Boreal-H3 Image to Video

Boreal-H3 Image to Video: Animate a Still into a Talking Clip

What is Boreal-H3 Image to Video?

Boreal-H3 Image to Video is Creatify's image-to-video model, a post-train of the MiniMax H3 base tuned for advertising and short-form social. Give it one still as the first frame and a prompt, and it returns a finished 1 to 15 second clip that opens on your exact image and animates it, with native dialogue and sound on every clip. Its signature is the look: real-camera, phone-footage realism, the vlog, UGC-ad and documentary feel that performs natively on social rather than a graded cinematic sheen. Add an optional last frame and the shot transitions from your first frame to it.

Key Features

• •First-frame keyframe control: the clip opens on the uploaded still and keeps its composition; the output follows that image's aspect ratio.
• •Optional last frame for a start-to-end transition.
• •Native dialogue and sound on every clip, no separate audio pass.
• •Output at 768p, 1088p or 2k (1440 px short edge), 24 fps, H.264 video with AAC audio.
• •Whole-second duration from 1 to 15; identical requests return identical clips (no seed).

Best Use Cases

Turn a product photo, a creator headshot or a scene still into native social and ad creative: talking-head UGC, product and lifestyle demos, and quick motion variants for Reels, TikTok and Shorts. Because the clip starts from your exact frame, it suits work where composition must be preserved, animating a hero shot, a packshot or a storyboard keyframe. In testing, a still of a ceramicist at her wheel animated faithfully: the wheel kept spinning, clay rose between her fingers, and she glanced up to speak a line, with clear native audio.

Prompt Tips and Output Quality

Describe what should move in the still, the camera, and any sound. Put a quoted line in the prompt to get native dialogue, and name ambient sounds to fill the audio bed. Keep the first frame clean, with obvious motion potential. Returned audio runs loud, so normalize it downstream. Use none for prompt_enhancement when you need literal, reproducible wording, and keep brands, logos and real people out of frame.

How to Use Boreal-H3 Image to Video

Boreal-H3 Image to Video animates a single still into a finished clip with native sound. Upload a first frame, describe what should move in it, and the model opens on that exact image and brings it to life, keeping its composition and aspect ratio.

Write the prompt as a shot

The first frame fixes the scene, so the prompt is where you direct the motion. Name what moves, the camera, and any sound: "the wheel keeps spinning, she glances up and says '...', wheel humming." Quote a spoken line to get native dialogue, and name ambient sounds to populate the audio bed. Concrete, physical actions animate more cleanly than vague mood words.

Choose a good first frame

Upload a clean still whose subject can obviously move: a person who can speak or gesture, liquid that can pour, a wheel that can turn. The output follows the first frame's aspect ratio, so frame vertically for social and horizontally for landscape. Keep brands, logos and real people out of frame.

Set where the shot ends (optional)

Add a last frame to transition from the first frame to it, for a reveal or a before-and-after. Match the framing between the two stills. Leave it empty for a single clean animate.

Recommended Settings

• •resolution: 768p to iterate quickly; 1088p or 2k (1440 px short edge) for finals.
• •duration: whole seconds from 1 to 15; default 5. A 15-second request returns about 14.4 seconds.
• •prompt_enhancement: auto (default) rewrites your prompt and adds latency; switch to none for literal, reproducible wording.
• •No seed: identical requests return identical clips, so vary the prompt to get a different take.

In testing, a still of a ceramicist at her wheel, run at 768p for 5 seconds, animated faithfully: the wheel kept spinning, clay rose between her wet fingers, and she glanced up to speak a line, with clear native dialogue and wheel-and-clay ambient. Returned audio runs loud, so normalize it before publishing.

Examples to try

Each example below ran end to end at 768p for 5 seconds. The clip opens on the exact still you upload, keeps its framing and aspect ratio, and carries native sound.

• •Talking-head UGC from a still. Prompt: "The potter's wheel keeps spinning and the tall bowl rises between her wet fingers, water running down the grey clay as she steadies the rim. She glances up at the camera and says, "Feel how thin that wall is getting." Her hands return to shaping the clay. Wheel humming and wet clay slapping, daylight from the studio window." The still comes alive — the wheel turns, the bowl rises and she speaks the line, lips and voice in sync. Tip: quote the spoken line and keep to one clear speaker.
• •Cause-and-effect liquid motion. Prompt: "The barista tilts the steel jug and pours steamed milk into the white cup of espresso; a leaf of latte art blooms across the surface as the milk folds in and steam curls up. The hiss of the steam wand, a cup clink and low cafe chatter." The milk pours like real liquid and settles into a latte-art leaf over cafe ambient, no speaker needed. Tip: name the action and the sounds to fill the audio bed.

Setting where the shot ends (two frames)

This example adds a last frame to the first: the clip transitions from the opening still to the closing one, so match their framing.

• •Start-to-end transition. Prompt: "The peony bud unfurls, its deep-pink petals opening layer by layer until the flower stands fully bloomed. A faint rustle of petals and quiet room tone." The clip opens on the closed bud and lands on the fully open bloom — a clean before-and-after reveal. Tip: leave the last frame empty when you want one continuous animate instead.

Try other single-still prompts the model is built for — "she turns the phone screen to the camera and nods", or "steam rises as the chef slides the plate forward" — one clear action plus the sound you want to hear.

Does it generate sound?

Yes, native dialogue and sound on every clip, with no separate audio pass.

What image should I upload?

A clean first frame; the clip opens on it and follows its aspect ratio.

Can I control where the shot ends?

Add an optional last frame for a start-to-end transition.

How long can a clip be?

Any whole number of seconds from 1 to 15; a 15-second request returns about 14.4 seconds.

Is there a seed?

No. Identical requests return identical clips, so vary the prompt instead.