MUDI video production (create by GPT)
For a new video brief, open PROMPT.md for the copyable request and script-approval follow-up.
Intake and references
- Start from the supplied footage, current campaign brief, product facts, approved copy, reference videos and guides. Locate source files by name in the supplied folder. In the original package,
pinkis Pink Bunny,greyis Moon Egg, and1105 (3) 成品.movis the finished reference. The two guides aremudi_video_ads_editing_skill_v1.mdandmudi_black_friday_video_ads_skill_v1.md. Read the latest supplied versions even when the filenames still end inv1. - Apply the user's current instructions and latest approved guide decisions. Use older footage and drafts to study technique. Verify product facts from supplied materials or the Mudi storefront where needed, and confirm an offer against the current campaign before using it. Leave unverified details out.
- Watch and listen to the reference. Record the time to first bead movement, how long the opening action holds, the change in shot lengths, where the words cue new visuals, VO pace and tone, caption density and placement, the music and contact sound, and how the CTA ends. Transfer the pattern to the new SKU and footage. If the reference cannot be played, state which parts of the analysis came from the guide.
- Review the raw clips and audible source tracks. Log usable source filename, in/out, visible SKU, action, angle, and contact sound. Look for front-facing bead rolling, palm or held detail, bead finish, full product, and wrist wear. Mark blur, dead time, and mixed-SKU frames. Use accurate footage for each named product.
1. Write and approve the copy
- Pick one clear purchase reason per version, such as tactile play, a moment between calls, a travel companion, gifting, styling, or a visible material detail. For a collection, connect the SKUs through one reason to choose and give each one a distinctive detail. Write in natural spoken English. Use Mudi and Mudi’s in customer-facing copy.
- Build one continuous read in five beats: hook, exact product name and type, verified material or finish plus a concrete bead action, one recognizable use or gift scene, and a direct purchase CTA. Describe what the fingers and beads actually do. Distinguish RuaRua rolling from Quiet Path 108's bead-by-bead handling. Keep relaxation language experiential and claims factual.
- Aim for about 110 spoken words. Match the exact total of the selected guide reference. Choose the closest angle when the user has not supplied one. The five reference totals are 107 (gift and keep), 111 (self-gift and styling), 111 (gift for a mountain lover), 109 (travel companion), and 113 (gift for someone always busy). Count only words spoken in the final script. Spell out figures for delivery, count contractions and possessives as one word, and count standard hyphenated compounds as one word.
- For a confirmed Black Friday campaign, name Black Friday in sentence one and give the exact SKU and confirmed discount by sentence two. Use “Early” and state the offer scope only when confirmed. Connect the offer to the reason to buy. For a Thanksgiving or evergreen campaign, use the actual campaign language and terms.
- For a full brief, plan two distinct versions per footage set unless the user requests a different count. Give each version its own hook, reason to buy, and scene or payoff. Reuse strong footage across versions with meaningful changes to shot order or pacing and the music treatment.
- Send the complete readable English script with its reference and word count for confirmation. Stop at copy if that is the requested stage. Use an already approved script as supplied. Keep the copy editable in CapCut, generate the agreed voice, and measure it at original speed. Trim excess silence or revise and reconfirm copy if duration requires it.
2. Edit and review in CapCut
- Use the editable CapCut project the user identifies, or start a clearly named draft for the new ad. Keep a recoverable copy of a prior approved cut when revising. Use the available CapCut controls and verify the actual timeline after imports. Use FCPXML only when the available editor supports it, and check linked media, cut order, source trims, and total duration. Use timed SRT or native editable captions based on the final VO waveform. A file-picker selection alone establishes only selection.
- Build 9:16 first, targeting 1080 × 1920 and around 35 seconds. Aim for 32–38 seconds, with the standard cut under 40 seconds, unless the brief overrides the format or duration. For a requested 1080 × 1080 derivative, reframe every shot and verify product visibility and caption safe area again.
- Show clear bead rolling in the first 0–3 seconds, preferably on the first frame. Start a single-SKU ad with one uninterrupted front-facing take around 5–6 seconds, with synchronized bead sound. Move into clean, distinct display shots around 3 seconds each. Use one complete wrist shot around the wearing line. Add at most one supplementary roll for an actual spoken demonstration or a specific closing beat. Close on a distinct product shot through the full CTA.
- For a collection, consider a split-screen opener with independently moving beads in every panel. Match every named SKU, color and action to the corresponding shot. For all cuts, change shots on movement or a change in spoken information, favor clean cuts, and check the source-range log for overlap or visual repetition. Mask or trim out other SKUs in a single-product ad.
- Use CapCut TTS with the approved voice, usually Natasha or Joanna / Joanne, at original speed and pitch. Keep video, caption text, VO, music and bead-contact audio independently editable. Preserve the actual contact sound wherever fingers touch, press, rub or roll the beads. Edit the contact audio with its picture and let it end with the action. Keep the VO intelligible, choose warm music with movement and changing texture, and duck it locally around VO or tactile details. Listen for clipping, doubling, abrupt changes and missing music. If a rolling shot needs a modest speed change, match and check its original sound. Preserve the narration at its recorded speed.
- Use only spoken captions. Place any approved discount in its corresponding caption. Split on natural phrase boundaries, usually 3–5 English words per cue, and keep cues close to the VO with a visually stable baseline. Use centered white Newsreader16pt-Rg, size 7 in CapCut. Under Captions → Text → Basic → Preset style, use the first-row fourth tile from the left, counting the no-style tile. Match the white Aa with dark lower-right shadow if the UI layout differs. Apply to all, recheck style, and keep each cue on one line. Raise the baseline clear of Reels controls while keeping the moving hand and bracelet visible. Reframe first and move captions only at a natural boundary when needed. Check white text on light fabric.
- Watch and listen to the whole timeline at normal speed. Verify the opening action, correct SKU, distinct follow-up shots, stable readable captions, no separate offer overlay, natural VO, audible separate music, contact sound in sync, and a complete ending. Use the reference's tight delivery as a benchmark. Its 33.8-second, 107-word version had 25 captions and mostly contiguous cues. Retain brief breaths or sound features and keep accidental VO dead air under two seconds.
- For a review-only request, leave the CapCut timeline open at the start with tracks visible and respect the user's directions about saving and export. For delivery, save the agreed editable version, export the authorized MP4 and play it back. Check dimensions, duration, first and last frames, captions, and audible mix against the timeline. Report the project, timeline, file location, and what was verified. A confirmed upload or publication is a separate deliverable when explicitly requested.
- If CapCut, permission, source files, a voice, or the required font is unavailable, continue the independent work. Provide the applicable script, timecoded selects, captions aligned to available narration, or track plan. Identify the precise step still blocked. Report observable timeline changes and verified renders as separate states.
Pink Bunny review lesson
The recorded test used a bead-rolling opener, palm and wearing shots, English VO and matching captions. The review identified slow altered VO, missing audible music, light-background caption contrast, a separate promotional overlay, and slow later display shots. For similar revisions, verify audible music both on its own track and during playback, restore original-speed narration, make the captions single-line and high enough for Reels, keep the first roll uninterrupted and tighten the following cuts. Treat the old suggestion of about four seconds per later shot as feedback on that test. Start around three seconds under the current guide. Inspect the timeline at the user's requested review point before export.