Can I use MiniMax H3 with my MiniMax API key?

Not yet. MiniMax’s own video generation API still lists only MiniMax-Hailuo-2.3, MiniMax-Hailuo-2.3-Fast, MiniMax-Hailuo-02 and the I2V-01 family as supported model values. H3 is reachable today through fal.ai, which was an official launch partner, so a fal key is the working route.

How much does MiniMax H3 cost per video?

On fal, H3 costs $0.26 per second of output at 2K. A 6-second clip is about $1.56 and a 15-second clip about $3.90. On reference-to-video the first 5 reference images are free, additional images cost $0.08 each, and audio references are free. Billing runs on your own fal key.

What are the MiniMax H3 model ids?

The fal endpoint paths are minimax/h3/text-to-video, minimax/h3/image-to-video and minimax/h3/reference-to-video. In OpenClaw config they carry a fal/ prefix, for example fal/minimax/h3/text-to-video. The path segment is h3, not hailuo-03.

What is the difference between MiniMax H3 and Hailuo 2.3?

H3 outputs up to 2K at 24 FPS with native 32 kHz stereo audio generated in the same pass, and accepts text, images, video clips and audio references together. Hailuo 2.3 is silent, lower resolution, and takes only text or a start image. Hailuo 2.3 is cheaper for short silent clips.

← Home

Guide

MiniMax H3 on OpenClaw: Give Your Agent 2K Video with Sound

MiniMax H3 (Hailuo 3.0) is selectable as your agent's video model today through fal.ai — 2K output at 24 FPS with native stereo audio, from a single request that can mix text, images, clips, and voice references. Here is how to point your bot at it.

Your Agent Has a Video Model Slot

Every OpenClaw and Hermes instance on OpenClaw Launch has three separate model slots on its dashboard card: Model for chat, Image for image generation, and Video for video. The Video dropdown is grouped by provider — Agnes (free), Google, fal.ai, MiniMax, xAI — and each provider group unlocks when you add that provider's key.

That is the slot MiniMax H3 lands in. Set it once and your agent can produce a clip in reply to a chat message on Telegram, Discord, WhatsApp, WeChat, or the web, without you touching an API yourself. For the full brief-to-MP4 workflow, see making videos with your bot.

Use fal.ai, Not Your MiniMax Key (For Now)

This is the part that trips people up. MiniMax H3 shipped as open weights and as a hosted API, but MiniMax's own video endpoint has not exposed it yet: the supported model values on the MiniMax video API are still MiniMax-Hailuo-2.3, MiniMax-Hailuo-2.3-Fast, MiniMax-Hailuo-02, and the older I2V-01 family. So the MiniMax provider group in your Video dropdown tops out at Hailuo 2.3.

fal.ai was an official launch partner and carries H3 on day one, which makes it the working route today. Add a fal key and H3 appears in the fal.ai group of the same dropdown.

The Three H3 Endpoints and What They Cost

Endpoint	Takes	Price
`minimax/h3/text-to-video`	A prompt	$0.26 per second at 2K
`minimax/h3/image-to-video`	A prompt plus a starting image (image input free)	$0.26 per second at 2K
`minimax/h3/reference-to-video`	Up to 9 reference images, 3 clips, 3 audio refs	$0.26 per second; first 5 images free, $0.08 each after; audio refs free

Do the arithmetic before you wire this into a busy group chat: a 6-second clip at 2K is roughly $1.56, and a 15-second one is about $3.90. Billing runs on your own fal key, so this is your spend, not a plan credit — and an agent that generates video on every casual request gets expensive quickly.

Reference-to-video is the one to care about. Feeding H3 the same character stills, the same reference clip, and the same voice sample across generations is what gives you continuity — same face, same voice, same look — instead of a fresh roll of the dice per clip. That is the difference between a novelty and a series.

How to Set It Up

Create a fal key at fal.ai/dashboard/keys.
Add it on your API Keys page under fal.ai. The fal group in the Video dropdown stays locked until the key is saved.
On your instance card, open the Video dropdown and pick the fal.ai group in the left column.
Choose a MiniMax H3 row. The fal list is refreshed daily from fal's trending endpoints, so the exact H3 variants shown can shift — use the search box at the bottom of the dropdown and type “H3” to find them.
Ask your agent for a clip in chat. The generated MP4 comes back in the conversation.

Self-Hosted Configuration

Running your own OpenClaw? The video model id is the fal endpoint path with a fal/ prefix — the same shape the dashboard writes:

fal/minimax/h3/text-to-video
fal/minimax/h3/image-to-video
fal/minimax/h3/reference-to-video

Set your fal credentials in the environment and point the video generation model at one of those ids. Note the path is minimax/h3/... — not hailuo-03, which is how several launch write-ups reported it.

H3 vs Hailuo 2.3

	Hailuo 2.3 (MiniMax key)	H3 (fal key)
Resolution	1080p class	Up to 2K at 24 FPS
Audio	Silent output	Native 32 kHz stereo, generated in the same pass
Inputs	Text or a start image	Text, images, video clips, and audio references combined
Open weights	No	Yes, on Hugging Face under MiniMax's community license

If your agent only ever needs a short silent clip, Hailuo 2.3 on a direct MiniMax key is cheaper and perfectly adequate. Reach for H3 when you need sound in the same pass, 2K output, or reference-driven continuity. For the architecture and open-weights side of H3, see our breakdown of the H3 release.

What's Next?

Make videos with your bot — the full brief-to-MP4 flow
MiniMax API key on OpenClaw — chat, image, video, and music on a direct MiniMax key
fal.ai on OpenClaw — the provider that carries H3 and 20+ other video models
All supported models — chat, image, and video across every provider