Section 6: Four Full Builds — a Faceless Hook in 20 Minutes, a '$300,000 Ad for Under $30', an Animated Music Video, and a Viral AI-Clone Action Sequence

Three pillars: idea, quality, packaging; AI enhances judgment, it does not replace it (Wendover Productions as the model). Script: ChatGPT iterated three times ('write a one minute hook' -> 'diving right into the moment' -> 'under 500 characters'). Narration: ElevenLabs TTS on PromptEdit. Music: Suno, 'no vocals, just a slow pulsing background track... suspenseful', two generations to choose from. Resolve: trim narration pauses, duplicate/reposition the music so the crescendo hits the most intense line, 6-frame crossfade at the music edit. Visuals: ChatGPT drafts a Seedance 2.0 prompt in 15-second segments with scene timestamps matching the voiceover; delete Seedance's baked-in audio and lay the clip over the narration; a too-short opening drone shot is screenshotted and regenerated longer in Kling; split and re-time. SFX from ElevenLabs' SFX tab in two buckets - seen (volcano, footsteps, cups rattling, dog) and felt (riser, heartbeat, impact) - with 'choose specific duration' checked, lengthened, volume down so they 'enhance... not distract.' Title card from the Content Creator Templates. Total ~20 minutes.

Recreating a Dr. Squatch-style spot (the brand's '$5 million to over $100 million in a year' is cited). Outline and script in ChatGPT. Casting: a character-prompt template on Higgsfield's Nano Banana Pro - the vague-vs-template comparison. Storyboard: paste the script into a Google Doc, drag only FINAL images under their lines (resize freely; quality is unaffected), later File > Download > Web Page to get a full-res image folder for Higgsfield. A scene-generation template (camera framing, subject/props, lighting/tone, background action, mood) is a free download. Video on Higgsfield: Veo 3.1 for dialogue and tone (but it drifts off a locked tripod and 'refuses to generate anyone appearing under 18' - the boy goes to Kling 2.6); Kling 2.6 for product and reaction shots, more detail retention, fewer credits; first/last frame on Veo for one-second magical transformations (weak soap to Dr. Squatch bar; young to aged dad). Voices: ElevenLabs Explore filtered by language/accent/age, '+' saves an Australian narrator and a deep dad voice; V3 model with bracketed director cues ('[depressed]', '[excited]'), 'way more expressive and human' than V2, two takes per request; Voice Changer re-voices Veo's baked-in dialogue to the saved voice, then frame-nudged to the waveform in Resolve. A long real-time Resolve assembly, layering tracks so a clip's own audio never covers the narration.

Never regenerate from scratch: reuse an approved output as the base and layer changes. Product swap: 'Replace the weak soap with the bar of Dr. Squatch soap' with a real product photo uploaded. Face swap: 'swap the face in image one with the face from image two, maintaining realism and matching the original lighting and expression.' Before/after: 'keep everything in the scene exactly the same, but replace the person... with the person from the second image.' The clone pipeline formalizes it: outfit transform -> environment transform -> face swap from multi-angle, multi-expression headshots, at 4K with four variations per pass on fal.ai, the same swap prompt reapplied to every scene 'until I got it perfect' - 'the secret sauce to making this look professional instead of AI slop.' Sam blends references image-to-image (a rooftop plus a street) for details he cannot phrase.

Early motion prompts 'absolutely sucked' until two changes: describe the exact start and end state of the motion, and use domain terminology - parkour ('kong vault', 'precision jump', 'cat leap', 'safety roll'), cooking ('julienne', 'saute', 'flambe') - because 'AI is trained on professional footage that uses correct terminology.' Kling 2.6 is the presenter's go-to 'including over Google Veo 3' for character detail and multi-object scenes; Cinematic scene stills came from ChatGPT-authored image-to-image prompts per storyboard bullet.

Seven steps for the viral 1M-view effect: bullet-list storyboard (position, angle, mood; stick figures optional); film base footage and multi-angle headshots on an iPhone 16 Pro Max; build the base clone in Nano Banana Pro on fal.ai; generate every scene still via ChatGPT prompts plus face swaps; animate scene to scene in Kling 2.6 with motion prompts; motion-transfer the real footage onto the clone with Wan 2.2 Animate in MOVE mode (explicitly not Replace, which 'only swaps the faces'); generate a Kling transition bridging the transferred clip into the animated sequence; edit in CapCut/Premiere with zoom-into-action cuts that hide face seams, a compound clip ('nest') for one global speed ramp, sound design from the Content Creator Templates plugin, Topaz upscale, 4K export.

Stack: ChatGPT for prompts, Nano Banana Pro for images, Hailuo (Minimax) for animation, Canva for manual tweaks, CapCut to edit; no upscaling - native 4K/1080p. He fed the years-old animatic MP4 to ChatGPT to extract characters and story, correcting misreads through Q&A (an 'angry dog' read as wholesome). One locked Pixar-style master prompt reused and simplified per character; front-view-only character sheets; environments built frame by frame from the sketches with image-to-image blending. Workflow: image gen, video gen and CapCut in parallel. Hailuo preset: 'no exposure change, make video smooth...'; he avoids Hailuo's end-frame-to-start stitching (visible exposure jumps). Failures: a 'walking past flowers' shot saved by mining a usable 2-second sub-clip; a flower pluck and a shooting star that took many attempts. CapCut beat markers exist, but he cut by feel.

DaVinci Resolve: free, one-time pro fee versus CapCut/Adobe subscriptions; trim pauses; waveform sync by nudging the playhead; split and delete generated audio so only the ElevenLabs track survives; push video tracks down so baked dialogue never covers narration; crossfades. CapCut/Premiere: zoom-into-action keyframes timed to mask face seams; compound clip / nest for one global speed ramp (fast through transitions, slow on action beats); the Content Creator Templates plugin for a deeper SFX library in Premiere; CapCut beat markers for music cuts. The 14-Day Filmmaker program and the course's editing section are the deep dives.