Claude video generation: seven Opus 5.5 workflows compared

Making a video with Claude can involve several different production setups, depending on what you want to show.

Some creators animate text and shapes with code. Others generate people, footage and audio in separate services, then use Claude to coordinate the work. Start with what you want to show, rather than the model name.

Code-only motion graphics do not require external video generation or MCP, a way of connecting Claude to outside tools. Claude Code, Anthropic’s command-line coding tool, can write the animation itself.

This guide starts with that Claude Code-only route, then compares six more setups up to avatar explainers, for seven in total, with links to practical setup instructions. These are proposed production workflows, not results from testing every combination ourselves.

How Claude video generation works, and where Opus 5.5 fits

Claude plans scripts, writes animation code, calls external tools when needed and helps assemble assets. For text and shapes, Claude Code can write code that a web browser draws on its own.

Think of Claude as planning the meal and coordinating the cooking.

For text and shapes, the browser draws the code Claude Code writes, so no external generation service is involved. A dedicated framework such as HyperFrames is optional and can make rendering easier.

A service name flying onto a screen can be made by changing the position and size of text over time. Creating new footage of a person walking down a street calls for a different approach, such as a video generation model.

Both appear under the broad phrase “Claude video generation.” In the comparison below, distinguish the tool making the visuals from the tool making the sound.

Seven ways to make videos with Claude

The seven patterns below differ in what appears on screen and how the assets are made. The first uses Claude Code alone; the other six add external tools. Compare each tool’s role to choose a setup for your needs.

The illustration below shows the six setups that rely on external tools. The Claude Code-only route is not pictured; it appears in the first table row and the next section.

Setup	Main visual material	Audio	Example use	How to make it
Claude Code alone	Text and shapes animated in code	None at first; optional synthesized sound	Short showreels and service introductions	Guide ↓
HyperFrames + Lyria	Animated text, shapes and assets	Lyria music	Service introductions and short promos	Guide →
Remotion + Node.js	Visuals drawn with code	Sound made with code	Numbers and interface explanations	Guide →
GPT Image + ElevenLabs	Animation built from generated images	Separately prepared audio	Illustrated explainers	Guide →
Higgsfield with other tools	Generated people and footage	Clip audio and background music	Live-action-style introductions	Guide →
HeyGen	A talking avatar	Spoken narration	News and explanations	Guide →
ElevenLabs	Images, footage and audio assets	Voice, music and effects	Consolidating asset creation	Guide →

Claude Code alone: motion graphics written in code

Claude Code can write text and shape animation on its own. For example: a 15-second reel where large words switch in rhythm, circles rotate, bar graphs grow and the brand name appears at the end.

Reference video published on X

{{tweet:https://x.com/stephanlivera/status/2103315922098470926}}

The creator says this video was made with Opus 5.5 on Max effort. Original post

Max effort is a setting for how deeply Claude reasons. It is not the Claude Max subscription.

In replies, the creator says no tools or APIs were supplied reply 1, and that no MCP servers or plugins reply 2, Skills reply 3 or Remotion reply 4 were used, just stock Claude Code.

This is the creator’s own account, not an audit of the source code. The rendering library and the path to MP4 are not identified in the posts and replies checked here.

The steps below are our proposed workflow for a similar type of video. They will not reproduce the creator’s code or result.

The idea is simple. Claude Code writes HTML, CSS and JavaScript that draw text and shapes, and a web browser plays them. Redrawing the picture a little at a time creates animation. MDN animation tutorial

Before you start: install Claude Code on your Mac using the official setup guide. Opus 5.5 requires version 2.1.280 or later. If yours is older, update it in Terminal, the Mac app for typing commands: run claude update for the native installer, or brew upgrade --cask claude-code if you installed with Homebrew.

1. In Terminal, create a project folder and start Claude Code.

mkdir showreel
cd showreel
claude

2. In the Claude Code conversation, switch the model. From here on, you are typing messages to Claude, not Terminal commands. Model configuration docs

/model claude-opus-5-5

3. Paste this prompt into the conversation. Replace [Your brand] with your service name. 30fps means 30 frames per second.

Make a 15-second showreel at 1920x1080,
30fps.
Keep it in one self-contained
index.html file.
Use only HTML, CSS, JavaScript and SVG
or Canvas.
Animate bold type and simple shapes in
rhythm.
End on the text [Your brand].
Do not use external AI or video
generation.
Do not use Remotion, HyperFrames or
frameworks.
Do not use MCP, CDNs or any external
URLs.
Add Play and Replay buttons.
Write a README with preview steps and a
plan
for rendering output/showreel.mp4
later.
Do not create the MP4 yet.
Before installing anything, name it and
why.
Do not run anything that could incur
charges.

4. Preview in a browser. When Claude Code finishes, open the showreel folder in Finder and double-click index.html. Check that Play starts the animation and Replay runs it from the beginning.

Success means text and shapes move for about 15 seconds and end on your brand name. If the page is blank or shows an error, save a screenshot in the showreel folder. Then tell Claude in the conversation: “The page is blank. The screenshot is [file name]. Find and fix the cause,” using the actual file name.

5. Optionally export an MP4. Export each frame as an image, then use FFmpeg, a video conversion program, to combine the numbered images into one video. FFmpeg cannot turn HTML into video directly. FFmpeg documentation Once you approve the preview, paste this:

I approved the preview. Render it frame
by frame
to numbered images, then encode
output/showreel.mp4.
Drive motion by frame number, not the
clock.
List the tools you need and wait for
approval.

The browser and FFmpeg run on your own computer. They are separate from external AI integrations, so this route uses neither a video generation subscription nor MCP.

HyperFrames + Lyria: promos with animated text and shapes

Reference video published on X

{{tweet:https://x.com/Majin_AppSheet/status/2103497319899693327}}

The linked tutorial offers a starter workflow for a similar type of video.

HyperFrames turns web technologies such as HTML into video. Official HyperFrames repository

This is also code-driven motion graphics. The difference from the first route is a dedicated workflow for producing and rendering video. Lyria music is optional; you can render locally without it.

Example video: a 15-second promo for a note-taking app. An opening line invites viewers to capture an idea. Three note cards then appear and move into an organized layout, followed by a closing message. Light background music accompanies the animated text and cards.

The setup guide includes a prompt for this sequence, HyperFrames installation, adding Lyria music and exporting an MP4.

Follow the step-by-step video tutorial →

Set up HyperFrames and add Lyria music

Remotion + Node.js: visuals and sound made with code

Reference video published on X

{{tweet:https://x.com/_nogu66/status/2103665780160421939}}

The linked tutorial offers a starter workflow for a similar type of video.

Remotion is another way to create video from code. One proposed setup combines its animation with sound prepared using Node.js. Remotion audio guide

Example video: a 15-second animation that turns scattered information into an organized layout. Three squares appear with an opening label, move into alignment at the five-second mark, and trigger a short sound. A large closing message then replaces the scene.

The guide explains how to install Remotion, create a short cue with Node.js and place it at the same moment as the visual change.

Follow the step-by-step video tutorial →

Install Remotion and add a sound cue

GPT Image + ElevenLabs: animating generated artwork

Reference video published on X

{{tweet:https://x.com/scisterna/status/2103600499861623106}}

The linked tutorial offers a starter workflow for a similar type of video.

This approach starts with still artwork, then changes its position and layering over time. GPT Image supplies the artwork, while ElevenLabs supplies assets such as narration. Official ElevenLabs connection guide

Example video: a short illustrated story with a bird moving through a forest and spoken narration. The forest background slowly enlarges while a separate bird image travels across the screen. A calm voice accompanies the scene, followed by a closing line of text.

The guide shows how to save the background and bird as separate assets, animate them with HyperFrames and add ElevenLabs narration.

Follow the step-by-step video tutorial →

Animate generated artwork and add narration

Higgsfield + image generation + Remotion: videos with people

Reference video published on X

{{tweet:https://x.com/showheyohtaki/status/2103552919249563691}}

The linked tutorial offers a starter workflow for a similar type of video.

Different tools handle character images, moving footage, music and on-screen text. The proposed setup uses Higgsfield for footage, image generation and Lyria for other assets, and Remotion to combine elements such as text. Official Higgsfield connection guide

Example video: a 10-second café ad combining a person on screen with an animated message. The first half shows a live-action-style scene of someone holding a cup. The second half presents a short invitation to take a break, with background music connecting the scenes.

The guide walks through Higgsfield sign-in and footage generation, preparing Lyria music and building the closing text scene with Remotion.

Follow the step-by-step video tutorial →

Generate Higgsfield footage and edit with Remotion

HeyGen: a presenter-led explainer

HeyGen’s MCP integration provides a route for requesting avatar videos from Claude. It serves a different purpose from HyperFrames, although both are offered by HeyGen. Official MCP overview

Example video: a talking avatar announcing a workshop in Japanese. Against a simple background, the presenter introduces an exercise: write down ideas, sort them into themes, then choose one action. Captions follow the spoken explanation.

The guide covers connecting HeyGen, choosing an avatar and voice, entering the script, checking captions and saving the MP4.

Follow the step-by-step video tutorial →

Connect HeyGen, enter a script and save an MP4

ElevenLabs: generating images, video and audio together

The official ElevenLabs MCP supports image and video generation as well as audio. It is an option when several asset types should be available through one connection. Official announcement

Example video: a roughly 10-second café introduction built around rising steam and calm narration. A plain white cup sits on a table in morning light. Steam moves gently while a voice invites the viewer to pause, accompanied by quiet piano music.

The guide explains how to prepare the image, short footage, voice and music through ElevenLabs, then arrange and export them in Studio.

Follow the step-by-step video tutorial →

Create ElevenLabs assets and finish in Studio

Skills, MCP and APIs in a video workflow

A Skill is a set of working instructions. MCP connects an assistant to external tools. An API is an interface through which software requests an operation from a service.

Adding a Skill that teaches animation techniques does not include a video model’s usage allowance. Connecting through MCP still involves the external account and its terms.

Check three things separately:

• What instructions does Claude need?
• Which service may Claude connect to?
• Which tool turns the assets into the final video file?

Answering these questions exposes missing steps, such as a setup that creates images but has no way to render a video.

The Claude Code-only route needs neither Skills nor MCP. The browser plays the code Claude writes, and FFmpeg can export it if needed.

Choose one first action from the table below. Each linked guide includes setup commands and a proposed starter prompt.

First exercise	Setup guide
Animate text and shapes without external services	Getting started with Claude Code alone
Animate text and save an MP4	HyperFrames installation and export
Align graphics and sound	Remotion plugin setup
Animate artwork with narration	Asset paths and ElevenLabs connection
Create a short character scene	Higgsfield sign-in and generation
Turn a script into a presenter video	HeyGen connection and authentication
Prepare images, voice and music together	ElevenLabs MCP command

During connection checks, ask the assistant to inspect available tools without generating anything. Then create and save one short asset. This order makes it easier to identify which stage is failing.

Where Claude video production costs arise

Claude’s subscription alone does not determine the total cost. If your setup uses external asset generation, that matters too, as does the rendering environment. The Claude Code-only route requires no external generation subscription.

Cost area	What to check
Claude	Plan, usage limits and extra-usage settings
Image, video and audio generation	External plan, credits and regeneration costs
Rendering	Local computer or cloud, plus software licensing
Revisions	Which assets must be made again

A service may use account-based OAuth authentication without requiring a manually entered API key. That does not make generation free.

For a trial, we suggest making one short finished scene before committing to a long video. It lets you assess pronunciation, character appearance and caption size, then decide what needs revision.

Choosing a setup for ads, service introductions and news

Choose based on what viewers need to see: text and numbers, people in a scene, or a presenter explaining something.

For testing ad headlines, favor editable text. For a person’s visual presence, consider generated footage. For spoken explanation, consider an avatar. This is a useful first decision before comparing tool details.

An ad test might change only the opening line while keeping duration, music and destination consistent. None of the tools described here guarantees a higher CTR or lower CPC.

Before commissioning a video, write down the audience, the single message and what should appear in each scene. Choosing the tools from that brief makes the desired result easier to share than starting with a list of services.