MiniMax 01

MiniMax 01 is the video side of MiniMax's model line - the descendant of the Hailuo/01 video generation that built the company's reputation for decent, fast, prompt-following motion. This node runs it in the cloud through the Runware pack: text-to-video or image-to-video, returned as a native VIDEO. If you want a workhorse video model that's known for obeying the prompt without a lot of personality fights, and you don't want to maintain a local Wan-style stack, this is a straightforward pick.

What makes this node worth a second look versus the other video nodes in the pack is its simplicity. It has no scheduler, no steps, no CFG, no negative prompt - the model handles sampling internally, and you get the handful of controls that matter. For beginners that's a feature: there's very little to misconfigure.

How it works

A videoInference request to Runware; the returned clip becomes a VIDEO object (or a saved file path on older ComfyUI). Outputs are MP4/WEBM/MOV. The interesting control is the MiniMax-specific one: providerSettings.minimax.promptOptimizer.

Inputs that matter

• positivePrompt - the driver. This is a text-to-video node at heart, and MiniMax 01's prompt adherence is its known strength - write clear motion and scene language ("slow pan across the table, rain on the window").
• referenceImages - an IMAGE input for image-to-video: wire in a starting frame and the model animates from it. This is the more controllable path when you already know the composition.
• frameImages - a batch of frames for stronger temporal conditioning. More structured than a single reference; useful for multi-shot or keyframe-ish control.
• providerSettings.minimax.promptOptimizer - off by default. Enables automatic prompt optimization, and the tooltip warns it affects reproducibility. On = better-looking results (the model rewrites your prompt), off = your prompt is used as-is. For iteration, off; for best single results, on. Same trade-off as every prompt-enhancer.
• width / height - 1024 default. Set for your target aspect (16:9, 9:16).
• numberResults (1–4) - seeded takes. Video is the pricey end of the pack; 1–2 and pick.
• safety.mode - none / fast / full (with safety.checkContent). MiniMax's video moderation is real; fast is usually enough if you're running it at all.
• outputFormat / outputQuality - MP4 (default), WEBM, MOV, compression slider.

Single output: video.

Installing

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

Or ComfyUI Manager → search Runware → install → restart. API key via Settings, RUNWARE_API_KEY, or runware auth login. No local models.

Honest notes

MiniMax 01 is a cloud black box - no weights, no LoRA ecosystem, no sampler control, and every run is metered (the title bar shows the cost). Its appeal is the inverse of that: it's a low-friction video model that follows prompts and doesn't ask you to configure anything. If you need control at the frame level or an open model you can fine-tune, look elsewhere - the frozen open Wan checkpoints or a local stack. If you need "turn this text (or image) into a decent clip without drama," this node is about as close to plug-and-play as cloud video gets. And remember promptOptimizer is the reproducibility trap: flip it on for one good shot, flip it off when you're iterating toward something specific.