From Images to Video: Seedance I2V — Props Race & Fight (Nezha & Ao Bing)
With a start image and an end image, you can ask AI to invent the motion between them. Seedance 1.5 Pro on FuseAI Tools accepts reference images and a rich text prompt to steer what happens in the middle—here, a 100m sprint upgraded into a prop-heavy, comedic duel (ribbon snag, ice on the track, weapon clash) before the finish.
If you have not generated the two stills yet, start with the paired prompts and parameters in the companion guide: AI First & Last Frame Images: Nezha vs Ao Bing’s 100m Sprint — Seedream Tutorial (tool: Seedream 5 Lite Text to Image).
Note: Myth-inspired characters in modern sportswear; not affiliated with any specific film IP.
I. What Is Image-to-Video (I2V)?
- Inputs: typically 1–2 images (your keyframes).
- Process: the model imagines intermediate motion and lighting changes.
- Output: a continuous clip—stronger character control than pure text-to-video when your frames already match.
II. Creative Upgrade: Plain Sprint → Props + Fight Beats
We layer beat-specific instructions onto the timeline so the model has “checkpoints” to hit:
| Time | Gag / beat | Motion cue |
|---|---|---|
| ~2s | Red ribbon toss | Ribbon wraps an ankle; rival steadies |
| ~4s | Ice counter | Frost on lane; jump or slide escape |
| ~6s | Side-by-side clash | Circular weapon vs ice hammer, sparks |
| ~8–12s | Drive to tape | Nezha noses ahead; Ao Bing close second |
III. Example Request Body (Seedance 1.5 Pro)
Use on Seedance 1.5 Pro. Field names match typical API shape: model, prompt, inputUrls, aspectRatio, resolution, duration (allowed values include "4", "8", "12"), fixedLens, generateAudio.
{
"model": "seedance-1.5-pro",
"prompt": "Nezha and Ao Bing explode off the starting line together. Around 2 seconds, Nezha throws a long red ribbon that snags Ao Bing's ankle; Ao Bing wobbles but recovers. Around 4 seconds Ao Bing retaliates with frost magic icing the lane; Nezha leaps and then skates on blazing wheel-like foot glow to stay upright. Around 6 seconds they sprint side by side: Nezha swings a circular ring weapon while Ao Bing raises an ice hammer—they clash mid-air with sparks. Final stretch: both accelerate; Nezha breaks the tape first by a hair, Ao Bing right behind. Camera tracks alongside the runners, cinematic sports broadcast meets fantasy action, crisp 1080p motion, readable faces, smooth continuity from start pose to finish pose.",
"inputUrls": [
"https://media.fuseaitools.com/image/bb25dd35ae8c5ca6e1049f1f3efd758b_1774686390_75vx5nfh_11de6c478eff479a9efaa0812681e62f.png",
"https://media.fuseaitools.com/image/d84617fbd61c9936888c12b60a49f0a2_1774686762_8yfb6ny4_36425f2733464b0d894420b2e419a917.png"
],
"aspectRatio": "16:9",
"resolution": "1080p",
"duration": "12",
"fixedLens": false,
"generateAudio": true
}Why these settings
- duration "12" — room for multiple beats without rushing.
- fixedLens: false — allows camera drift/parallax for action.
- generateAudio: true — synced SFX / ambience when enabled (uses more credits per pricing rules).
- inputUrls — same two frames as in the Seedream first/last frame article.
IV. Example Video
Example output — props race + clash + finish (sample)
V. Storyboard & Sound Cues (Expected)
| Segment | Picture | Sound (when audio on) |
|---|---|---|
| 0–2s | Launch from blocks | Starter signal + spikes on track |
| 2–3s | Ribbon catches ankle | Fabric whip + stumble grunt |
| 4–5s | Ice spreads on lane | Crystallize SFX + jump |
| 6–8s | Weapon clash beside sprint | Metal ring + impact sweetener |
| 8–10s | Neck-and-neck drive | Rhythmic percussion bed |
| 10–12s | Tape break; rivalry smile | Crowd swell + light sting |
VI. Why This Concept Helps Discoverability
- Unique hook — “myth rivals + modern track + fantasy props” is a memorable bundle.
- Long-tail queries — image-to-video tutorial, two-frame workflow, AI fight scene, etc.
- Session time — readers who watch the sample clip signal quality to search and social algorithms.
VII. Tuning Tips
| Goal | What to change |
|---|---|
| Longer brawl | Try duration "12" with more mid-prompt detail, or split into two clips. |
| Slower hero hits | Add “brief slow-motion on the clash” to prompt. |
| More gadgets | Name props explicitly once each; avoid ten new objects in one pass. |
| Comedy tone | Add “lighthearted, exaggerated reactions.” |
VIII. Simple Sprint vs This “Props + Fight” Version
- Plain sprint — shorter prompt, fewer failure modes.
- This build — higher entertainment value; needs clearer time-stamped prompt language.
IX. FAQ
Will the model always nail the weapon clash?
Concrete nouns (“ring weapon,” “ice hammer,” “sparks”) help. Regenerate or tighten timing phrases if a beat slips.
Are two input images enough for two characters?
Both should already appear in the start and end frames; I2V inherits that casting.
Can I add a third character?
Not without new key art. Generate new Seedream frames that include everyone, then re-run Seedance.
X. Summary
Seedance turns your Seedream first/last frames into motion when you (1) keep frames consistent, (2) write a time-structured prompt, and (3) pick duration, camera lock, and audio to match the story budget. Reuse this skeleton with your own cast—swap URLs, keep aspect ratio aligned, and iterate.
Frame generation walkthrough: Seedream — Nezha & Ao Bing first/last frames.