MiniMax 01 Director

"Director" in a video model name usually means one thing: it wants to be told what to do, scene by scene. MiniMax 01 Director is that node - the MiniMax video tier pitched at multi-shot, direction-heavy generation, where you're less "give me a video of a cat" and more "shot 1: the cat wakes up. Shot 2: it stalks across the kitchen. Shot 3: it dramatically ignores the camera." It's the tool you reach for when a single motion prompt isn't enough and you're essentially storyboarding in text.

Like every Hailuo-family node in the Runware pack, it's closed, cloud, and metered: MiniMax doesn't ship these weights, so the API node is the only way to drive it from ComfyUI, and each run costs money. The "Director" framing means the model is built to hold a longer narrative through the generation - that's where it beats the plain prompt nodes, which can lose the plot if you try to cram a sequence into one paragraph.

What you set

• positivePrompt (required) - this is where your direction lives. Structure it like a shot list; the model responds to explicit scene transitions.
• frameImages - image-to-video input for character or scene continuity across shots.
• width / height - 1024 default.
• numberResults - 1-4.
• providerSettings.minimax.promptOptimizer - automatic prompt rewriting; the tooltip warns it affects reproducibility, so off if you need stable seeds.
• safety.mode - none/fast/full.
• outputFormat MP4/WEBM/MOV, outputQuality 20-99.

Note there's no duration slider here - this tier uses the model's own pacing for the sequence, which is part of why it's called Director.

Output is video (VIDEO).

How it works

Same pack machinery as every node: model AIR + taskType over REST via the Runware SDK, media converted back to a native VIDEO, cost printed on the title bar. The difference is entirely in what the model is tuned to do - hold a directed sequence together rather than produce a single motion clip. In practice that means your prompt does the heavy lifting: explicit scene structure gets you coherent multi-shot output; a vague prompt gets you a vague video at Director pricing. The character-consistency catch is worth remembering, too - for a character who must survive multiple shots, feeding a frameImages reference helps far more than prose alone.

Installing

ComfyUI Manager → search Runware → install → restart. Manual:

cd ComfyUI/custom_nodes
git clone https://github.com/Runware/ComfyUI-Runware
pip install -r ComfyUI-Runware/requirements.txt

No model downloads; dependencies are runware-sdk, pillow, soundfile. API key via Settings → Runware API key, RUNWARE_API_KEY, or runware auth login.

Troubleshooting

Set the key first - it's the pack's #1 error. If your "directed" video comes out incoherent, re-read your prompt as a shot list: the model needs explicit scene breaks, not one run-on sentence. And if you're not actually doing multi-shot work, this node is overkill - the plain Hailuo nodes cost less for a single clip.