Inspect the real reference.
Retrieve the exact post, download its media for private analysis, inspect timed frames and map shot changes. Read the live Makeroll site before writing product claims.
treg · X metadata FFmpegOne chaotic assistant. One impossibly calm coordinator. Four original comedy shorts, built with an agentic pipeline and an open production record. Visual comedy with music and effects; no spoken dialogue.
Elliot creates the problem. Kai makes the correction look effortless. Mara directs the moment. Three original fictional adults; four visual jokes designed to read without dialogue.
Original fictional cast, generated with AI. Elliot, Kai and Mara are independently designed characters, not portrayals of real performers. Each short uses a 12-second 720p generated scene plus a 2-second makeroll.com ending. Production masters were locally upscaled to 1080 × 1920; all players and downloads on this page use lighter 720 × 1280 web copies. The 34-second social cut runs all four actions at 1.5× speed. Prompts and observed deviations are documented below.
The reference is a paired studio performance. Our campaign applies its visual economy to a different cast, four original jokes and a Makeroll finish.
The linked post depicts likenesses of two real public figures performing together, alternating medium and wide two-shots. Dark suits, an orange background and a yellow floor keep attention on their gestures.
Watch the original on X ↗Source: @SashaNris · posted 21 Sep 2026 · retrieved 22 Sep 2026 through treg. Reported views at retrieval: 888,181. Historical snapshot, not a forecast.
Both performers share the frame. Their contrast is understood before exposition is needed. The brief proposed one character’s accidental chaos against another’s precision. Our original cast develops that comic contrast with different character designs and no celebrity impersonation.
One background, one hero prop and consistent screen positions reduce visual noise. The woman initiates or resolves each gag; she has a story function.
The source’s eight shots average about 3.64 seconds. That suggests an editing rhythm, but does not prove eight separate generations. Our short scenes favor a single legible cause-and-effect event.
The post does not establish its model, prompts, editing tools, cost or time. The workflow below is our independent production plan and attempt, not a claim about the source creator’s methods.
The source has audio. Its music, speech, rights and creation method were not identified. Our ending uses an original locally synthesized sonic signature.
Codex coordinates the build. treg connects research and generation. Local finishing makes the branding exact. Every claim has a place to be checked.
Retrieve the exact post, download its media for private analysis, inspect timed frames and map shot changes. Read the live Makeroll site before writing product claims.
treg · X metadata FFmpegDefine setup, mistake, correction and payoff. Give each adult character a role. Lock the setting, wardrobe, prop count and movement before submitting a generation.
Codex · creative directionGenerate four 12-second, 9:16 scenes with original fictional adults through Seedance 2.5 at 720p. Preserve prompts and job handles; inspect the actual returned footage before finishing.
Seedance 2.5 treg CLIUpscale to 1080 × 1920 locally. Composite makeroll.com and an AI-fiction disclosure. Add a two-second end card and an original sonic logo; normalize the final mix.
FFmpeg · H.264 / AACCheck story, identities, hands, prop contact, timing, watermark, ending and sound. Probe metadata, decode every frame, measure loudness, and record checksums.
Visual review ffprobePackage the films with prompts, settled costs, observed limitations and a social compilation. Test local playback and mobile layout. Share the build files; external posting is a separate action.
HTML · local media Browser QAProduction sequence: research → creative development → footage → sound → branding → packaging → export QA. The record includes successful generations, refunded attempts and paid takes that were not selected.
Generation window: about 20 minutes. The first original-cast job was submitted at 09:41:40 UTC and the final selected balloon take completed at 10:01:39 UTC on 22 September 2026. That covers six successful 12-second generations, including two discarded takes. Research, the earlier rejected attempts, editing and QA took additional time; this is not an all-in production-time claim.
Read the exact original-cast generation prompts and compare their intended action with the outputs. Successful requests and export checks are recorded separately from creative intentions.
Brand lettering is deliberately excluded from generated scenes and added in post. “Make the unexpected.” is proposed campaign copy, not an asserted existing Makeroll slogan.
The original brief inspired a contrast between chaos and precision. These original-cast films have no measured audience results. Review the creative choices, limitations and export checks for yourself.
Iteration 1 — Four celebrity-cast Seedance requests failed at the copyright gate; two Hailuo 3 Max trials failed with “new_sensitive”. All six reservations were refunded. Iteration 2 — With authorization to proceed using best assumptions, the production moved to independently designed Elliot, Kai and Mara. This is an explicit casting change, not a claim to have generated the requested celebrities. Iteration 3 — Two paid balloon takes lacked a clear obstruction and were discarded. The third take establishes it. Finishing — The umbrella mix initially exceeded the audio target and was corrected with two-pass normalization.
Creative review ↗ · Web-export checks ↗ · Full production pack and cost ledger available to members or partners only ↗