How to Chat-Generate AI Short Videos with Lovart — From Script to Visual Storyboard
Your content calendar demands three short videos this week. You open your video editor. The timeline is empty. You have no footage, no B-roll, no idea where to start. You spend two hours scrolling stock footage libraries, finding clips that sort of match your script but not really, and the result will look like every other stock-footage compilation on the internet — generic, disjointed, forgettable.
Short-form video is the dominant content format of the decade. Instagram Reels, TikTok, YouTube Shorts, LinkedIn video — every platform is pushing vertical video as its primary content type. But production has not democratized the way distribution has. Most creators have distribution access but not production capacity. An AI design agent bridges this gap: you describe the scene sequence, the agent generates the visual assets, you assemble them into a video.
Lovart is the AI design agent trusted by 10M+ creators. Try Lovart Free →
Related: 方法 Create a ブランド Kit for a Jewelry デザインer | How Wedding Planner Olivia Rousseau デザインed 62 Invitation Sui
Why AI-Generated Visuals Are Changing Short Video Production
Traditional video production requires you to either (a) film original footage (expensive, time-consuming, requires equipment and skill) or (b) license stock footage (limited selection, generic feel, licensing complexity). Both pathways create a ceiling on how much video content you can produce.
AI-generated visuals remove the ceiling. You are not limited by what you can film or what exists in a stock library. You describe the scene you need, and the agent generates it. This is not theoretical — content creators are already producing AI-generated video content at scale, using AI visual assets as keyframes, background plates, B-roll inserts, and animated scene elements.
The Chat-Generate AI Short Videos Workflow
Step 1 — Write a Visual Script
Do not start with prompts. Start with a script that maps words to visuals:
SCRIPT: "3 AI Tools That Changed My Workflow" (60-second vertical short)
SCENE 1 (0-5s): HOOK — ME, shocked expression, looking at my phone.
On screen text: "I tested 47 AI tools. 3 actually work."
Visual style: Phone-shot selfie aesthetic, my living room, natural light.
SCENE 2 (5-15s): TOOL 1 — A screenshot of the tool's interface,
floating in a clean 3D space with a subtle glow.
My voiceover explains what it does.
On screen text: "Tool #1: [Name]"
SCENE 3 (15-25s): TOOL 2 — Same visual treatment, different tool.
SCENE 4 (25-35s): TOOL 3 — Same visual treatment, different tool.
SCENE 5 (35-50s): DEMONSTRATION — Split screen. Left: me doing a task
the old way (struggling, exaggerated). Right: me doing it with the AI tool
(easy, exaggerated relief). Comedy beat.
SCENE 6 (50-60s): OUTRO — Me, relaxed, confident.
Text: "Which one would you try first? Comment below."
End card with CTA.
This script format gives the agent everything it needs: scene duration, visual description, text overlay, style reference, and emotional tone — per scene, in sequence.
Step 2 — Generate Keyframe Images for Each Scene
For each scene in your script, generate a keyframe — the defining still image of that moment:
Scene 1 Keyframe Prompt:
Vertical 9:16 frame for a short video. ME — shocked expression, wide eyes, mouth open,
looking at my smartphone in my right hand. Phone-screen selfie style — I'm holding the phone
as if filming myself. Living room background, morning light from a window on the left,
slightly messy (lived-in, not staged). On the phone screen, visible overlay text:
'I tested 47 AI tools. 3 actually work.' The overall aesthetic is TikTok-native —
real, unpolished, authentic. Warm color grade, slight film grain.
Scene 2 Keyframe Prompt:
Vertical 9:16 frame for a short video. A sleek software interface floating in a 3D space —
abstract dark gradient background with subtle blue particle effects. The interface shows
a clean dashboard with metrics and a glowing 'Generate' button.
Overlay text at the top: 'Tool #1: [Name]'. Below it, smaller text: 'What it does in 5 seconds.'
Cinematic, high-contrast, the interface is the hero. Dark mode aesthetic.
Scene 5 Keyframe Prompt:
Vertical 9:16 frame for a short video. Split screen composition —
left half: ME looking frustrated, papers everywhere, pulling my hair (comedic exaggeration).
Label at bottom: 'Without AI.'
Right half: ME looking relaxed, one click on a clean interface, everything organized.
Label at bottom: 'With Lovart.'
Bright, punchy, comedy-ad style. Same me in both panels for continuity.
Generate each scene's keyframe separately, reviewing and refining before moving to the next. This gives you fine control over the visual arc of the video.
Step 3 — Generate Transition and B-Roll Assets
The spaces between keyframes are where videos feel either professional or amateur. Generate transition assets:
Transition wipe asset: A smooth animated-style gradient swipe from left to right,
brand blue to transparent. Vertical 9:16. Designed to overlay between Scene 1 and Scene 2.
B-roll asset for Tool 1: Close-up macro shot of a cursor clicking a 'Generate' button
on a clean interface. Cinematic depth of field — button sharp, background soft.
Short looping motion suggestion.
End card asset: Brand gradient background (brand colors from kit).
Product logo centered. Text: 'Try Lovart free at lovart.ai'.
Subtle animated glow behind the CTA. Vertical 9:16. Clean, minimal, high-contrast.
Step 4 — Assemble and Time
Export all keyframes, transitions, and B-roll assets. In your video editor (CapCut, Premiere, DaVinci Resolve, or even Canva):
- Place keyframes on the timeline in script order.
- Add transitions between scenes.
- Dial in each keyframe's duration to match the script timing.
- Record and lay in your voiceover.
- Add motion — simple Ken Burns zooms and pans on static keyframes create the illusion of video.
- Export at 1080x1920, 30fps or 60fps.
Step 5 — Repurpose Across Platforms
Your 60-second vertical short is one asset. From the same visual library, generate:
HORIZONTAL CUT: Same keyframes, reformatted to 1920x1080 landscape.
Reposition compositions for horizontal framing. For YouTube.
SQUARE CUT: 1080x1080, tighter crops on key elements. For Facebook Feed and LinkedIn.
THUMBNAIL: Horizontal 1280x720 YouTube thumbnail using the best single frame,
with text overlay '3 AI Tools I Actually Use.'
One content piece, four platform formats, one visual library.
AI Short Video Prompt Cheat Sheet
E-E-A-T: Evidence and Platform Insights
The short video generation workflow in this article combines content production techniques from social media creators with AI-assisted visual generation. The "script-first, keyframe-by-keyframe" approach emerged from Lovart users producing consistent short-form video content across TikTok, Instagram Reels, and YouTube Shorts in 2026.
Lovart provides full commercial rights on all generated visuals used in your videos. Your video content is your asset — publish it, monetize it, repurpose it without restriction.
Frequently Asked Questions
Does Lovart generate full videos or just images?
Lovart generates the visual assets — keyframe images, transition elements, end cards, B-roll frames. You assemble these assets into a video using any video editor. This workflow gives you complete control over timing, pacing, and audio while eliminating the hardest part — creating the visuals.
What video editors work best with Lovart-generated assets?
Any editor that supports image sequencing works. CapCut (free, mobile-first) is the most popular among short-form creators. DaVinci Resolve (free, professional) for higher production value. Canva (drag-and-drop) for beginners. Premiere Pro for those already in the Adobe ecosystem.
Can the agent generate animated elements?
The agent generates static images with motion suggestions. For simple animations, export your keyframes and apply Ken Burns effects (zoom/pan) in your editor. For more complex animation, use the generated images as reference in animation software or AI video tools that accept image input.
How many keyframes do I need for a 60-second video?
Minimum 6-8 keyframes (one every 7-10 seconds). More keyframes = more visual variety = less boring. For fast-paced content, generate 10-15 keyframes and spend 3-5 seconds on each. The visual change itself holds attention.
Can I use the same visual style across multiple short videos?
Yes — this is the purpose of the style reference approach. Define a "channel style" at the start of your prompt series: "All scenes in this series: dark mode aesthetic, brand blue accents, high contrast, cinematic lighting." Every keyframe will share this visual DNA.
What about audio and voiceover?
Audio is the domain of your video editor. Generate your visuals in Lovart, then record voiceover, add music, and sync everything in your editor. The visual pipeline is decoupled from the audio pipeline — each does what it does best.
How does this compare to fully AI-generated video tools like Sora or Runway?
Sora, Runway, and similar tools generate complete video clips — motion and all — from text prompts. Lovart generates the high-quality static visual assets that you assemble into video. The Lovart approach gives you more control over composition and branding but requires manual assembly. The fully-AI approach gives you complete clips but less control. They are complementary — many creators use both.
