Video summary

Fable 5 Built a GTA Clone in One Prompt

Main summary

Key takeaways

Technology

Summary of the video (tech concepts + product features + evaluations)

  • The creator tests Claude “Fable 5” by giving it six “impossible” one-shot prompts (each in a different app/game/product domain) and watching the output end-to-end to see whether the model can autogenerate working interactive software.
  • The video frames the experiment as an evaluation of whether Fable 5 is “worth the hype,” claiming it outperforms previous models by a large margin (compared to earlier “Opus” access).
  • Access timing is called out: availability is mentioned until June 22, after which access is removed for some subscription tiers.

Access / setup mentioned

Claude desktop

  • Look for Fable 5 / “5 of 1 million context” availability.

Terminal

  • Use the model command and select Fable as the option.

The six one-shot prompt demos (what was built + what was evaluated)

1. “Factory” factorial-style game / content automation channel

  • Build time: ~13 minutes 44 seconds
  • What it produced: a fully running automation game that generates and “ships” videos.
  • Notable internal components shown:
    • Thumbnail maker / publisher
    • Ideas sourced from a newsletter, plus an ideas dashboard
    • A “shipped videos” pipeline where winning formats flow back to the idea source
    • A format selector (e.g., choosing best format or running comparisons)
  • Creator review/analysis:
    • Emphasizes it “keeps growing” and behaves like a real simulation.
    • Claims it looked better than similar behavior with Opus 4.8 (allegedly more fixing needed / looked “jank”).
    • Mentions strong performance metrics (e.g., views per minute) implying output is tracked at scale.

2. GTA-like driving chaos game (GTA clone)

  • Build time: ~47 minutes total (includes hitting a token limit then continuing)
  • What it included:
    • Third-person driving
    • Arcade-like physics and NPC reactivity
    • Wanted levels that rise with chaos (wanted level 1 → 2)
    • NPCs running away and the ability to drive into people/cars
    • Car-jacking
    • Drifting (with a control mapped to a spacebar-related input)
  • Creator review/analysis:
    • Highlights it feels “already working” versus expected Opus quality (less janky, more playable).
    • Biggest limitation: no weapons (“not so GTA-like”).

3. Luxury watch reseller website (e-commerce)

  • What it produced: a site from a short prompt; surprisingly generates something like a 3D watch model (not just flat images).
  • Website features shown:
    • Product display with multiple watch “pieces”
    • Add to cart
    • Checkout flow including placing an order
  • Creator review/analysis:
    • Questions whether the watch renders are accurate (suggests some shapes may be too simplistic or not matching real models).
    • Requests improvements such as a better render engine, more realistic 3D, and more convincing presentation (e.g., showroom/counter or on an arm).
    • Overall verdict is very positive for one-prompt functionality.

4. “Intelligence engine” dashboard for creator KPIs

  • What it was: a dashboard-style app to summarize social metrics.
  • Features shown:
    • Charts such as YouTube views over time
    • A “Cumulative” toggle that changes chart behavior (with layout issues—goes out of bounds)
    • A results checker for yesterday’s results
    • Per-video results, sorted by top viewed videos
    • Publish timestamps
    • Buttons to connect Instagram and connect Facebook
  • Creator review/analysis:
    • API wiring for social accounts isn’t completed yet (“we haven’t wired the API key”).
    • The goal is to evolve into a learning loop: use results to feed back into future video idea generation.
    • Described as “absolutely gorgeous,” but needs refinement (thumbnails, chart/layout boundaries, more stats coverage).

5. Drum kit

  • Build time: ~7 minutes 40 seconds
  • What it allows: interacting with drum elements via clicking and keyboard shortcuts.
  • Creator review/analysis:
    • Notes performance/latency issues: it lags slightly in headphones, making it harder to play.
    • Creative potential: could expand into a full looping system (drums → piano/guitar/bass).

6. Rally simulator (driving game)

  • Build time: ~30 minutes before hitting a limit; ~33 minutes total
  • Gameplay controls:
    • Mobile-first: left thumb drag to steer; right thumb for gas/brake
    • Keyboard shortcuts also supported
  • Creator review/analysis:
    • Demonstrates drifting and tilt steering behavior.
    • Includes toggles like sound on/off and restart/repeat.
    • Suggests future improvements: more maps/cars and deployment to iOS/Android.

Overall conclusions from the creator

  • Fable 5 is portrayed as capable of “one-shot” generation of working interactive apps/games, not just mostly broken prototypes.
  • Favorite builds:
    • Factory: wanted to build larger content automation.
    • GTA clone: praised NPC reactivity and driving/physics, but criticized for lacking guns.
    • Website: praised functionality; wants better rendering fidelity.
    • Dashboard: wants more complete KPI integration and UI polish.
    • Drums: wants more musical expansion (e.g., a looper).
    • Rally: wants more content (e.g., maps/cars like Norway).
  • The creator suggests that with a few more prompt iterations, these could become production-like (e.g., adding maps, more features, and deploying to app platforms).

Main speakers / sources

  • Main speaker: the video creator/experimenter who runs the tests and narrates each prompt outcome (no specific name given in the subtitles).
  • Model being tested: Claude Fable 5 (by Anthropic, referenced indirectly via “Claude” and comparisons to “Opus”).
  • Mentioned products/contexts:
    • Google Doc (prompts link)
    • Claude Desktop
    • Terminal model selection
    • iOS/Android app stores

Original video