Jason Lee's Opus 5.5 animated-app tutorial follows a useful three-pass build: make the hydration tracker work, replace the character motions that look stiff, then build a phone-tilt scene. The interesting lesson is not that one model animated everything. It is that Jason assigns different jobs to code, generated video, and a static vector character, then reviews the result on an actual phone.
Opus 5.5 gets Jason to an interactive browser prototype quickly: drink selection, a volume slider, rising water, onboarding, and settings. Its first character animations look rigid. He replaces those with Higgsfield clips, uses a narrated screen recording to request repairs, and separately builds a code-driven Expo scene that responds to touch and phone motion. The video demonstrates a strong prototype workflow, not a shipped hydration business or a controlled Opus-versus-Fable-versus-Astra test.
Watch Jason Build the App
Credit: The app, prompts, decisions, and results are Jason Lee's demonstration. The Gamma segment at 05:09 is sponsored. His original prompt pack is linked below; its page requests an email address for the download.
1. Map the Reference App Before Prompting
Jason studies Waterllama, a one-purpose hydration tracker with a daily target, beverage logging, characters, and progress feedback. He calls his prototype Brownie and gives it a capybara character. Waterllama's site confirms the product mechanics, including custom drinks and hydration ratios. Jason repeats an approximately $100,000-per-month revenue claim in the video, but this article has not independently verified that figure. It should not be used as a market-size estimate.
A home-page screenshot does not explain an app's navigation or motion. Jason uses Mobbin's screen library and MCP to inspect onboarding, home, and paywall references, then adds a phone screen recording to show how the original app responds to taps. Mobbin says its MCP is available on Pro and Team plans, so this part of the workflow is not a free prerequisite. The transferable brief is a map of states and interactions, not a request to duplicate another product pixel for pixel.
Jason uses a Pinterest character image and asks the agent to follow motions from his screen recording. That is useful for a private learning exercise; a commercial app needs original or licensed character art, copy, sound, and animation. Study the interaction pattern, then make a distinct visual system and movement language.
2. Build the Functional Skeleton, Then Click Through It
In the first pass, Jason asks Opus 5.5 for onboarding and a home screen with code-made motion. At 06:26 he demonstrates a working drink picker, volume slider, water-level response, settings, and onboarding. That is more evidence than a static mockup: we see him tap through the flow and identify failures.
The failures matter. He cannot initially swipe to additional drinks. Some challenge controls are placeholders. The jumping and swimming capybara motions are stiff, and one water animation glitches. The trial paywall is explicitly a mockup; clicking it is not proof of subscriptions or payment processing. Jason's first pass is a screen-and-logic scaffold with motion quality still to solve.
3. Give Code and Generated Video Different Jobs
Code for responsive controls
What Jason keeps
The water level, slider, bubbles, drink choices, and button responses stay code-driven. They need to react immediately to a touch or value change.
Why it fits
Code keeps these elements stateful and sharp without storing a separate clip for every possible input. Their actual performance and accessibility still need device testing.
Video for expressive character action
What Jason changes
He asks Claude to call Higgsfield for sipping, leaping, and swimming clips. Seedance 2.5 creates video from media references; Genjutsu is presented as motion transfer. He requests a green background so the character can be separated for the app.
Why it fits
The first code-only body movement looks puppet-like. Generated clips look smoother in the second build, but they introduce compositing, loop, file-size, load-time, and rights checks. The video does not publish a measured mobile performance budget.
A static SVG inside a moving scene
What Jason changes
For the pool screen, the capybara on a float does not need a complex body animation. Jason uses an SVG character while code positions it, draws the water and ripples, and responds to cursor, touch, or phone tilt.
Why it fits
Separating a mostly static figure from interactive scene motion keeps that screen responsive. SVG is resolution-independent, but a detailed SVG is not automatically smaller than a well-compressed bitmap; measure the actual asset.
Jason reports using different Higgsfield credit amounts for the generated clips; those are his demo-account observations, not a stable price per finished app. The real budget also includes discarded variations, upscaling, asset cleanup, and human review.
4. Turn a Screen Recording Into a Precise Revision
Rather than list vague complaints, Jason records the prototype while speaking and pointing at problems: untouched buttons should invite a tap, the water glitch should disappear, a capybara icon should be removed, and the drink row should swipe. He gives Claude the recording path and asks it to integrate the new character clips. At 15:25, he walks through the second build: leaping and swimming look smoother, the glitch is gone, and swiping reveals more drinks.
The reusable technique is to tie feedback to a visible screen, action, and expected result. A voice note alone can still be ambiguous; a screen recording makes it easier to locate the state that needs changing. Re-test each affected path after the revision, including navigation and touch behavior.
5. Preview the Tilt Screen on a Real Phone
The final scene is a separate Expo React Native screen. Jason gives Claude an SVG character, a pool reference, layout instructions, and desired touch/tilt behavior. The resulting code moves the floating character and ripples the water. He opens it through Expo Go using the development-server QR code and demonstrates both tilt and finger input on his phone. This is the clearest on-device validation in the video.
Expo DeviceMotion supports motion readings, availability checks, and permission requests. A production version should ask for motion access in context, work when the sensor is unavailable or permission is declined, offer a touch fallback, and respect reduced-motion preferences. An Expo Go preview does not establish that the whole browser prototype has been packaged, tested, and released as a native app.
Where to Get Jason's Exact Prompts
Jason links Interactive Animation at The Breadcrumb as the source for his three animated-app build prompts. When checked on 14 October 2026, that page asked for an email address before download. We have not reproduced its gated text. The video does show the essential ingredients:
- Structure: define the original mascot and app name; give the agent screen-flow references, a recording of interaction states, and a concrete scope of onboarding plus home.
- Motion repair: identify the three actions that looked stiff, supply the allowed character reference, select an appropriate video or motion-transfer model, and then test the composite in context.
- Device interaction: specify the Expo screen, source assets, pool behavior, touch and tilt response, and the need for an on-device preview.
These are our summaries of what is visible in the tutorial, not Jason's verbatim prompts. A different app should use its own artwork and acceptance tests.
What This Demonstration Does Not Prove
Jason has built an engaging prototype, not shown paid conversion or retention. Waterllama's reported monthly revenue is not independently checked here, and a mock paywall does not process money. The video also does not show account persistence, accessibility review, consent and privacy design, offline behavior, store approval, or performance measurements on a range of phones. The work to test those parts is where a polished demo becomes a dependable product.
Nor does this establish that Opus 5.5 is categorically better than Fable or Astra. Jason does not run the same assets, prompt, tool access, time budget, and evaluation rubric across models in this episode. His earlier Fable-based animated-app build is useful context, but it is a different app and workflow.
Jump to the Video
| Time | Segment | Time | Segment |
|---|---|---|---|
| 00:00 | Introduction | 00:47 | The app and reference |
| 02:20 | Structure and Mobbin | 05:09 | Gamma sponsor |
| 06:26 | First build | 10:33 | Improving character motion |
| 15:25 | Second build | 16:22 | Phone-tilt scene |
Business Idea: A One-Purpose Progress Companion
The useful product pattern is narrow: take one action people already want to repeat and make progress feel visible. A small music studio, sports coach, or skills instructor could test an original practice companion without building another general habit tracker. Start with one week of real use, not a paywall.
Test a one-purpose companion
Find a reachable niche, then prototype one satisfying interaction.
Sources and Resources
Creator: Jason Lee's tutorial and email-gated prompt pack. Product references: Waterllama, Mobbin MCP and plan availability, Higgsfield video models, and Expo DeviceMotion. Creator links: Gamma sponsor, Higgsfield, and Mobbin discount. Sponsor and referral offers can change; check terms before buying.