long-form AI UGC workflow: short form survives small inconsistencies because nobody has time to register them. at 60 seconds they register everything. so the build is different. step 1: the multi-angle reference set. generate 6 to 8 character variants, judge on realism and whether they read as a real person in 2 seconds cold. upscale the winner. then generate 3 more angles off it: - side profile - front - three quarter a 60 second video runs 5 to 8 distinct shots. without angle references, every transition is a chance for drift. step 2: the script. 6 sections, in order: 1/ hook, 0 to 5s 2/ persona signal, 5 to 12s, lived experience language pulled from real reviews 3/ pain amplification, 12 to 25s, name the failed alternatives 4/ mechanism bridge, 25 to 35s, why everything else didn't work 5/ product and proof, 35 to 55s, 2 to 3 proof points with real numbers 6/ CTA, 55s+, calmer than short form, a hard close breaks the intimacy then read the whole thing aloud. a script that makes you stumble produces robotic voiceover, and a stilted phrase compounds across a minute in a way it never does across 5 seconds. step 3: production, shot by shot. each shot gets its angle reference, scene reference, and the full clause stack. shot by shot means you regenerate 1 failed shot, not the whole video. 2.5 changed the math here though. 30 second single pass with multiple connected shots in one render means fewer stitch points to manage. step 4: the 2 longform-specific quality checks. character consistency against previous shots. audio register continuity, since a shot where the voice is suddenly more energetic breaks coherence even if each shot sounds fine alone. step 5: audio. native sync handles short clips well. past 15 seconds, vocal energy flattens and breath patterns get too regular. the fixes: - emotional arc spec on the audio not just the visual - pause direction marked in the script - cleanup pass on the finished track past 60 seconds, generating voice through a dedicated tool and running lip sync mode beats native noticeably. step 6: the continuity pass. lighting, emotional state, and vocal energy across every transition. then watch the finished piece once with sound completely off. 50 to 80% see it muted first. if the visual doesn't earn the tap, the audio never gets heard. 3 to 4 hours per video, $0.15 to $3 in generation. short form is still right for cold discovery. longform converts the warmer audience your reels already pulled in, and almost nobody produces it at quality.