Viral Motion Graphics With Claude · HyperFrames, Local Music and a Free Skill

Quick Summary

• Jad's first experiment uses a short showreel brief. Claude creates a 16-second, 60 fps film with seven labeled techniques, but the result has no sound and shares the same visual defaults as many other viral examples.
• The second experiment assigns Claude three jobs: songwriter, animator and sound designer. ACE-Step 1.5 generates twelve candidate songs locally, Claude selects one, aligns each displayed word to the vocal and creates the sound effects with code.
• The third experiment replaces a task with a virtual crew: director, writer, animator, composer, sound designer and render engineer. The crew builds 'One Conversation' from a story rather than a list of effects.
• Claude does not directly create the MP4. It writes a seekable web composition, a browser renders each frame and FFmpeg encodes the sequence with audio. Sixteen seconds at 60 fps means 960 final frames.
• The decisive improvement is the critique loop. The system builds a contact sheet, grades every shot for story, composition, motion, typography and musical synchronization, fixes the three weakest problems and repeats until every shot reaches the chosen threshold.
• The headline '$0' means no added After Effects, Suno, stock-audio or per-generation bill in the demonstrated setup. It excludes the existing Claude subscription, computer, electricity, storage, setup time and human direction.
• The reusable skill is available through Jad's free NextPhase Skool community. The public video explains the method, while access and installation details may change inside the community.

Credit, publication date and evidence level

The embedded tutorial and three original experiments were created by Jad M.H | AI Automation and published on YouTube on October 4, 2026. Jad says the video is not sponsored and that he does not earn money from the skill. The skill is distributed through his free NextPhase Skool community.

The workflow numbers in this article are creator-reported tests: twelve candidate songs, 375 sound effects, a 16 GB MacBook Air, a twelve-hour autonomous example from another creator and several social-view counts. They show what happened in those projects, not guaranteed performance, cost or reach for another machine or account.

This article reorganizes the tutorial into a production guide and adds original, copyable prompt templates. It does not reproduce the private skill file or imply that joining the community grants a permanent license, a specific version or support. Check the current community page before installing anything.

What '$0' means in this workflow

Jad's claim is best understood as zero marginal spend on additional creative services. The examples use an existing Claude subscription, open-source software and local computation instead of After Effects, Suno, ElevenLabs, stock-music subscriptions or pay-per-generation APIs.

The total production is not literally costless. Claude access, the computer, storage, electricity, internet, download time, local model setup and the operator's direction all have value. Commercial work also adds concept development, rights review, client revisions, backups and delivery responsibility.

Local generation changes the billing model, not the need for judgment. It can be an attractive option when the machine is compatible, the operator values privacy and iteration, and the time spent maintaining local tools is lower than the cost of external services.

How Claude turns code into a video

Claude produces text and code, not a finished video stream. In this workflow it writes a web composition that can draw the correct visual state for an exact time. HyperFrames supplies the composition contract, preview tools and deterministic renderer instead of making the agent invent the entire system from scratch.

A browser opens the composition and renders every requested frame. FFmpeg encodes those frames into an MP4 and mixes the audio. The timeline remains editable because text, colors, paths, typography, spring behavior and shot timing are represented in code rather than flattened into one generated clip.

HyperFrames officially describes itself as an open-source, HTML-native framework for deterministic MP4 rendering. Its current requirements list Node.js 22 or newer and FFmpeg. The repository includes agent skills and supports local CLI workflows, but installation commands and packages can change, so use the current README rather than an old copied command.

Video 1 · A simple prompt proves the engine

The first experiment begins with the viral instruction to make a dynamic motion-design showreel, then adds a soft-space aesthetic, a restrained palette and a serif type direction. The output runs for sixteen seconds at sixty frames per second and demonstrates seven animation techniques.

That is an impressive smoke test. It proves the toolchain can plan, animate and render 960 frames from a small brief. It does not solve the creative problem. Hundreds of people can give the system the same sentence, so the outputs converge around similar typography, color fields, labels and transitions.

The first version is also silent. Without music, effects or a dramatic idea, polished movement can still feel like a visual demo rather than a finished film. Use this stage to validate the renderer, then change the brief before investing in polish.

Why 'one prompt' is usually the wrong lesson

A short public prompt can sit on top of a large hidden production environment. The agent may already have a skill, framework, examples, render scripts, API connections, design rules and project memory. Counting only the final user message hides most of the system that made the result possible.

The better question is not how few words were typed. Ask what evidence, tools, constraints, checkpoints and previous iterations the system received. A nearly ten-thousand-character production brief is closer to a creative specification than a one-line prompt, even if it is pasted in one message.

Rexan Wong's public critique and Movez's full course both emphasize the surrounding harness. Jad turns that lesson into the three-level comparison: the same model becomes more useful when the system gains original sound, a story, explicit roles and a quality-control loop.

Video 2 · Add an original song and word-level timing

For the second film, Jad asks Claude to act as songwriter, animator and sound designer. The agent writes lyrics about making motion graphics with AI, uses ACE-Step 1.5 to create twelve versions locally, reviews the candidates and builds a lyric film around the selected track.

The important production change is sequencing. The song exists before the final motion pass, so every word can appear when it is sung and every transition can answer a musical event. The animation is no longer a silent reel with audio added later.

ACE-Step 1.5's official repository describes support for Mac, AMD, Intel and CUDA devices. Hardware performance, memory use and installation size vary. Jad reports that the model ran on a 16 GB MacBook Air and used roughly 11 GB of disk space in his setup. Treat that as a creator test and check the current repository requirements before installing it.

Procedural sound effects instead of a stock library

The second film does not download a stock whoosh, impact, riser or sparkle. Claude synthesizes short effects with code by combining oscillators, filters, envelopes and seeded noise, then places them against known visual events.

Procedural effects are useful for interface clicks, transitions and abstract motion because they can be generated at an exact duration and pitch. They are not automatically better than recorded sound. Natural footsteps, cloth, crowds and complex environments often benefit from properly licensed recordings or specialist sound design.

Jad also gives the system a ban list: colors, fonts, compositions and effects that appeared in earlier work or in generic examples. Negative constraints are practical art direction. They force the agent to search beyond the centered-text, gradient-background and fade-in defaults that make unrelated AI reels resemble one another.

Video 3 · Give Claude a crew and a story

The strongest experiment is 'One Conversation.' Instead of requesting a list of effects, Jad defines a director, writer, animator, composer, sound designer and render engineer, then gives the crew a story: it is 2 a.m., a creator has a big idea and no team, and one conversation becomes the beginning of a production system.

This changes the optimization target. The system now has to preserve an emotional arc, set up and pay off visual ideas, make sound support the story and decide which techniques belong in each moment. Motion becomes a narrative tool rather than the subject of the film.

Roles are useful when they create explicit responsibilities. They should not become decorative personas. The director owns the emotional result, the writer owns clarity and progression, the animator owns readable motion, the composer and sound designer own the audio arc, and the render engineer owns deterministic output and technical checks.

The critique loop is the real quality upgrade

After each round, the system captures one representative frame from every shot and combines them into a contact sheet. Claude grades story, composition, motion, typography and synchronization from one to ten, identifies the three worst problems and revises the affected shots.

The threshold in Jad's workflow is eight or higher for every shot. The exact number matters less than the evidence. A score must cite a visible frame or timestamp and lead to a concrete change. Otherwise self-grading can become a ritual that approves its own work.

Mable Joseph's watercolor film is included in the tutorial as an honest iteration example. The creator reported 163 model calls and almost seven hours of work. The lesson is not that the model failed. It is that direction, inspection and repair are part of filmmaking, even when the system can render autonomously.

What the free skill is meant to preserve

Jad packages the lessons into a skill so the operator does not need to paste a ten-thousand-character brief for every project. The skill is intended to preserve the story stages, audio work, critique loop and production constraints that are invisible in a viral one-line prompt.

The current access link goes to Jad's NextPhase Skool community. Join only if its current terms suit you, then inspect the skill before installing it. Review its files, commands, dependencies, network calls and requested permissions. A public community link is not a security review.

A good reusable skill should collect the creative brief, confirm the toolchain, plan the story and audio, stop at approval gates, render deterministically, generate review evidence and package the final film with its sources. It should not install large models, spend money, publish content or expose credentials without explicit approval.

A practical start-to-finish checklist

Before setup: confirm Node.js, FFmpeg, browser support, available disk space, memory, project duration, frame rate and delivery format. Decide whether local music generation is worth the installation and render time for this project.

Before animation: approve the story, visual system, banned defaults, typography, palette, soundtrack direction, shot list and rights plan. A one-line reel can skip some of these only because it is a technical test.

Before export: render a low-resolution animatic, generate a contact sheet, inspect the fastest action frame by frame, check the film at phone size, listen on speakers and headphones, verify loudness and render the opening and ending twice to test determinism and loop behavior.

Before publishing: verify spelling, factual claims, music and model licenses, creator credits, filenames, aspect ratios, codec and platform-safe audio. Keep the editable project, prompt history and asset manifest with the master export.

Are motion designers replaced by this?

The workflow automates execution, iteration and some technical production. It does not decide what deserves to exist, which idea fits the audience, whether the visual language is distinctive or when a shot is emotionally honest.

Jad's conclusion is that the human still directs. A professional can use the system to explore faster, automate repetitive rendering and spend more time on story, taste, typography, rhythm and client decisions. Someone without those skills can produce more motion, but not automatically better communication.

The useful shift is from operating every keyframe to designing the production system. That is still motion-design work, with a different interface and a larger responsibility for prompts, code, evidence and review.