Claude Fable 5 Made This Entire Video By Itself.

Summary
Nate Herk presents a demonstration of an end-to-end autonomous YouTube video generated by Anthropic’s Claude Fable 5 using Claude Code’s /goal command. After an introduction, Herk plays the completely AI-produced video segment (featuring a synthetic avatar, cloned voice, script, and code-rendered motion graphics), before returning to analyze the Claude Code execution log, prompt structure, token usage, and costs.

---

What is shown

• [00:00 - 00:06]: Real Nate Herk introduces his experiment: giving Claude Code a single prompt via the /goal command and leaving for the gym.
• [00:06 - 03:23]: The autonomous video generated by Claude Fable 5 plays:
  • [00:06 - 00:30]: Meta-reveal showing an AI avatar of Nate Herk with on-screen HUD tags confirming synthetic avatar, cloned voice, and Claude-written script.
  • [00:31 - 00:53]: Overview of Claude Fable 5 as the first publicly available model in Anthropic’s “Mythos” tier above Opus.
  • [00:54 - 01:23]: Benchmark and case study animations: Stripe migrating a 50M-line Ruby codebase in 1 day; converting screenshots into source code; autonomously beating Pokémon FireRed from raw screenshots alone.
  • [01:24 - 01:44]: Long-horizon task abilities (3M+ token context, file-based memory notes, reaching the final act of Slay the Spire 3× more often than Opus 4.8) and pricing overview.
  • [01:45 - 03:03]: Four-station production breakdown explaining the autonomous pipeline: script fact-checking, ElevenLabs audio chunking (<60s to prevent voice drift), HeyGen Avatar 5 rendering (via Playwright browser automation and direct API), and FFmpeg assembly with GSAP/HTML Hyperframes motion graphics verified through automated visual self-critique loops.
  • [03:04 - 03:23]: Autonomous outro and sign-off mimicking Nate’s standard channel ending.
• [03:24 - 05:46]: Real Nate returns to inspect the Claude Code session in VS Code:
  • Terminal log showing the session completed in 1 hour, using 380k tokens across 58 tasks.
  • Account usage page showing the run consumed ~40% of his $200/month plan.
  • The exact text of the /goal prompt detailing formatting, styling constraints, avatar chunking, verification rules, and reputation risk context.

---

Claims & numbers

• Anthropic released Claude Fable 5 on June 9, 2026, marking the first time the Mythos model tier above Opus was made available to all paid plan users (previously restricted to vetted security partners) (narrator / visual [00:31 - 00:43]).
• Pricing for Claude Fable 5 is stated as $10 per million input tokens and $50 per million output tokens (narrator [01:38 - 01:42]).
• Stripe used Fable 5 to compress months of engineering into days, completing a full migration of a 50-million-line Ruby codebase in 1 day, a project originally scoped at 2+ months for an entire team (narrator [00:56 - 01:07]).
• Fable 5 beat Pokémon FireRed from start to finish using raw screenshots alone without maps or navigation aids (narrator [01:13 - 01:22]).
• Using file-based scratchpad memory over 3M+ tokens, Fable 5 reached the final act of Slay the Spire 3× more often than Claude Opus 4.8 (narrator [01:24 - 01:37]).
• Voice generation with ElevenLabs was split into chunks under 60 seconds each to eliminate voice drift over long takes (narrator [02:04 - 02:14]).
• The entire agent execution took 1 hour, consumed 380,000 tokens (381k context tokens), ran 58 tasks, and utilized ~40% of Nate’s $200/month Claude subscription limit (Nate Herk [03:48 - 04:34]).

---

Notable quotes

• “What you’re watching right now was not filmed. This avatar is AI. The voice you’re hearing is a clone of mine, and every single word of this script was written by Claude.” — AI Avatar / Narrator [00:06 - 00:14]
• “I just typed one prompt into Claude Code and walked away. And everything else—the research, the script, the voice, the avatar, the motion graphics—all of it happened on its own.” — AI Avatar / Narrator [00:20 - 00:30]
• “One prompt went in, and a finished, fully edited YouTube video came out the other side. That’s what a Mythos-class model does the same week it comes out.” — AI Avatar / Narrator [03:02 - 03:11]

---

Assessment
This is a genuine hands-on workflow demonstration showcasing an autonomous agentic media pipeline orchestrated via Claude Code and Claude Fable 5. While the video rendering pipeline leverages third-party tools (HeyGen, ElevenLabs, FFmpeg, Hyperframes) scripted and inspected by Claude rather than generating raw video pixels natively, the execution logs and prompt proof confirm an entirely autonomous multi-modal agent run.

---

Lyrics & themes

• Narration Outline:
  • Meta-Reveal [00:06 - 00:30]: Disclosing the artificial nature of the video segment.
    • Quote: “I didn’t write this, I didn’t film it, I didn’t edit it, and while it was being made, I never saw a single frame of it.” [00:14 - 00:20]
  • Fable 5 Overview & Coding Capabilities [00:31 - 01:08]: Launching the Mythos tier and highlighting enterprise coding feats.
    • Quote: “Stripe said Fable 5 compressed months of engineering into days.” [00:56 - 01:00]
  • Vision, Gaming & Context Benchmarks [01:09 - 01:44]: Visual reasoning (Pokémon FireRed), scratchpad memory (Slay the Spire), and API token costs.
    • Quote: “It reached the final act three times more often than Opus 4.8.” [01:34 - 01:37]
  • The 4-Station Autonomous Pipeline [01:45 - 03:03]: Dissecting script generation, voice anti-drift chunking, browser automation for avatar generation, and GSAP/Hyperframes programmatic editing with automated visual QA.
    • Quote: “It rendered out frames from every scene and visually reviewed them... until it all passed.” [02:55 - 03:01]
  • Channel Outro [03:04 - 03:23]: Standard YouTuber call-to-action seamlessly mimicked by the AI.

---

Lore & references

• Mythos Tier: Anthropic’s flagship intelligence tier placed above Opus; previously held in closed safety testing (Project Glasswing / security partners) before the Fable 5 release.
• Claude Code & /goal: Anthropic’s terminal-based agent tool equipped with long-horizon execution hooks, file-based memory, and stop-task verification loops.
• Slay the Spire & Pokémon FireRed: Prominent long-horizon computer-use and visual reasoning benchmarks for multi-modal frontier models.
• Playwright Fallback: Reflects real-world agentic behavior where the AI circumvents unexposed API endpoints by spinning up headless browser automation to click web UI buttons manually.

---

Visual style & craft

• The inner video adopts a clean, dark-mode tech aesthetic matching professional motion design standards: animated vector diagrams, code diffs, stylized retro Game Boy graphics, and floating PIP (picture-in-picture) avatar positioning.
• Motion graphics are not generated as diffusion video clips; they are programmatic web-rendered animations constructed in HTML/CSS using GSAP (GreenSock) inside the Hyperframes rendering framework, synced to word-level audio timestamps.
• Visual self-correction is highlighted via automated inspection contact sheets, where the model took snapshot frames across the render to detect and repair bounding box overflow or clipping errors before final encoding.

Described by gemini-3.8-flash on 2026-09-29 from the video’s audio and frames.