How to Use MiniMax H3 Max in 2026: Fast AI Video with Sound

Zoey Zhang

Oct 2026

MiniMax H3 Max turns a start frame into a video with sound, up to 15 seconds long, and it is fast: our 5 second 768p clip came back in 16 seconds. It is fal's post-trained version of MiniMax H3 (Hailuo 3), tuned for prompt adherence and looks. On Flick it runs in the image-to-video node and in Omni Reference, and it is available on every plan, including Free.

Quick answer: how do you use MiniMax H3 Max?

Connect a start frame to a video node, pick MiniMax H3 Max, and describe the action. Then:

  1. Start at 480p or 768p. 20 or 40 credits for 5 seconds, back in seconds.
  2. Name the action. It acted on "the woman is laughing" but barely moved on "animate it".
  3. Save 2K for the keeper. 2K adds a second pass: about 2.5 minutes in our runs.
SpecMiniMax H3 Max on Flick
Base modelMiniMax H3 (Hailuo 3), post-trained by fal
ReleasedMiniMax H3: July 31, 2026; H3 Max on fal: August 27, 2026
Modes on FlickImage to video (start and optional end frame), Omni Reference
Length5 to 15 seconds
Resolution480p (default), 768p, 2K
AudioNative, generated with the picture
Price20 credits for 5 s at 480p; double at 768p, four times at 2K
Prompt expansionBalanced (default) or Quality
PlansEvery plan, including Free
Measured time768p, 5 s: 16 s. 2K, 5 s: 2 min 34 s and 2 min 41 s

Animate a still in seconds

MiniMax H3 Max: sound included, from 20 credits.

What MiniMax H3 Max is

MiniMax launched H3 on July 31, 2026 as a multimodal model that generates video with native stereo sound, up to 15 seconds and 2K by default. fal then released H3 Max, "a post-trained variant of MiniMax H3" tuned for "stronger prompt adherence and better aesthetics" and paired with fal's own inference stack for throughput.

Flick's version follows your start frame, adds sound, and offers three resolutions. 480p and 768p come back in seconds; 2K runs a second pass and takes minutes.

Test 1: one clear action

On an illustrated portrait of a woman in a floral dress before pink flames, the prompt was four words: "the woman is laughing, fire". 15 seconds, 768p.

MiniMax H3 Max animates an illustrated woman in a floral dress before pink flames: the camera pushes in from a wide profile to a close-up of her laughing with her eyes closed

H3 Max did exactly what was asked, and more: the camera pushed in from the full figure to a close-up of her laughing face, and the flames were redrawn in a cleaner, flatter style than the painted original. A second take from the same prompt made the same push-in.

Test 2: an action with a prop

On a photographic office scene, "The man angrily throws his notebook into the air while everyone else looks at him. no background music". 5 seconds, 768p.

MiniMax H3 Max shows a man in a patterned orange jacket in a crowded office hurling his notebook upward; the shot cuts to the notebook flapping open against the ceiling, then to his colleagues looking up

It cut the 5 seconds into three angles: the throw, the notebook flapping open against the ceiling lights, and the colleagues watching it fall. It finished in 16 seconds.

Test 3: a vague prompt

"animate it" (15 seconds) and "enjoy your life" (5 seconds), both at 768p on painted stills:

MiniMax H3 Max on a painted portrait of a woman among pink flames and deer, prompted only with "animate it": the flames flicker and the figure stays almost still for 15 seconds
MiniMax H3 Max on a painted scene of two people on an orange sofa under a white umbrella in a green field, prompted with "enjoy your life": the umbrella sways and the figures barely move
First and last frames of the two vague-prompt clips: the flame portrait and the sofa scene look almost identical at the start and end
First and last frames of the two vague-prompt clips: the flame portrait and the sofa scene look almost identical at the start and end

With nothing to act on, H3 Max barely moved: the flames flicker, the umbrella sways, the people stay put. That is useful when you want a living still, and a waste of 15 seconds when you wanted a scene.

What it costs and how long it takes

Length480p768p2K
5 s204080
10 s4080160
15 s60120240

Credits on Flick. Rates for every model are on model credit costs; a failed generation is refunded (how credits work).

Our timed runs: 16 seconds for 5 seconds at 768p; 2 min 34 s and 2 min 41 s for 5 seconds at 2K; 42 to 52 seconds for five 15 second reference-to-video runs at 768p. Output at 768p is 1344 × 768.

Prompting MiniMax H3 Max

  • Write the action, not "animate". A named action produced a full performance; "animate it" produced flicker.
  • Expect camera moves. It pushed in on a laugh and cut between angles on the notebook throw without being asked; add "(fixed camera)" if you need it locked.
  • Check the style on illustrated frames. It redrew painted flames in a flatter style.
  • Know which prompt expansion you ran. Balanced is the default and every test here used it; Quality is the other option in the settings.
  • Write sound into the prompt, or rule it out. "no background music" is a valid instruction.
  • Motion not what I asked covers the usual fixes.

Errores comunes

  • Rendering 2K while exploring. It takes minutes instead of seconds; find the take at 768p.
  • Filling 15 seconds with one word. Long clips need a sequence of actions or they stall.
  • Assuming it keeps a painted style exactly. Compare the last frame to your start frame.

Sources

Preguntas frecuentes

What is MiniMax H3 Max?

How much does MiniMax H3 Max cost on Flick? 20 credits for 5 seconds at 480p, 40 at 768p and 80 at 2K, up to 240 for 15 seconds at 2K.

How much does MiniMax H3 Max cost on Flick?

How fast is it? Our 5 second 768p run took 16 seconds. 2K took about 2.5 minutes.

How fast is it?

Is it on the Free plan? Yes, on every plan.

Is it on the Free plan?

MiniMax H3 Max or Seedance 2.5? H3 Max for fast, cheap takes up to 15 seconds; Seedance 2.5 for longer clips and more elaborate direction. See our comparison.