What MiniMax-H3 does on RelayMe
MiniMax-H3 is a video model from MiniMax. In RelayMe it appears as a video node in the workspace, so you can call it from a browser without installing anything. Each run turns a text prompt into a clip between 5 and 15 seconds long, and you can choose any whole number of seconds in that range. Output comes in two resolution tiers, 768P and 2K, and every clip is generated with an audio track; there is no silent option for this model.
The node accepts aspect ratios of 16:9, 9:16, 1:1, 4:3, 3:4 and 21:9, plus an adaptive setting. What sets it apart from the other video models on RelayMe is the number of references it takes in one request: up to 9 images, 3 videos and 3 audio clips. That makes it a good fit when the shot has to follow existing material, such as a product photographed from several angles, a character sheet or a sound you want the clip to match.
Running MiniMax H3 without a local GPU
Many people searching for MiniMax H3 are trying to work out how to run it in ComfyUI or how much graphics memory it needs. This guide does not cover a local install. The point of the hosted route is that the GPU is not your problem: RelayMe sends the request to the model provider and returns the finished clip to your task history.
To start, open the workspace, add a text node for the prompt and a MiniMax-H3 video node, and connect the text output to the video node’s prompt input. If you have reference images, videos or audio, add them as inputs and connect them to the node’s reference ports. Set the duration, resolution and aspect ratio on the node and run it. The clip appears in the task history with the credits it actually used, and you can download it from there. Generation takes minutes rather than seconds: our 5-second 768P test on 30 September 2026 took 174 seconds.
If you are weighing a local ComfyUI setup against hosted credits, see ComfyUI alternatives without a local GPU for what each option costs and what it cannot do.
Choosing duration, resolution and references
Duration and resolution are the two settings that change the price, so decide them on purpose. Five seconds is enough to judge motion, lighting and whether the subject stays consistent; longer clips are worth it only after the prompt works. 768P is the drafting tier. Move to 2K for the final version that will be published or edited, because the 2K rate is about two thirds higher per second.
References are where MiniMax-H3 earns its place. Use images to lock what must not change, such as the product, its label, a face or a set. Use a reference video when the motion or camera move matters more than the look. Use an audio clip when the soundtrack has to follow a given rhythm or voice. Only upload material you have the rights to use. More references are not automatically better: start with the one or two that define the shot, and add more only when the result drifts away from them.
How to write a MiniMax H3 prompt
A prompt that names the subject, the action, the camera and the sound gives the model fewer things to guess. A structure that works well is: subject and setting; what happens over the clip, in order; camera framing and movement; lighting and style; and what the audio should contain. For example: “A ceramic coffee mug on a wooden kitchen counter in morning light. Steam rises as a hand slides the mug toward the camera. Slow push-in from a medium shot to a close-up. Warm, soft natural light, shallow depth of field. Audio: quiet kitchen ambience and a soft ceramic clink.”
When you use references, say what each one is for, for example “keep the mug shape and logo from the reference images” or “match the camera move in the reference video”. Keep one idea per clip: if the shot needs a product turntable and a lifestyle scene, make two clips. Change one thing between runs, so you know which change fixed the result. Once a prompt works, keep it in a text node and reuse the workflow instead of retyping it.
What a MiniMax-H3 clip costs
MiniMax-H3 is billed per second of output. On RelayMe the current rate is 120 credits per second at 768P and 200 credits per second at 2K, with audio included in both. One thousand credits correspond to one US dollar of balance, so a 5-second 768P clip costs 600 credits (about US$0.60), a 10-second 768P clip 1,200 credits, and a 15-second 2K clip 3,000 credits. Check the pricing page before a large batch, because rates can change.
Our dated run: on 30 September 2026 we generated a 5-second 768P clip from a product scene image and a prompt. It was charged 600 credits and took 174 seconds to finish. Treat this as one example rather than a quote. The practical habit is to budget for a few drafts at 768P and five seconds, then one final render at the length and resolution you will publish.
Check the current pricing and credit rules before selecting a model; a dated example is not a current quote.
When to choose another video model
MiniMax-H3 is not the only option in the same workspace, and the cheapest choice depends on the job. Grok Imagine is 150 credits per second and also goes up to 15 seconds, which suits longer text-to-video drafts. Veo 3.1 Fast is 150 credits per second with audio or 100 without, makes 4, 6 or 8 second clips and can render 4K, which suits short ads where a silent version or 4K output matters.
Choose MiniMax-H3 when the clip has to stay close to several reference images, a reference video or a reference sound, or when you want audio on every take without paying a separate rate. If you are unsure, connect the same prompt to two video nodes and run both once; the task history shows each result next to the credits it used. The Grok Imagine vs MiniMax-H3 comparison covers the differences in more detail.
Check Veo 3.1 vs Veo 3.1 Fast and Grok Imagine vs MiniMax-H3 for the inputs, duration settings and prices of each model.