Video summary

FABLE 5.1 Is Here And It's PERFECT For Vibe Coding Games (FULL Unity + Blender Workflow)

Main summary

Key takeaways

Technology

AI-generated full Unity game inside the Unity Editor

  • The speaker demonstrates a tennis game being built within Unity.
  • Assets, environment/scene, character models, game logic, and UI are created using AI tools.
  • The game is deployed to the web and runs in a browser.

Fable 5.1 (Anthropic) as the main “productizing” model

The speaker claims Fable 5.1 took an earlier prototype and brought it to life, including:

  • Generating the full environment/scene content
  • Adding day/night mode (lighting and environment changes)
  • Iteratively improving game mechanics, visuals, and missing elements
    • Example: creating a racket when none existed
  • Creating UI assets and integrating them into the HUD
  • Adding audio (sound effects + music) via an audio tool

Prototype using multiple AI tools before Fable 5.1

Before switching to Fable 5.1, the prototype relied on several tools:

  • Grok 4.6
    • Used to get a simple Unity prototype working.
  • “GPT 5.6 SolMax Fast”
    • Used for tennis-specific animations, including:
      • idle
      • forehand
      • backhand
      • serve
      • smash

Then the game was handed to Fable 5.1 for continued iteration toward an MVP.

Unity Editor advantages for AI workflows

The speaker highlights Unity’s Scene View and its usefulness for AI-assisted editing:

  • Panning/inspecting scene elements
  • Clicking objects and viewing properties
  • Using many config “knobs” to tune parameters for AI control/adjustments

They also mention they didn’t touch Unity Editor manually much—only adjusting settings when needed.

Asset pipeline guidance (mock-ups → Blender assets → Unity integration)

The speaker emphasizes an early “concepting” step:

  • Mock-ups / mood boards
    • Generate multiple variations from image models
    • Select/lock a concept before producing final assets
    • This reduces expectation mismatches—even with similar prompts

Image generation tools explored for mock-ups

  • Grok Imagine 2.0 (image-to-image style mock-ups)
  • NanoBanana 2 Lite
  • GPT Image 2.0
    • Note: can support transparency for UI

Guided asset creation from mock-ups

From mock-ups, the workflow includes:

  • AI crops character/stadium components
  • Generates character turnaround sheets (A-pose, multiple angles)
  • Generates multiple stadium angles for environment references

Blender as the core modeling/animation tool (not Unity)

The workflow is split across tools:

  1. Unity CLI / Unity tooling
    • Creates a placeholder scene and scale using primitives.
  2. Blender
    • Produces the real assets and animations.
  3. Unity
    • Imports the generated assets afterward.

The speaker uses Blender MCP tools to enable AI to:

  • Modify Blender models headlessly (fast, automated)
  • Or modify models live in an open editor (so the user can watch changes)

Fable/AI iterative environment building

The environment build is described as iterative:

  • Start with blockouts in Blender
  • Then replace placeholders with more detailed Blender outputs
  • The speaker notes it’s not perfect at first, but improves significantly once Fable 5.1 takes over.

Character creation and animation details

Character generation tools

For character generation, the speaker uses third-party tools:

  • Meshy (also mentions Tripo)

Meshy process:

  • Drag-and-drop an A-pose image
  • Enable multi-view
  • Generate the character model
  • Export into Blender

Rigging and animation approach

  • Grok 4.6 rigging
    • Attempted but didn’t produce good animation results (per speaker).
  • GPT 5.6 SolMax Fast
    • Generated the tennis-specific animations via a prompt listing required animations.

Important Blender IK note: add joint “kinks”

  • The speaker advises adding slight bends (“kinks”) in joints.
  • Reason: perfectly straight joints can confuse inverse kinematics (IK) behavior.

Common AI rigging artifact: mesh tearing/holes

  • Explained as vertex-to-bone weight assignment issues during rigging.
  • Fix strategy: instruct the AI to correct weight painting to eliminate tearing/warping.

Using Unity CLI to coordinate AI agents

The speaker describes Unity CLI (and a “Unity CLI Plus” add-on they created) to enable AI agents to:

  • Open projects
  • Create scenes
  • Interact with the Unity editor programmatically
  • Parallelize tasks, scaling from:
    • placeholders first
    • then refined assets after Blender imports

Game iteration results / MVP

After handing the project to Fable 5.1, the speaker claims the pipeline achieved:

  • Correct integration of characters into Unity
  • Tennis mechanics, including:
    • collisions
    • iterative mechanics improvements
  • Visual polishing, such as:
    • crowds added (with an initial bug where crowds were “asleep,” later fixed by AI)
    • rotation/import fixes between Blender and Unity
    • improved sky/lighting: golden hour + night mode
    • added props/effects: billboards, confetti, ground effects
    • improved racket model

They conclude the game became “fairly decent” for about a day’s work, while noting it wasn’t made in a single prompt.

UI and audio generation

UI assets

  • Used GPT Image 2.0 for 2D HUD assets with transparency
  • Fable 5.1 integrated and “skinned” placeholder UI into a working HUD
  • HUD includes shot feedback, such as:
    • timing too early/late
    • smash/backhand information

Sound effects and music

  • Used an “ElevenLabs sound effects and music skill”
  • Prompt included:
    • generate multiple candidates
    • use a sub-agent to blindly assess appropriateness
    • avoid too-sharp/piercing audio

Resource / toolkit mentioned

The speaker promotes vibegamedev.com as an “all-in-one toolkit” including:

  • Agent skills for AI game dev
  • Projects, prompts, and source code
  • Example project mentioned: RoboTennis

Main speakers / sources

  • Main speaker: A YouTube creator (unnamed in the subtitles) demonstrating the pipeline.

AI model / tool sources mentioned

  • Fable 5.1 — Anthropic (primary agent for game completion)
  • Grok 4.6 — prototype, rigging attempt, tennis test scene
  • GPT 5.6 SolMax Fast — animations, weight-paint/tearing fixes
  • Grok Imagine 2.0, NanoBanana 2 Lite, GPT Image 2.0 — mock-ups and UI assets
  • ElevenLabs — sound effects and music skill

Software mentioned

  • Unity (engine + CLI)
  • Blender (assets/animations)
  • Meshy/Tripo (character generation)
  • Blender MCP

Original video