AI Video Script Generator for TikTok, Reels, Shorts
Short-form video scripts that work aren't paragraphs, they're scene lists. Each scene has a duration, a voiceover line (or no voiceover and silent on-screen text), and what actually appears on screen. A 30-second TikTok is 5-7 scenes; a 15-second Reel is 3-4. scrb generates scripts in that exact shape, tuned per platform - TikTok leans native-sounding and self-deprecating, Reels leans polished, Shorts leans punchy and informational. Each script is structured JSON you can hand to a creator or paste into CapCut.
What scrb knows about short-form video scripts that a chat model doesn't
Chat models write you a monologue. Video creators need a scene list with per-scene timing and split audio / text tracks.
- Scene-list JSON:
{scenes: [{duration_s, voiceover, on_screen_text}]}. Each scene is one camera cut. Durations sum to the target length (15, 30, 45, 60s). - Platform-aware: TikTok scripts lean lowercase, native voice, hook in 0-2s. Reels scripts lean polished, brand-visible, CTA in the last 3s. Shorts lean informational, often a single-shot monologue with title overlay.
- Hook-first discipline: scene 1 is always a hook - a question, a counter-intuitive claim, a fast product cutaway. Never "Hi guys, welcome back to my channel". Never "Today I'm going to talk about". Scene 1 voiceover hits before second 2 or you lose the scroll.
- Voiceover separate from on-screen text: on-screen text is shorter (≤4 words) and often says something different from the VO. This split matches how the top-performing creators actually edit - the voice carries the narrative, the text carries the punch line or stat.
- Silent-video mode: opt-in mode where voiceover is blank and all the narrative is in the on-screen text. Matches the ~60% of viewers who watch muted. Each scene's on-screen text becomes longer (up to 12 words) and the scene duration bumps 0.5-1s.
- Category fit: product-demo scripts differ from founder-story scripts differ from behind-the-scenes. Scrb picks a structure template based on the input context.
- CTA per platform: TikTok = "save this" / "comment X". Reels = "follow for more" / "link in bio". Shorts = "subscribe" / "watch the next". Scrb picks the native-feeling CTA per platform instead of pasting a generic "like and subscribe".
- 25+ languages native: scripts generated directly in French, German, Japanese, Portuguese, etc. Not transliterated captions.
Before / after: real examples
Generic AI output:
Scene 1: "Hi everyone! Welcome back to my TikTok. Today I want to talk about how AI is revolutionizing the Etsy selling world!"scrb output:
Scene 2: "With so many sellers struggling, AI tools can unlock your potential..."
Scene 1 (0-2s):Why this wins: scene 1 gives a full stop hook in under 2 seconds. Concrete numbers in every scene (140 chars, 13 tags) instead of vague claims. On-screen text is short enough to read on mobile. CTA is native TikTok (comment to get DM'd).
VO: "stop using ChatGPT for your etsy listings"
On-screen: "STOP."
Scene 2 (2-7s):
VO: "etsy titles are capped at 140 chars. chatgpt writes you 180. you're leaving keywords on the table every time."
On-screen: "140 chars. chatgpt gives 180."
Scene 3 (7-14s):
VO: "also etsy wants all 13 tags. chatgpt gives you 5. that's one of the biggest ranking factors."
On-screen: "13 tags > 5 tags."
Scene 4 (14-22s):
VO: "use a tool trained on etsy specifically. it fills every field. every time."
On-screen: "fill every field."
Scene 5 (22-30s):
VO: "comment 'etsy' and i'll send you what i use."
On-screen: "comment etsy ↓"
scrb output:
Scene 1 (0-2s):Why this wins: concrete stat (40%), specific regulatory fact (FDA top allergen, "parfum" loophole), brand CTA that sounds native to Reels (link in bio, not "follow for more"). 4 scenes over 15s is the right pacing for Reels - not so fast it blurs, not so slow the scroll kicks in.
VO: "Why we don't put fragrance in our moisturizer."
On-screen: "Why no fragrance?"
Scene 2 (2-6s):
VO: "Fragrance is the top allergen the FDA tracks in skincare."
On-screen: "#1 tracked allergen"
Scene 3 (6-11s):
VO: "It's in 40% of moisturizers. Usually not listed by name on the label."
On-screen: "'parfum' = 100+ chemicals"
Scene 4 (11-15s):
VO: "Ours smells like nothing. Your skin will thank you. Link in bio."
On-screen: "Link in bio."
FAQ
Which platforms does scrb target?
TikTok, Instagram Reels, and YouTube Shorts as first-class platforms - each has a different voice calibration and CTA style. The output format is the same (scene list JSON), so you can generate the same script three times for the three platforms if you're cross-posting.
Does scrb produce videos or just scripts?
Scripts only. scrb writes the voiceover + on-screen text + per-scene durations. You shoot or render the actual video in CapCut, DaVinci, Premiere, or your tool of choice. The script is deliberately structured so it maps to a shot list.
Can I request silent (mute-ready) scripts?
Yes - set silent: true and the script returns with empty voiceover fields and longer on-screen text per scene (up to 12 words). Matches the ~60% of viewers who watch without sound, especially on Reels and Shorts.
How does billing work?
Each script counts as one generation regardless of duration (15s, 30s, or 60s). A creator posting 3 scripts per week fits comfortably in Starter ($9.99, 100 gens) with generous headroom.
What tones / styles are supported?
TikTok mode defaults to "native creator" voice (lowercase, self-deprecating, fast). Reels mode defaults to "polished brand" (brand-visible, professional). Shorts mode defaults to "informational" (a creator explaining something to camera). You can override with a tone parameter: casual, professional, humorous, educational.
Can the hook be a specific line I want to test?
Yes - pass a hook field and scrb builds scene 1 around that line, then generates scenes 2+ that flow from it. Useful for A/B-testing a specific hook across variations of scene 2-5.
How to plug scrb into your workflow
- Web app: pick "video script" mode, choose platform + duration + topic. Copy the scene JSON or the human-readable shot list.
- REST API:
POST /api/v1/generatewith output kind video_script_tiktok / video_script_reels / video_script_shorts. - Creator pipeline: schedule weekly topic drops via a Sheets or Notion DB, hit scrb's API on each row, get scripts ready for the shoot day.
- Agency workflow: bulk CSV of topics for a DTC client, generate the scene lists, hand to the editor.
Pricing
- Free: 5 scripts / mo.
- Starter ($9.99/mo): 100 scripts = creators posting daily across 2 platforms.
- Pro ($24.99/mo): 500 scripts + bulk CSV + REST API = agencies with 3-5 creator clients.
- Business ($49.99/mo): 3,000 scripts = multi-brand video teams.