Boreal-H3 Text to Video

Boreal-H3 Text to Video: AI Text-to-Video Model

What is Boreal-H3 Text to Video?

Boreal-H3 Text to Video is Creatify's text-to-video model, a post-train of the MiniMax H3 base narrowed for advertising and short-form social video. Give it a written prompt and it returns a finished clip, 1 to 15 seconds, at 768p, 1088p or up to 2k, with native dialogue and sound on every clip. Its signature is the look: a real-camera, phone-footage aesthetic — vlog, UGC-ad and documentary realism — rather than a graded cinematic sheen, which is exactly what reads as native social and ad creative.

Key Features

• •Prompt-only text to video, with native dialogue and sound baked into every clip.
• •Phone-footage, UGC realism by default — handheld and documentary rather than a polished film grade.
• •Follows the brief as a shot: subject, action and camera direction, with legible cause-and-effect motion.
• •Output at 768p, 1088p or 2k (1440 px on the short edge), 24 fps, H.264 video with AAC audio.
• •Whole-second length from 1 to 15, and six aspect ratios including vertical 9:16 for social.
• •Fast to a finished clip, so you can iterate many variants; prompt_enhancement rewrites or keeps your prompt literal.

Best Use Cases

UGC-style ads and social posts are the sweet spot: talking-head creators, lifestyle and product demos, and quick shot variants for Reels, TikTok and Shorts. Because it renders people and motion that read as genuinely filmed on a phone, and carries its own audio, a single prompt can produce a social-ready cut — a maker talking to camera while they work, a pour that behaves like liquid, a demo where the described action actually happens. The tested default, a ceramicist at a wheel talking to camera, shows the pattern: a real person, a spoken line, and believable motion in one pass.

Prompt Tips and Output Quality

Write the prompt as one shot — subject, action, camera move, and any spoken line in quotes. Name ambient sounds (a wheel humming, a sizzling pan) to fill the audio bed; returned audio runs loud. There is no seed, so vary the prompt to explore different takes. In testing, a 9:16 clip of a potter talking to camera came back with convincing handheld realism, synced speech and clean hand-on-clay motion. Start at 768p to iterate, then re-run the same prompt at 2k for delivery.

How to Use Boreal-H3 Text to Video

Boreal-H3 turns a written prompt into a short, finished clip with its own dialogue and sound. Think of the prompt as a shot list in one paragraph: who or what is on screen, the action, the camera move, and any spoken line or ambient sound.

Writing the prompt

Lead with a concrete subject and a single clear action, then the camera move (handheld, slow push-in, locked off). Put spoken dialogue in quotation marks so it lands as speech, and name the sounds you want in the mix. Keep brand names, logos and real people out of the frame — the model can render a logo unprompted, and generic subjects stay more reliable. The default aesthetic is real-camera and handheld, so you rarely need to ask for a cinematic look.

Choosing settings

Resolution sets detail: 768p to iterate quickly, 1088p or 2k for a final. Duration is any whole number of seconds from 1 to 15; a 15-second request returns close to 14.4 seconds. Aspect ratio follows your platform — 9:16 for Reels, TikTok and Shorts, 16:9 for landscape, 1:1 for feed; auto is 16:9 when no image is supplied. Leave prompt_enhancement on auto to let the model rewrite the prompt into the form it works best with, or switch to none when you need the wording kept exactly and the result repeatable, since there is no seed.

Recommended Settings

The default example — a ceramicist at a pottery wheel talking to camera — was rendered at 768p, 9:16, 5 seconds, with prompt_enhancement on auto, and came back with convincing phone-footage realism, synced speech and clean motion.

• •Social UGC or talking-head: 9:16, 768p, 5–8 seconds, with a quoted line in the prompt.
• •Product or lifestyle demo: 16:9 or 9:16, 1088p, naming the cause-and-effect action you want to see.
• •Final delivery: 2k, re-running the same prompt you proved at 768p.

Examples to try

Each example ran end to end at the settings shown — 768p, 5 seconds — with its own dialogue or sound baked in, so what you see and hear is one prompt's return.

• •Talking-head UGC with native dialogue (default). Prompt: "Handheld phone video of a potter at her wheel shaping a tall bowl from spinning wet clay, water running over her hands, the rim rising between her fingers. She glances up at the camera and says, "Feel how thin that wall is getting." Daylight through a studio window, the wheel humming and wet clay slapping." She keeps throwing while she looks up and speaks, lips tracking the voice over the wheel's hum. Quote the spoken line and keep to one clear speaker.
• •Fluid motion and ambient sound, no dialogue. Prompt: "Close handheld phone video of a barista pouring steamed milk from a metal jug into a white cup of espresso, a leaf of latte art blooming on the surface as the milk folds in. Steam rises from the cup. Daylight from the cafe window; the hiss of the steam wand and clink of the cup, low cafe chatter." The milk pours like real liquid and settles into a finished rosetta, cafe sound under it. No speaker needed — name the action and the sounds and the audio fills in.
• •Wide 16:9 landscape with a chain reaction. Prompt: "Wide handheld phone video, landscape, of a hand tipping the first tile in a long row of red and white dominoes on a wooden table; the tiles topple one after another in a running chain down the line, each knocking the next. Daylight across the table; the rapid clack of falling dominoes." One tap sends the whole row down in sequence across the wide frame. Set aspect_ratio to 16:9 for landscape, and expect the audio back loud.

Try other single-shot ideas the model is built for: "A mechanic wipes her hands and explains a quick fix to camera", or re-run a proven prompt at 2k once it's locked at 768p.

Does Boreal-H3 make sound?

Yes — native dialogue and sound on every clip. Include a spoken line in the prompt and it comes back as speech.

Can I keep a product or person consistent across a clip?

Use the image-to-video or reference-to-video modes, which take a start frame or reference media; this text-to-video mode is prompt-only.

Is there a seed?

No — vary the prompt rather than a seed to get a different result.

What length and resolution can I get?

Any whole number of seconds from 1 to 15, at 768p, 1088p or up to 2k.

Which aspect ratios are supported?

auto, 16:9, 9:16, 1:1, 4:3 and 3:4; auto is 16:9 when no image is supplied.