guide
4 min read

MiniMax H3 on film.fun: Multimodal Video with 2K Output and Reference Everything

F

Film Fun Academy

August 19, 2026

MiniMax H3 is now live on film.fun.

MiniMax's new multimodal video model is wired into Splice as minimaxH3 on the Replicate path — available across the video generation surfaces alongside Seedance, Kling, LTX, and the rest. It's the sharpest option in the lineup, with true 2K output.

What it does

H3 is a genuinely multimodal video model — it takes text, images, video, and audio as inputs and turns them into a single generation:

  • Text-to-video — describe the shot and go.
  • Image-to-video with first/last frame. Set a first-frame image to drive the opening (and the aspect ratio), add a last-frame image to control where the shot lands, and H3 fills the motion between.
  • Reference images, videos, and audio. Feed public references for subject and style, motion, or sound — H3 conditions the generation on all three.
  • Resolution up to 2K. Choose 768P for speed and economy, or 2K when you need the crispest frame in the toolbox.
  • Shots from 4 to 15 seconds, so you can size a clip to the beat.

How to prompt it

  • Describe the frame like a cinematographer. Subject, lens feel, light, movement — H3 rewards concrete visual direction over story summary.
  • Use a first-frame image to anchor identity and ratio. For image-to-video, the first frame sets the aspect ratio automatically, so you don't have to fight the crop.
  • Layer references for consistency. A reference image keeps a character on-model across shots; a reference video transfers a camera move; reference audio shapes the mood.
  • Go 2K for hero shots, 768P for drafts. Iterate cheap at 768P, then re-run your keeper at 2K.

When to reach for it

Reach for H3 when you want maximum sharpness or heavy reference control — a 2K hero shot, or a scene where you're steering with images, a motion clip, and audio all at once. For long, audio-first takes, Seedance 2.5 is the sibling to try; for crisp, reference-driven shots, H3 is the pick.

Priced per second of output — 768P is the economical default, 2K costs a little more for the extra resolution. Pick MiniMax H3 in Create → Video, start at 768P to dial in your prompt, then re-run your favourite at 2K.

📬 Enjoyed this? Get weekly AI filmmaking tips

Join thousands of creators getting guides like this delivered to their inbox every week.