Fable 5 drops included Claude access on July 7 at 11:59:59 p.m. PT

• Included subscription access now has a hard timestamp: 11:59:59 p.m. PT on July 7, according to trq212's timing note, after the original blog post framed Fable 5 access as capacity-limited.
• Creators treated the last day like rationed studio time, with petergyang's task list saving Fable for project triage, business planning, bug hunts, big project plans, and AI skill refactors.
• MCP workflows did the flashiest work: magnific's thread ran Fable through Magnific for sites, 3D, video, and campaign mockups, while hellorob's ComfyUI run automated the prep work around a generated clip.
• The strongest design pattern was prompt-first production, with MengTo's Koisei prompt specifying the stack, assets, shaders, motion system, sections, and performance checklist before generation.
• Community builds ranged from a SNES game port in AIandDesign's Rotatron post to a platform-prediction train app in om_patel5's train app thread and a WebM/MOV counter tool in ozansihay's counter demo.

A Magnific workflow bundled a 3D game, responsive website, product site with 3D objects, script-to-video, and luxury skincare campaign mockups into one last-day thread from magnific. hellorob used Comfy MCP to pull a shot from the web, trim it, build pose and depth references, swap himself in, write Seedance 2.0 prompts, and stitch a comparison without touching ComfyUI nodes in the ComfyUI run. MengTo published the opposite end of the workflow: a full Awwwards-style prompt with three.js, GSAP ScrollTrigger, Lenis, scroll-scrubbed video, WebGL petals, shader dissolves, and a 60fps acceptance checklist in the Koisei prompt.

Fable 5 leaves included Claude subscription access at 11:59:59 p.m. PT on July 7, per trq212. The same thread says the goal is to restore Fable 5 as a standard subscription model once capacity allows.

The original availability graphic said Fable 5 was fully available on the API and consumption-based Enterprise plans, while Pro, Max, Team, and seat-based Enterprise access was staged because demand was expected to be high and hard to predict. The last-day scramble came from that capacity window becoming a clock.

petergyang's prompt list framed Fable as the model for jobs where a cheaper model could waste the user's time.

The five jobs:

1. Find Fable-worthy work: scan projects, docs, and memory, then rank the top five tasks worth running on Fable.
2. Get life and business advice: read a plan doc and connected tools, then write a one-page assessment with three priorities and what to drop.
3. Make a project ship-ready: inspect the full codebase for real bugs, edge cases, user-facing failures, reproduction steps, and fixes.
4. Plan the next big thing: map phases, key decisions, risks, and open questions so a cheaper model can execute later.
5. Refactor AI skills: audit overlapping or conflicting skills, cut bloat, standardize structure, and check in at milestones.

Allie K. Miller's Fable prompting notes added three setup moves: audit claude.md for conflicting instructions, make skills less prescriptive, and include the why behind the task, including goals, incentives, success metrics, and user context.

magnific's last-day thread used the Magnific MCP as a creative surface, not just an image endpoint.

The thread's use cases:

• Developing a 3D game.
• Building a fully interactive and responsive website.
• Making a product website with 3D objects.
• Creating video from a script.
• Generating four photorealistic 16:9 mockups for a luxury skincare campaign with consistent packaging, palette, typography, and lighting.
• Connecting the Magnific MCP with Claude in a few steps.

hellorob's Comfy MCP run turned the hidden labor of AI video into an agent task list.

Fable pulled the source shot, detected cuts, trimmed footage, ran saved Depth Anything v3 and OpenPose workflows, selected the sharpest frame, directed a GPT Image 2 character swap, wrote Seedance 2.0 prompts, ran generations, and assembled a three-panel comparison. The useful phrase in the post was simple: the preprocessing and prompt writing were usually most of the work.

MengTo argued that with a capable and expensive model, the prompt becomes the valuable asset. His Koisei prompt reads less like a request and more like a production bible.

The prompt specified:

• Stack: vanilla HTML/CSS/JS or Vite, three.js, GSAP ScrollTrigger, Lenis, scroll-scrubbed video.
• Assets: named hero, koi, temple, kimono, lantern, night, and film clips.
• Design system: exact color tokens, typography, grain, borders, and vignette behavior.
• Sections: preloader, fixed nav, hero, manifesto, scroll film, horizontal gallery, day-to-night transition, night film, footer.
• Signature effects: WebGL petals, water ripples, shader dissolve, inner-parallax gallery cards, light/dark nav switching, progress rail.
• Performance rules: one shared WebGL renderer, capped device pixel ratio, lazy video loading, reduced-motion fallback, mobile adjustments.

MengTo later described the recipe as screen recording to prompt, video generation, and a detailed Fable prompt covering every scroll interaction, library, and story beat.

His replies sharpened the economics: he preferred one detailed prompt over ten prompts that could send the agent into a rabbit hole, and he said going into HTML too early costs more tokens.

Fable's game demos clustered around old-school constraints and new-school browser ambition. AIandDesign said Fable ported the game Rotatron to SNES C, then noted the impressive part: Fable had to express a rotating triangle animation with SNES background tiles because the console could not draw it the original way.

Min Choi's roundup collected ten frontier-demo categories in under 99 hours:

1. Command & Conquer ported to native ARM64 iPhone and iPad with RTS controls.
2. Subway Surfers cloned in one hour, from idea to art to 3D models to code.
3. A kid's toy turned into a mobile game, then expanded with a volcano prompt.
4. An ocean wildlife app with AI videos and transitions.
5. Meta Ray-Bans turned into a 4D vision weekend project from a research paper.
6. A 3D world built with math only.
7. A procedural FPS mechanic with no assets.
8. A real-time dungeon generator in Three.js.
9. Dense worlds compared to GTA 6, still running at high FPS.
10. Game-ready Blender assets from simple prompts.

David Comfort used Fable as an evaluator while comparing Kling and Seedance 2.0 on cinematic video shots.

The evaluation called out obedience over texture: Seedance turned an almost imperceptible push-in into a stronger camera move, while Kling stayed closer to the written direction. The OCR'd verdict put Seedance at roughly 2.4x the temporal activity, 0.098 versus 0.040 mean inter-frame score.

A separate character-performance pipeline used Midjourney for the original character, GPT Image 2 for character sheet and scene expansion, Seedance 2.0 for performance, and Claude Fable 5 for orchestration.

The most useful last-day builds were not all spectacle. om_patel5 described a UK train app that pulls Network Rail APIs, tracks trains live, predicts the hidden platform about 75% of the time, and improves by logging advertised versus actual platforms after each journey.

A webcam typing app turned blinks into Morse code: short blink for dot, long blink for dash, pause to commit a letter, head shake to backspace, and sliders for dot length, dash length, eye-closure threshold, and noise filtering.

ozansihay built a browser counter tool with Fable 5 through ChatLLM for editors who need animated number counters without After Effects. The tool lets users set start and end numbers, duration, color, font, then export as WebM or transparent-background MOV.