Batch Video Generation Made Easy: Project Queues and Single-Shot Workflows
If you've ever tried to produce an AI short drama, you already know the pain: one shot looks great, the next shot's character has a different face, and by the time you've generated twenty clips you've lost track of which version belongs to which scene. The bottleneck in AI video isn't usually the model — it's the workflow.
That's the problem 墨川导演台 (MiniMax H3 AI 导演台) is built to solve. Instead of treating every clip as a one-off experiment, it organizes your work around projects, queues, and single-shot control. Think of it less like a slot machine and more like a proper editing desk — you still get the magic of AI generation, but you also get a place to put everything.
In this tutorial, we'll walk through how to combine single-shot generation for precision with batch generation for volume, all managed inside a project queue. By the end you'll have a repeatable rhythm you can apply to any short film, ad, or episodic series.
---
H2: Start With One Shot — The "Rehearsal" Mindset
Before you ever hit batch generate, generate a single shot. This is the equivalent of a film director doing a rehearsal before rolling the camera on the whole scene. It costs you almost nothing in time, and it saves you from queuing up fifty clips that all share the same mistake.
Here's the practical flow inside the director's desk:
- Write or import your script. Use the AI 编剧 (AI screenwriter) to generate a six-段 (six-part) script and storyboard draft through conversation. It supports Chinese, English, Spanish, Arabic, and Japanese, and — importantly — the AI actually writes the draft into your 剧本库 (script library), it doesn't just chat at you. You can also import an existing script or start a new one manually.
- Break it into storyboard cards. Inside the script library you manage scripts, episodes, and individual storyboard cards. Each card is a shot. This is your shot list.
- Generate one frame first. Use AI 制图 (AI image generation) — text-to-image or image-to-image — to create the first frame, last frame, and character reference images for that shot. This is where you lock in the look.
- Attach your references. In the 资产库 (asset library) you manage characters, scenes, props, and reference videos. Upload reference images here so your character and scene stay consistent across shots. This is the single most important step for avoiding the "why did her face change?" problem.
- Generate the single shot. Now run that one shot through the project. Watch the AI 日志 (generation log) to see what's happening in real time — if something's off, you fix it here, at a cost of one clip instead of thirty.
Only once a single shot looks right should you scale it up. This is the core discipline: rehearse one, then perform many.
---
H2: Build the Project Queue — Where Batch Generation Actually Lives
Here's where the workflow gets satisfying. The 项目库 (project library) is where you organize generation tasks and versions by project. It's the container that holds everything: your single shots, your batches, and every version you've ever made.
Think of the project queue like a restaurant kitchen during dinner service. The single-shot workflow is you, the chef, carefully plating one dish to check the seasoning. The batch queue is the line cook firing twelve orders at once — same recipe, same standards, just at volume.
How to set up a batch run:
- Organize by project. Create a project for your short film, episode, or campaign. Every generated task and version lives inside it, so you never lose track of which clip belongs where.
- Confirm your assets first. Before batching, double-check that every character, scene, and prop referenced in your storyboard cards exists in the asset library. Consistency is inherited from references, not invented per shot.
- Queue the shots. Batch generation lets you fire off multiple video generations from your storyboard cards at once. The system handles the queueing so you're not babysitting each clip.
- Let the queue work. Because generation runs through a ComfyUI workflow on the inference backend, the queue supports reference-image consistency, storyboard continuity (分镜衔接), and batch queueing. In plain terms: the queue doesn't just generate clips, it tries to keep them coherent with each other.
- Review versions, not chaos. Every run is stored as a version inside the project. If shot 7 came out wrong, you regenerate shot 7 — you don't redo the whole scene.
The mental shift is this: single-shot generation is for decisions, batch generation is for execution. Mixing them up is how creators burn hours.
---
H2: Choose Your Engine — Local, Cloud, or API
Batch generation is only as good as the machine running it. The director's desk supports three real modes, and picking the right one changes how smoothly your queue flows.
Local deployment runs on your own GPU computer, with inference happening locally. Great when you want full control and don't want to depend on a connection. Your queue lives on your hardware.
Cloud deployment connects to a GPU cloud server you rent yourself, with inference running over an SSH tunnel. The desk supports general cloud GPU deployment across AutoDL, Alibaba Cloud, Tencent Cloud, 仙宫云 (Xiangongyun), and custom SSH servers. It connects to a cloud ComfyUI over SSH, and — this detail matters for teams — local tunnel ports are automatically isolated, so multiple users don't step on each other's queues.
API calls let you access the director's desk capabilities programmatically. Useful if you're building batch generation into a larger pipeline.
If you want to test the cloud path, here's the real flow for 仙宫云:
- Connect to the 仙宫云 private image (it's not public).
- Register at 仙宫云 through the official link and use invitation code B1H9M6 — you must register fresh with this code: https://www.xiangongyun.com/register/B1H9M6
- After registering and completing verification, send your 仙宫云 account ID to customer service (for example:
ac401472-db02-5df3-xxxx-xxxxxxxxxxx). - Once your test account is opened, you'll see the director's desk image — deploy an instance with it.
- After the instance runs, click SSH on the right side of the instance to get the connection host, port, and password. Enter these into the 云部署 (cloud deployment) page's SSH Host / SSH Port / SSH password fields.
- Click 验证 SSH (Verify SSH). Once it passes, click 启动云端 ComfyUI (Start Cloud ComfyUI) and wait for the green success indicator.
- Now you can write a script and test video generation.
For anything pricing- or spec-related, check the official site or ask customer service — those details change and are best confirmed directly.
---
H2: The Audio Layer — Don't Forget It When Batching
A batch queue that only produces silent video is half a film. The 音频库 (audio library) handles voice cloning, lip sync, background music, and sound effects, and AI 配音 (AI dubbing) uses IndexTTS 2.5 for multilingual dubbing with emotion control. There's also AI MV for music video generation.
Practical tip: lock your voice and audio choices before you batch your video shots. If your character's voice changes between episodes, the consistency work you did in the asset library gets undone at the last step. Treat voice as part of your asset setup, not an afterthought.
---
H2: Putting It All Together
Here's the rhythm that works:
- Rehearse — generate one single shot, check the frames, references, and audio.
- Lock — confirm your characters, scenes, and props are in the asset library.
- Queue — batch generate the rest of the storyboard cards inside the project.
- Review versions — fix individual shots, not whole scenes.
- Repeat — the project library remembers everything, so your next episode starts from a proven setup.
The whole point of a project queue is that it turns AI video from a series of lucky accidents into an actual production line. Single-shot generation gives you the control of a director; batch generation gives you the throughput of a studio. Use them together and you stop fighting your tools.
If you want to try it, start at the official site, and if you get stuck, the fastest path is usually the customer service WeChat: ymcandai. You can also find the desk on Taobao by searching 「杨墨川AI」, and on Douyin, Xiaohongshu, and Video Account by searching 「杨墨川」. For anything else, email info@ymcdirector.com.
