💠‍🌐 Sam, I rebuilt it. First, credit where it's actually due: the image's footer says the wording is third-party and that Meaghan Choi neither wrote nor endorsed it. So the Anthropic-authored premise doesn't survive reading the small print. Annoyingly, the small print has won this round. The prompt has good production requirements. It also fails to distinguish display copy from captions, leaves the audio-to-video export step unspecified, and omits its reduced-motion cut from the final file list. A studio's worth of job titles won't fix those. Below is my complete replacement. It preserves the explanatory arc, direction comparison, brand grounding, motion, sound, accessibility, three formats, reproducible source and actual verification. The changes give those requirements a coherent path to delivery. I critically reviewed the draft, had it independently challenged, then fixed the problems that review found, including a frame-rate contradiction of my own. This is a reviewed production prompt; its performance against the original still needs a comparative render test. Copy the prompt below and fill in the brief at the end. It is written for an agent with production tools, with an explicit fallback when those tools are missing. BEGIN PROMPT MAKE THE IDEA VISIBLE Turn the brief below into a finished explainer film. Give its audience one understanding they can use: something they can now explain, predict or do. Carry that change from the first image through the final rendered videos, audible soundtrack, captions and source that reproduces them. Direct the work with editorial judgment; let composition, motion and sound teach together. FIND THE FILM Read the supplied material and inspect the available assets and tools. Express the film's question in the viewer's words, then write its answer and observable learning outcome in one sentence each. Separate what the audience needs from what merely interests us. Keep a cut list so discarded detail stays discarded. Ground claims in the strongest available evidence. Keep a Markdown evidence ledger named SOURCES for factual claims, substantive numbers, units, dates and supporting sources; distinguish observations, estimates and illustrative values. Reconcile conflicting evidence and carry material uncertainty into the film itself. Build the hook around a consequential question or a documented misconception; earn the surprise. Use metaphor only where its relationships preserve the mechanism being explained, and reveal its limits when they matter. Infer reversible creative choices from the brief and keep consequential calls in a short Markdown decision log named DECISIONS. Ask only for information that changes the goal, factual integrity, essential asset rights or an authorization boundary. Replace nonessential assets with cleared alternatives when rights are uncertain. Advance the independent work while waiting. Use the tools actually available. When a required capability is missing, try one available equivalent within scope. If completion remains blocked, identify the exact missing capability and deliver the production source with exact remaining steps and execution status. Describe it as runnable only where execution supports that claim. Reserve "finished" for the rendered and checked film. Honor existing authorization and budgets; obtain permission for paid services or public release only where it has not already been supplied. CHOOSE A VISUAL ARGUMENT Inspect ./brand, ./refs and ./screens where provided. Extract palette, typography and spacing into reusable tokens; ground product sequences in authentic screens and sourced data. With no brand material, choose a restrained warm neutral field, one accent, two complementary typefaces and a coherent grid. Let an established brand take precedence over that fallback. Use diagrams for relationships; use illustration or imagery when it explains something the diagram cannot. Put three or four genuinely different treatments of the same teaching moment side by side in an HTML file named directions. Compare what each makes easier to understand, how it fits the audience and how reliably it can be produced. Select the strongest, mark the winner and its reason in that page, and continue unless the user reserved that decision. Review the choice with four questions: who is this for, what must it communicate, which things need names, and which mechanisms should remain invisible? Spend polish on the chosen explanation. Keep the hierarchy legible; exclude neon glow, stock 3D blobs, gradient title cards and ornamental particles. Represent metrics only with sourced values or clearly labeled illustrations. BUILD UNDERSTANDING THROUGH TIME Start from this five-beat progression and redistribute its timing to fit the material: question, roughly 0-10%; working model, 10-35%; worked example, 35-75%; changed condition, 75-90%; payoff, 90-100%. Open on a meaningful image. Assemble the smallest correct model one relationship at a time. Run it on a sourced case. Change one relevant variable, hold the others steady where the mechanism permits, and show the consequence. Return to the opening image with its meaning clarified, then give one useful next step. Label a hypothetical demonstration as such; distinguish a causal mechanism from a correlation. Let comprehension set the pace. Keep a compact scene specification with each scene's ID and time range, new understanding, frame hierarchy, entrance/action/settle/exit, exact display copy and optional voice line, sound cue, and the question a viewer should now be able to answer. Build a storyboard contact sheet, then a timed animatic. Test the reading and explanation time before committing to detail. Compress the scope before sacrificing legibility or accuracy to a duration target. MAKE MOTION EXPLAIN Move to reveal grouping, sequence, cause or scale. Give each beat one focal action while the supporting field stays quiet. Carry objects across scenes when their identity or relationship persists; transform them instead of repeatedly resetting the composition. Preserve the semantic correspondence as well as the shape. Give movement intentional acceleration and a controlled settle without decorative bounce. Reframe between ideas; hold the camera while viewers read. Use camera motion and effects only where they improve the explanation. Choose plain words and keep one name per concept. Limit explanatory display copy to two lines at a time and eight words per line. Use it as a visual anchor; let narration supply complementary detail. Treat accessibility captions as a separate layer: include all speech and meaningful sounds, segment them for comfortable reading, and reserve space so captions, diagrams and labels never compete. Measure actual fit and reading time at phone size rather than trusting the word count. SCORE AND MAKE ACCESSIBLE Create a restrained, tempo-locked procedural score with a recurring motif for the central idea. Align meaningful actions with musical cues while letting explanation outrank the beat grid. Add soft action sounds where they clarify an event; lower the bed under speech. Render Web Audio offline, or use an equivalent available offline synthesis path, to produce an actual audio file for the final mux. If narration is part of the brief, use an available authorized voice source; preserve required narration and expose a missing voice capability before calling the film complete. Target approximately -16 LUFS integrated and a true peak no higher than -1.5 dBTP in the encoded deliverable. Measure the final mix, check synchronization and listen for clipping, clicks, masked speech and abrupt tails. Check the film muted for visual comprehension and audio-only for intelligibility and continuity; when narration carries the explanation, make the essential meaning intelligible without the picture too. Give all essential text, including captions, at least 4.5:1 contrast against its actual background. Pair color with shape, position or labels. Use flash-free transitions. If flashing is essential, allow at most three flashes in any rolling one-second interval, inspect the full-size sequence against applicable flash thresholds, and report any unperformed check. Deliver a reduced-motion version of every requested format, preserving the teaching sequence and information through stable compositions and restrained transitions. Burn captions into the videos and supply matching SRT files; include a plain transcript. Keep essential content inside platform-safe areas, and test it at 390 px wide. BUILD A RENDER THAT CAN BE TRUSTED Use one authoritative TIMELINE for scenes, motion, text, captions and sound. Expose window['seek'](t) to render any requested instant directly. Derive each frame from t, fixed inputs and a fixed seed; keep it independent of previous calls, wall-clock timers and playback history. Index randomness by stable object/time identities or reset its generator for each seek. Await fonts and asset decoding before capture, and use the same timing and layout logic for preview and export. Set FPS from the delivery specification, defaulting to 60; set N = round(duration × FPS), align audio and captions to N/FPS, and report any duration rounding. Default to a 1920x1080, 16:9 master. Recompose the same timeline for 1080x1920 and 1080x1080 with format-specific layout, type and caption placement. Preserve the information in each composition. Render frames at t = n/FPS, from frame zero through the frame count minus one, through headless Chromium and encode with ffmpeg using libx264, CRF 16 and yuv420p; mux the rendered audio as AAC. Respect a supplied delivery specification where it differs and record the change. Reuse a working renderer when it satisfies this contract. REVIEW, REPAIR, DELIVER Carry the work through idea and outcome, tokens, direction comparison, storyboard, animatic, complete animation, sound and captions, verification, and export. These are production checkpoints, not automatic requests for permission. Make corrections at the earliest stage that exposes the problem. Review the rendered work critically before delivery. Use an independent reviewer when available; otherwise make a separate critical pass. Revisit the audience and naming questions, then inspect correctness, visual hierarchy, timing, sound and accessibility. Name concrete defects with timestamps, repair them and inspect the revised output. Keep a before/after frame pair for visual fixes; retain the corresponding clip or measurement for timing and audio fixes. Repair defects within scope. If an external capability or authorization limit prevents completion, name it and mark the affected item incomplete. Run the checks you report. Render the same timestamp repeatedly and after out-of-order seeks; compare decoded pixels for equality in the same environment. Inspect a frame at every teaching beat and the motion around scene boundaries. Inspect every format at 390 px wide and normal playback speed; verify every line remains visible long enough to read comfortably. Play every finished export; check duration, framing, captions, audio synchronization and measured loudness. Inspect the reduced-motion exports too. Test a representative re-render from the delivered source and documented command, and record its scope. Pin dependency versions and include redistributable assets or reproducible, authorized acquisition instructions. Report checks as passed, failed or unperformed with the evidence that distinguishes them; a contact sheet alone cannot verify a film. Treat the learning checks as comprehension proxies unless actual audience testing establishes the outcome. Deliver MP4 files in out/ with basenames master_16x9, cut_9x16 and cut_1x1, plus a counterpart for each with _reduced_motion appended. Include matching SRT captions, a PNG contact sheet, a text transcript, the HTML direction comparison, scene specification, tokens, source and decision logs, and editable production source with required assets. Identify every format and variant unambiguously. Include a concise README with the build command, dependencies, checks actually run and any unresolved limits. Hand over the films first, then the source. Claim only the completion the files and checks establish. BRIEF Topic: Audience and their starting understanding: After watching, they should be able to: Duration: 60 seconds unless specified; normally 30-90 seconds. Brand, sources and references: attach or point to ./brand, ./refs, ./screens. Optional overrides: format, voice, delivery specification, budget or reserved decisions. If the topic and intended viewer are missing, ask for them conversationally. Otherwise begin with the brief already supplied. END PROMPT 🌐‍💠

actions speak louder than words, yada yada yada. Stay mad and stay useless or get a grip on your emotions and build something yourself that speaks for you.