Write better prompts.
Get better videos.
A practical reference for every tool we ship. Formula, examples, and what works best with each model.
The prompt formula
Every great prompt has the same backbone. Subject + Motion are required — without them the model has nothing to anchor the scene. The other four tokens make it cinematic.
Read left to right and the model has a story arc: who, doing what, where, looking like, shot how, sounding like. Skipping any of the optional pieces is fine — but don't skip Subject or Motion.
Example, broken down
A muscular hummingbird hovers near a glowing crystal flower at twilight, soft volumetric light streaming through palm fronds, anamorphic 35mm lens, slow dolly-in, faint ambient pad with insect chirps.
General tips
Which model when?
Belvora ships three families. Each excels at different things — picking the right one costs you less time and credits.
| Model | Strengths | Best for | Tier |
|---|---|---|---|
| Seedance 2.0 | Multi-modal refs (9 imgs + 3 vids + 3 audios), highest control, all aspect ratios. | Reference-to-video, ad creatives, brand consistency. | Fast or Standard |
| Runway Gen-3 | Cinematic motion, strong camera moves, fast. | Story scenes, action, real-world look. | Fast only |
| Luma Dream | Stylized / oneiric output, soft motion. | Mood pieces, animated styles, low-stakes iteration. | Fast only |
Working with references
References pin specific aspects of the output (subject identity, motion, style) so the model doesn't drift. Use them when text alone can't carry the idea.
Images
- 1 image — anchors the subject. Use Image → Video Fast/Standard or add as the first reference in Reference → Video.
- 2-3 images of the same subject from different angles — locks identity across the clip.
- Up to 9 images mixing subject + style — composes a scene. Reserved for Reference → Video.
Videos (Reference → Video only)
- Upload a clip whose motion or camera move you want to mimic.
- Max 15s total across all reference videos.
- Tag in prompt:
"same dolly-in as the reference"or"copy the choreography from the video".
Audio (Reference → Video only)
- Up to 3 audio files (MP3, total ≤15s).
- Useful for beat-synced cuts or to drive lip-sync.
Adding text to a video
Seedance and Runway can render legible text inside the frame. Use this for slogans, captions, lower-thirds.
Pattern
[Text content] + [Timing] + [Position] + [Style]Example
Opening shot of a city at golden hour. A title card "Made by Belvora" fades in at 0:02 in the lower third, white serif on a thin red bar, holds for 2 seconds, fades out.
Tips:keep text under 4 words for legibility; specify font weight ("bold sans-serif") and position ("upper third", "lower right").
Example library
Three prompts you can paste into Studio right now.
Cinematic close-up
Macro close-up of a dewdrop sliding down a green leaf at sunrise, shallow depth of field, golden rim light, slow 35mm tracking shot, gentle ambient hum.
Action / motion
A breakdancer mid-spin in slow motion, neon-lit underground tunnel, smoke curling around her, dramatic key light from above, anamorphic flares, bass-heavy electronic pulse.
Stylized / animated
Watercolor animation of a paper boat drifting on a stream, light brush strokes, soft pastel palette, top-down camera that slowly pans alongside, music-box melody.
Common mistakes
Why: Generic. No subject identity, no environment, no lens, no mood. Output will be average.
Why: "Cinematic" alone is meaningless. The model can't pick which 1000 references to lean on.
Why: Seedance and Runway reject real-people likenesses by partner policy. You'll burn 10+ minutes of compute and get refunded.
Why: "The camera moves around" can mean dolly, truck, orbit, handheld, drone. Pick one.
Ready to try?
Copy any example above and paste it into Studio. 50 free credits on us.