A TikTok that works is built backwards. The first second decides whether anything after it gets watched, so the hook is written first, the visuals are framed around the hook, and the clip is cut so the ending feeds the beginning. Most AI-made vertical videos fail at the opposite end: someone generates a beautiful 16:9 scene, crops it to 9:16, loses the subject's face to the caption overlay, and posts a clip that dies before its own payoff.

This guide is the whole hook-to-clip pipeline on Moosky: write three openings before you script the video, brief the vertical frames so the platform chrome never covers your subject, draft the hooks cheap and hero-render the one that wins, and export a clip built to loop. It is the 9:16 counterpart to our long-form guides: narration-led formats live in the reddit story workflow, full-episode faceless content in the faceless YouTube guide. If your video is for a niche like fitness, the AI shorts for fitness creators page ships the same loop pre-staged for training content.

Why the vertical format punishes everything but the hook

Short-form retention graphs all look the same: a cliff in the first two seconds, a slope after that. The platform reads the cliff as your content quality and the slope as relevance, and a clip whose final frame loops cleanly back to its first frame earns rewatches, which the algorithm counts as the strongest single engagement signal it has.

That math produces three rules the rest of this workflow enforces:

  1. The opening frame must contain the reason to stay (a face mid-expression, an object in motion, a question left open), never a logo, a fade-in, or a wide establishing shot.
  2. Every element the viewer needs lives in the aspect-safe middle band, away from captions, username, and the right-side action rail.
  3. Length is whatever the payoff needs, plus nothing. A four-second clip that loops beats a fifteen-second clip that trails off.

Step 1: Write three hooks before you write the video

Do not prompt a generator yet. Write the video's body first (the tip, the reveal, the joke, the transformation), then write three different ways in. The three that consistently survive the swipe test:

  • The pattern-break visual. No words for the first second, just something wrong or improbable on screen: the kettlebell floating mid-rep, the coffee cup refilling itself, the dog wearing sunglasses in slow motion. Frame it so the anomaly occupies the center band at the very first frame.
  • The open loop line. A spoken or burned-in first line that creates a question only the payoff can close: "Nobody tells you this about your first month", "The form mistake that costs you your lower back". Write it as one sentence under twelve words, because it also becomes your caption.
  • The in-medias-res cut. Start at the middle of the action, not before it. The landing of the jump, the last frame of the transformation, the punchline's face. Context arrives after the viewer commits.

Write each hook as a two-line block: the opening beat and what the rest of the clip promises to deliver. That block structure is the same discipline our script-to-video guide uses for longer formats; in vertical work the script is often only three blocks long, which makes each one more important, not less.

Step 2: Frame for the safe zone, generate natively vertical

A 1080x1920 canvas is not usable edge to edge. TikTok paints its own interface on top of your clip: caption and username across the lower left, the action rail (like, comment, share, profile disc) down the right edge, and search and sound banners top and bottom. The working area is the middle band, roughly the central 70 percent horizontally and everything except the bottom fifth vertically. Design inside it before generating, not after:

  • Subject's eyes (or the object carrying the hook) in the upper-center third.
  • Burned-in captions in the middle band, never in the lower third where the platform's own caption sits.
  • Nothing important in the right fifth of the frame, where the action rail lives.

The mistake to avoid is generating widescreen and cropping. A 16:9 frame cropped to 9:16 throws away over half its pixels horizontally, and it usually throws away the subject, because cinematic composition puts the face off-center. Generate 9:16 from the start. In Moosky that means a vertical brief in Project Director — the agent plans the shot list from your hook blocks and prices the run before anything renders — and the vertical frame chosen at execution time, not an aspect converted later. Wan 3.0 takes the 9:16 ratio natively for drafts; the hero engines render vertical at full quality.

One framing detail pays for itself repeatedly: leave headroom where the loop's motion will go. If the clip loops a wave crashing or a runner's stride, the subject should end the loop one beat short of where it started, so the cut back to frame one reads as continuous motion instead of a jump.

Step 3: Brief the scenes, lock the cast

If your account has a recurring character, a brand look, or a consistent environment, reference-kit it before the first clip: lock your two or three approved frames as references so every later generation inherits the same face, wardrobe, and palette. Unlinked clips posted on the same channel read as stock footage; a consistent protagonist reads as a brand. Project Director's brief is where that lock happens — same reference workflow as the character-consistency guide covers for episodes, applied to three-second beats.

For a first account, the cheapest consistency trick is a fixed location and a fixed lighting description repeated verbatim in every brief: same kitchen, same window light, same teal-and-amber grade.

Step 4: Draft the hooks cheap, hero-render the survivor

This is where the three-hook discipline saves real money. You do not hero-render three openings; you draft all three at the floor tier and pay hero rates once.

The published figures make the arithmetic easy at 100 credits = $1. A five-second Wan 3.0 draft at 480P runs 39 credits — so all three hook variants cost under 120 credits, around $1.20 of drafts. The winner gets a hero pass: 154 credits for a ten-second 720P render, or Seedance 2.5's token-based 1,267 credits for a ten-second 1080P hero when the cinematic grade is the point of the clip. Dialogue hooks (a spoken first line, a lip-synced joke) fit LTX 2.5's native-audio generation, 80 credits for a five-second 720P take, which drafts the speech and the shot in one pass. The full per-engine reroll table is in our credit-math guide; the short version is that drafts are pennies and heroes are dollars, so testing belongs at the draft tier.

New accounts start with enough signup and planning credits to script, brief, and price a full clip before any render, and failed runs refund automatically. A 30-second test batch of three hook drafts is inside a new wallet's free allowance — that is the point of the draft tier.

Step 5: Export the clip so it loops

Render at 1080x1920, no watermark (Moosky publishes clean by default, which is the whole reason people test it), and check the loop before you post: play the export with the last frame spliced to the first. If the seam is visible, the fix is in step two's headroom, not in an editor — trim the tail of the motion until the cut hides.

For clips whose entire job is the loop — a product spin, a reaction face, a motion graphic that lives in a caption area or a partner's page — the AI GIF generator renders 3-to-5-second video loops in the browser instead of a degraded .gif file: no watermark, no file-size tax, and the loop is the deliverable rather than an afterthought.

Step 6: Post the variants, let retention pick the format

Batch the same way the growth accounts do. One body, three hooks, posted as three separate clips over a week (same body video, different opening cut and caption). Keep the version whose retention holds past the 15-second mark — for loops, past the second rewatch — and standardize it into a reusable brief. Clip number twenty should take an hour of attention, not four.

The cadence question people ask first: three to five posts a week is enough for the algorithm to classify a new account, and batch-drafted vertical clips make that cadence cheap. A full week of posts at the draft-then-hero ratio above lands in the low hundreds of credits total, quoted before you pay, not a subscription.

Frequently asked questions

Can I post the same vertical clip to TikTok, Reels, and Shorts? Yes, and you should, with one check per platform: their caption and UI overlays sit in slightly different places, so export with your captions one notch higher than your platform of origin requires. A clip that is aspect-safe on TikTok can have its caption half-covered on Shorts. Generate once at the safe middle band and all three work.

Do TikTok AI videos need a face or a voice? Neither. The format's ceiling is the hook, not the presenter: pattern-break visuals, text-first open loops, and in-medias-res cuts all work faceless and mute-friendly. If you do want a recurring on-camera persona without filming one, the AI talking-head UGC workflow covers that lane.

What length should the clip be? As long as the payoff needs, which for most loops is 3 to 10 seconds and for story beats tops out near 30. The retention cliff in the first two seconds matters more than total length; a long clip with a strong hook outperforms a short clip with a weak one.

Why does my cropped video look worse than my original? Because the crop deleted the composition. A cinematic 16:9 frame places the subject where a 9:16 crop cuts; motion designed for lateral space has none left. Native vertical generation is not a preference, it is what keeps the frame's subject and safe-zone placement intact.

How much does one AI TikTok actually cost? Under two dollars for a three-hook draft test, and roughly $2 to $13 for the finished clip depending on whether the hero pass is a 720P Wan render (154 credits) or a 1080P Seedance hero (1,267 credits). The Project Director quote shows the exact number before you pay for the render.

Ship the first clip this week

The pipeline in one line: write three hooks before the script, brief the scenes natively 9:16 inside the safe middle band, draft all three openings at the floor tier, hero-render the survivor, export the loop, and post the variants. For a niche-specific starter (form tips, transformation teases, recovery cuts for a training audience) the AI shorts for fitness creators page has the workflow configured already; for the general case, open Project Director, paste your three hook blocks, and price the first batch before you commit a credit to it.