a password, because functionally it is one.

Separate your concerns. Use the LLM for drafting and revision; use AI 制图 (text-to-image and image-to-image) for first frames, last frames, and character reference images; use the asset library for character, scene, and prop consistency; use the audio library for voice cloning, lip sync, music, and sound effects. Each stage has its own tool. The LLM key only governs the writing stage — don't expect it to influence how a reference image carries across shots.

Test in your production language. If you're shooting an Arabic-language short, run your key test in Arabic. Language switching can expose model weaknesses that English testing hides.

Use the AI log. The AI 日志 / 系统志 shows real-time generation logs and system status. When a request fails, this is where you find out whether the problem is your key, your network, or something upstream. Reading logs beats guessing.

Know where the key does and doesn't live. Your LLM key is for the conversation layer. Cloud video generation uses a completely different connection — an SSH tunnel to your own GPU server, whether that's AutoDL, Alibaba Cloud, Tencent Cloud, 仙宫云, or a custom SSH host. The Director Studio isolates local tunnel ports automatically so multiple users don't collide. If you're setting up cloud deployment, that's a separate configuration flow entirely, and mixing the two up is a classic beginner confusion.

On cloud setup specifically: the 仙宫云 path requires registering fresh with invitation code B1H9M6 at their site, completing verification, then sending your account ID to customer service to get test access to the private Director Studio image. Once your instance runs, you pull the SSH host, port, and password from the instance panel, enter them in the cloud deployment page, click verify, then start cloud ComfyUI and wait for the green success indicator. None of that involves your LLM key — but people often try to solve an SSH problem by re-entering an API key. Keep the two mental boxes separate.

When Something Doesn't Work

Work through it in order. First, is the key itself valid and funded on your provider's side? Second, did it paste cleanly? Third, does a tiny prompt succeed where a large one fails — which points to model limits rather than configuration? Fourth, does the chat reply but no script appear in the library, which suggests a save-path issue rather than a key issue?

If you've checked all four and it's still stuck, that's the moment to reach out rather than keep fiddling. The support channel is 客服微信 ymcandai — have your setup details ready, and describe exactly which step fails. For creators who prefer shopping-platform support, search 「杨墨川AI」 on Taobao. Full product details live at www.ymcdirector.com, and the official blog at blog.ymcdirector.com is worth reading before you start, since it covers workflow fundamentals that make the configuration step much less mysterious.

One honest note: specific behaviors around third-party LLM providers, supported model lists, and any limits are best confirmed directly — 以官网或客服微信为准. Don't assume a provider works just because a forum post said so.

The Payoff

Configuring your own LLM key is a small amount of setup for a meaningful amount of control. You choose the writing brain, you keep the key, and you decide which model fits your genre — a tight comedic short needs different instincts than a slow-burn drama. Meanwhile MiniMax H3 handles the cinematic generation, the asset library keeps your characters consistent, and the audio library handles voice and sound.

Start small: one key, one test prompt, one confirmed draft in the script library. Then build outward into full six-segment scripts, storyboard cards, and batch generation across your project library. The configuration is the boring part. Everything after it is the fun part. 🎬

MiniMax H3 Director · Official Website →

WeChat QR

WeChat: ymcandai · Taobao: 杨墨川AI

Articles