Paul J Lipsky's tour collects 18 Claude Fable 5.2 examples, from a playable first-person shooter to a gesture-reactive flock of birds and a draft circuit board. Four are Paul's own experiments. The most useful way to watch is to ask three questions of each: What was actually delivered? Which other tools and human decisions were involved? What would it cost or take to repeat?
Try Fable for an ambitious, bounded visual prototype: a small game level, a short animated scene or a motion-graphics pass on footage you own. Budget the full run and inspect the result. Do not interpret "one prompt" as "one tool" or "production ready". Paul himself revised his Godot game, plans to repair awkward video cuts and still favors Opus 5.5 for routine writing.
Watch the 18-Build Tour
The X links below were resolved from the video's full description on 12 November 2026. Where Paul shows his own work rather than a creator post, the source opens at that point in the episode. The X embeds include direct-link fallbacks.
All 18 Builds at a Glance
| Build | What it shows, and the check | Original |
|---|---|---|
| 1. First-person shooter Felix Rieseberg | Playable Claude artifact; Paul reports about eight hours and a $550 API-equivalent estimate. | Creator post |
| 2. Boat ride through eight worlds Kevin Ngo | Navigable browser experience and a separate video of the worlds. | Creator post |
| 3. The AI dream film Anabology | Fable-directed result with a separate video generator and Suno music. | Creator post |
| 4. Changing-style animated short Paul J Lipsky | A 90-second film from his brief; some styles and shots visibly need revision. | Video chapter |
| 5. Lantern Wake adventure game Paul J Lipsky | Godot and Blender build, improved through a second round of human feedback. | Video chapter |
| 6. Animated zombie hand Higgsfield AI | Gemini brief, Fable and a Blender-connected Higgsfield workflow; not a model-only render. | Creator post |
| 7. JavaScript-drawn short Kevin Ngo | Cinematic coded animation whose frames were drawn by JavaScript. | Creator post |
| 8. Editing this very video Paul J Lipsky | Cuts and motion graphics from Fable; Paul identifies awkward cuts he will change. | Video chapter |
| 9. Interactive bird flock Gregory Wieber | Real presenter footage composited with a simulated flock that reacts to gestures; Paul reports about $23 usage. | Creator post |
| 10. Istanbul 3D recreation Peter Gostev | Detailed explorable city with visible simulation errors; Paul cites a $2,618 build estimate. | Creator post |
| 11. Battle of Geonosis scene Aditya | A fan recreation shown as a rendered scene, not cleared commercial IP. | Creator post |
| 12. Video script in his voice Paul J Lipsky | Uses his old notes and transcripts; he still prefers Opus for routine writing cost. | Video chapter |
| 13. Tamagotchi-style device design Adam | Circuit-board, case and render design; the video does not demonstrate a tested manufactured unit. | Creator post |
| 14. Pong surprise Richard Davey | A simple paddle game that evolves into a more elaborate playable experience. | Creator post |
| 15. Gilt & Gutter shooter Kevin Ngo | A browser-playable shooter with its own art direction and smooth interaction. | Creator post |
| 16. AI sandbox-escape music video Wes Roth | A narrative music-video example; separate model, audio and asset roles should be checked in the original. | Creator post |
| 17. Animated chalk drawing Dan Fein | A chalk-art animation with sound design attributed by Paul to the Fable-led workflow. | Creator post |
| 18. Clawd animation Other Reality | A short character animation about Fable itself. | Creator post |
Games and Interactive Worlds
The opening Felix Rieseberg shooter is the most dramatic example of broad prompting: a tiny, high-detail level that viewers can play as a Claude artifact. But Paul's reported eight-hour run and roughly $550 API-equivalent cost matter as much as the final image. A short prompt can delegate many implementation decisions; it cannot make the compute, testing or iteration free. Kevin Ngo's eight-world boat ride and Gilt & Gutter show contrasting quiet and action-oriented browser play. Richard Davey's Pong surprise is a reminder that a familiar mechanic can be a useful test bed.
Paul's own Lantern Wake is a clearer beginner workflow: he supplied a game direction, let Claude Code work with Godot and Blender, then asked it to improve the graphics. The result is a playable prototype, not a finished commercial game. Review installed tools and permissions yourself before allowing an agent to download software. Peter Gostev's Istanbul demonstrates scale and detail, but characters walk through columns in the shown footage. A visual recreation is not a geographically accurate digital twin.
Animation, Compositing and Editing
The film examples do not share one pipeline. Anabology's dream film combines Fable's direction with a separate video generator and Suno for music. Higgsfield's animated hand starts with a Gemini production brief and uses a Blender-connected plugin. Kevin Ngo's short draws frames with JavaScript. They are different forms of AI-assisted production, not three examples of a model natively exporting finished video by itself.
Paul's animated chase explores style changes, including one yarn-like scene he says did not work as well. His most practical test is editing this episode: Fable selects preview clips, cuts and creates motion graphics without a detailed edit map. Paul likes some choices but flags a generic "Claude look" and awkward cuts that he intends to change. For creator work, preserve the editable project, watch the whole cut with sound and check caption accuracy, pacing and rights before publishing.
Gregory Wieber's flock overlays simulated birds on real talking-head footage and uses tracked hand gestures to influence their motion; Paul relays an approximately $23 usage figure. Aditya's Geonosis scene, Wes Roth's music video, Dan Fein's chalk animation and Other Reality's Clawd clip broaden the range. The Star Wars fan recreation raises an additional commercial-rights question; a demo is not a license to reuse characters or footage.
Writing and Hardware: The Less Flashy Tests
Paul also asks Fable to draft an intro in his voice using notes and transcripts from past videos. He likes its structure and language, but says Opus is good enough for his everyday writing. That is a sensible model-choice test: compare the final approved paragraph, revision count and token bill, not only the first draft. For private transcripts, keep access and retention in mind.
Adam's Tamagotchi-style hardware design moves from idea to board layout, case and product render. It is a design package, not evidence that the shown device was built and electrically validated. Before ordering boards, an engineer should review component availability, schematic, battery and charging safety, enclosure tolerances, firmware, test fixtures and applicable compliance requirements. Anthropic's separate Diode Computers research describes physical prototypes, but it does not validate Adam's design.
The Cost Question Behind the Spectacle
As of 12 November 2026, Anthropic lists Fable 5.2 at $8 per million input tokens and $40 per million output tokens, versus $4 and $20 for Opus 5.5 on the base API. Those are token rates, not flat project prices. Paul's cited $550 shooter and $2,618 Istanbul figures are creator-reported API-equivalent estimates in the video; they should not be treated as universal costs or confirmed subscription charges. Tools such as rendering, image and music generation can add separate costs. Check Anthropic's current pricing before a long run.
Anthropic recommends starting with Opus 5.5 for most coding and knowledge tasks and moving to Fable for ambitious, open-ended technical work. The practical test is a capped head-to-head: same brief, same assets, one revision allowance, and a score for quality, human cleanup time and total cost. A result that looks better but needs ten times the budget may still be right for a launch film and wrong for a weekly script.
A Bounded First Build
- Choose a ten-second animation or one-room game level, not an entire series or open world.
- Give the agent your assets, target platform, visual references, constraints and explicit acceptance criteria.
- Use official installs for Godot or Blender if needed, and approve any new tool or paid service before it runs.
- Set a time and token budget. Require checkpoints, a runnable project and a list of unfinished or substituted parts.
- Review on the actual target device, fix the weakest scene, and compare the result against an Opus run.
Original Creator Posts
Each panel embeds the post Paul cites. X may block embedded playback in some browsers; every panel has a direct source link.
Games and interactive worlds
Films, animation and compositing
Hardware design
Video Chapters
| Time | Segment |
|---|---|
| 00:00 | Intro and model context |
| 01:07 | Games and simple prompts |
| 04:58 | The AI dream video |
| 06:00 | Paul's animated short |
| 08:50 | Paul's Zelda-style game |
| 11:26 | 3D models and animation |
| 13:28 | Editing and recreations |
| 18:16 | Writing and costs |
| 21:11 | Hardware, games and music |
Sources and Editorial Scope
We used Paul's full video and transcript, the creator links in its expanded description, Anthropic's launch post and its model documentation. Costs and build durations attributed to creators were not independently reproduced. A polished clip establishes what was shown in that run; it does not prove repeatable quality, accurate simulation, production safety or clearance of third-party IP.