APIMODELSAPIMODELS

21+ Claude Fable 5.2 Blender 3D with Python Prompts

21+ prompts. 145 things people made with Claude Fable 5.2 in its first week, from Blender films and Three.js scenes to games, fight animations, music and paintings, each credited to its creator and linked to the original post, with the prompt where the creator published one. Fable 5.2 does not draw pixels: every piece here was rendered from code it wrote.

← All 145 Claude Fable 5.2 prompts

Last updated: · Claude Fable 5.2 API — pricing & playground · API docs

TL;DR

Claude Fable 5.2 (model id claude-fable-5-2, released by Anthropic on 9 November 2026) is a language model: it takes text and images and returns text, with a 1M-token context window and up to 128K output tokens. It does not render pixels or synthesise sound by itself. Every film, game, song and painting on this page came from code it wrote, such as Blender Python, Three.js and raw WebGL, Canvas, Web Audio, numpy or LilyPond, which an agent like Claude Code ran, rendered and usually encoded with FFmpeg. Many of the best-known results started from one or two sentences, but the runs behind them were long: the pelican film wrote about 4,800 lines of Python in a session of about 1.1M output tokens, and the FPS level ran for 8 hours at about $550 by its author's account. The 145 cases were posted on X between 4 and 11 November 2026; the few posted before 9 November are pre-release tests their authors labelled "Fable 5.5". 38 cases have the prompt as the creator published it (9 long briefs are excerpted, with our outline), 20 have only the creator's one-line request, and 87 were never published. On apimodels.app, Fable 5.2 costs $4.80 per 1M input tokens and $24 per 1M output tokens, against Anthropic's $8 / $40 list.

Source: Anthropic · Claude Fable 5.2 (the model vendor’s official site)

Hover to preview · click for sound

Photoreal 3D pelican riding a bicycle, driven from Claude Code

Make a video of a 3D pelican riding a bicycle, as realistic as you can make it: believable anatomy and feathers, a real bicycle with working drivetrain motion, physically plausible riding animation, real-world lighting and camera. Take the time it needs. Deliver a finished video file.

What it is:
A photoreal short of a 3D pelican riding a bicycle. Fable 5.2 drove Blender 5.2.3 headlessly from Claude Code, writing about 4,800 lines of Python for a 21-bone rig, feathers and a drivetrain that actually turns, and Cycles OptiX rendered the 444 frames on an RTX 5080 in 56 minutes (technical details in the author's reply, x.com/techfren/status/2119926681716834440).
Clip:
Video, 18 s. About 163K views and 910 likes as of 11 November 2026.
Note:
The prompt comes from the full session trace the author published as a Hugging Face dataset; a block of runtime-environment constraints that follows it is not included here. The author also posted a "receipts" post: x.com/techfren/status/2120203574483550694.
Prompt source:
https://huggingface.co/datasets/techfren/pelican-fable-trace
by @techfrenSource
Hover to preview · click for sound

Prehistoric watering hole with dinosaurs, built in Blender with Python

Prompt not published

No prompt published. "Welcome Fable 5.2, creator of worlds": a prehistoric world with dinosaurs gathered at a watering hole. In a reply, the author says it was "Built from scratch with Fable writing python to control blender" (x.com/jeremyberman/status/2119870801730245009).

Clip:
Video, 18 s. About 282K views and 1,344 likes as of 11 November 2026.
Hover to preview · click for sound

13-shot Blender short with a MacBook and Clawd, made on a livestream

Prompt not published

No prompt published. A Blender short generated in one go during a livestream: 13 shots featuring a desk, a MacBook and a little Clawd figure. Rendering it in 4K on 10 GPUs on RunPod took 52 minutes and cost $5.75.

Clip:
Video, 59 s. About 810K views and 6,802 likes as of 11 November 2026.
Note:
The post itself does not include the prompt; the "Full details" link the author gave is a claude.ai artifact whose content could not be retrieved.
video by @rileybrownSource
Hover to preview · click for sound

Steam locomotive transforms into a robot in a 2D anime short

Excerpt

Create a 2D anime short. It must look completely like real Japanese anime, as if it were taken directly from a finished theatrical anime film. This is traditional-looking, hand-drawn 2D anime, not a 3D animation, a moving illustration, or a stylized web graphic. …

This is an excerpt. Read the full prompt in the original post →
What it is:
A 90-second short in the style of 2D Japanese anime, in which a steam locomotive transforms into a robot. Everything was built in Blender with Python and rendered in Eevee with line art, and even the score was written in code, synthesized with numpy oscillators (see x.com/0kncn/status/2120060332194287853).
Clip:
Video, 90 s. About 69K views and 410 likes as of 11 November 2026.
How the brief is built (our outline):
Opens with the deliverable and the bar: one 90-second MP4 at 1920x1080 and 24 fps that must look like hand-drawn theatrical 2D anime, built in Blender from Python, rendered in Eevee, then inked and painted frame by frame in a second pass, with a synthesized soundtrack. HERO fixes the design with hard numbers: a 1940s 2-8-2 freight locomotive 23 m long that becomes a 21 m robot, down to the rust, the moss and a stork's nest on the cab roof, and requires an original design. TRANSFORM splits one 120-frame transformation into five 24-frame stages and rules out cheating: every part is rigid, nothing stretches, vanishes or appears, and each piece moves on a hinge or slide you can see. WORLD and STORY lay out a real-scale mountain railway (a 126 m viaduct of seven 18 m spans) and the plot beat by beat, from the rockfall to the robot staying behind as the new pier. LIGHT and SOUND set the look (high-key cel light, gouache backgrounds, effects drawn on twos, no particle mush) and a 144 bpm grid where every cut lands on a multiple of 5 frames and 16 music hits land on picture. FORMAT asks for a Python "story clock" with one state per frame; CHECKS has a script test every frame for parts passing through each other, wheel slip, sliding feet and the camera entering a mesh before the final render.
Full prompt:
6,390 characters, at https://x.com/0kncn/status/2119901786060259612
by @0kncnSource
Hover to preview · click for sound

Game-ready 3D zombie war eagle built through Blender MCP

Excerpt

You are building a game-ready 3D creature in Blender through Blender MCP (execute_blender_code for actions, renders and viewport screenshots to check your work). The attached image is the reference. Goal: an outstanding, realistic, production-quality model that matches the reference as closely as possible. …

This is an excerpt. Read the full prompt in the original post →
What it is:
A game-ready 3D zombie war eagle built through Blender MCP and matched to a reference image part by part.
Clip:
Video, 21 s. About 21K views and 304 likes as of 11 November 2026.
How the brief is built (our outline):
Sets up the agent before the art: which Blender MCP tools to use, a working folder, a saved .blend after every stage, and "never ask me questions": when unsure, pick whatever is closest to the reference and continue. WHAT IT IS gives size, units and orientation (about 1.6 m tall hunched, about 4.2 m wingspan, metres, facing -Y, feet at Z = 0) and asks for a static model in the reference pose, with no rig and no animation. THE REFERENCE, PART BY PART walks through the skull, beak, eyes (one blind, with a scar cut into the geometry), body, shoulder wound, wings and the natural and mechanical legs, naming sides as "its left (right side of the image)" so they cannot be confused. HOW TO BUILD IT orders the work: a blockout against an orthographic reference camera, a subdivision-ready body mesh, feathers from a 2048 px atlas of 10-16 feathers it draws itself in numpy (flagged as the step that decides the look), the wound, both legs, then PBR materials. CHECKS, GAME LIMITS and OUTPUT close it: at most 80,000 triangles and 4 texture sets of 2048 px, a GLB under 30 MB with named objects, six named renders including one beside the reference, and a short report.
Full prompt:
9,411 characters, at https://x.com/maxt3chno/status/2119896909062058411
by @maxt3chnoSource
Hover to preview · click for sound

Jerusalem in the final week of Jesus's life, rebuilt in Blender

Build an immersive, historically accurate 3D model of Jerusalem during the week of Jesus's final visit using Blender.

What it is:
A 3D fly-through video of Jerusalem as it stood during the last week of Jesus's life, rebuilt in Blender.
Clip:
Video, 30 s. About 10K views and 40 likes as of 11 November 2026.
by @kplikethebirdSource
Hover to preview · click for sound

Photoreal Bugatti Tourbillon render from one prompt and a reference image

Prompt not published

No prompt published. A short, photoreal 3D render of the Bugatti Tourbillon, made from a single prompt plus a reference image in a run of about 2.5 hours.

Clip:
Video, 33 s. About 112K views and 1,236 likes as of 11 November 2026.
Note:
The author only says that one prompt and a reference image were used; the prompt itself has not been made public.
video by @xikharSource
Hover to preview · click for sound

Star Wars Blender Bench: the First Battle of Geonosis

Prompt not published

No prompt published. A Star Wars "Blender Bench" piece: the First Battle of Geonosis in 30 seconds, which took 1.5 hours to make.

Clip:
Video, 30 s. About 44K views and 292 likes as of 11 November 2026.
video by @adityavg13Source
Hover to preview · click for sound

Rigged cartoon zombie hand that crawls and does a backflip

Excerpt

Using Blender's Python API (bpy, bmesh, mathutils), build a complete animation pipeline from scratch and get it right on the first pass: 1. Model an original stylized cartoon zombie hand with stylized wounds, stitches and a severed stump at the wrist. 2. …

This is an excerpt. Read the full prompt in the original post →
What it is:
From Higgsfield: a rigged zombie hand built with Blender Python that crawls and does a backflip, with the keyframes refined in stages.
Clip:
Video, 19 s. About 63K views and 950 likes as of 11 November 2026.
Note:
The prompt is transcribed from a third-party repository, github.com/yihui-dev/awesome-fable5-2-videos, which cites the Higgsfield blog as its source; this page shows only an excerpt.
How the brief is built (our outline):
Opens with a five-step pipeline in Blender's Python API, to be right on the first pass: model an original cartoon zombie hand, build a production-style rig with custom controllers (IK/FK, gizmo rings, curl sliders), animate it, shade and light it, and render on a cloud GPU. Modelling covers the silhouette, the zombie details on the back of the hand and the stump, and animation-ready topology. Animation runs in two passes: key-pose blocking first, then curve polish in the Graph Editor for overlap, weight and floor contact. The output is one 20-second, 1920×1080 video that shows both the rig and the render. Before the full render it must render six check frames locally (including the mid-air backflip and a pose with the controllers visible) and review each one for twisted fingers, torn mesh and loose wounds. The cloud render has a cost gate: list GPUs and prices, check nothing is already running, quote the job with expected and maximum credits, and start only after a yes; then follow the job until there is a finished video.
Full prompt:
8,138 characters, at https://higgsfield.ai/blog
Hover to preview · click for sound

Transformer truck with an animation of it transforming

Make me a transformer truck with animations of its transforming.

What it is:
Part 4 of camperaigg's test series: a transformer truck that can change shape, plus an animation of the transformation, made in Blender.
Clip:
Video, 12 s. About 3,471 views and 80 likes as of 11 November 2026.
Note:
Pre-release test: the post is labeled "Fable 5.5" and was published before the official release.
by @camperaiggSource
Hover to preview · click for sound

Howl's Moving Castle with its internal machinery shown

Make me Howl's moving castle and also show its internal machinery.

What it is:
Part 5 from camperaigg: Howl's Moving Castle, including a look at the machinery inside it.
Clip:
Video, 35 s. About 3,019 views and 51 likes as of 11 November 2026.
Note:
Pre-release test; the post is labeled "Fable 5.5".
by @camperaiggSource
Hover to preview · click for sound

Mechanical heart with its inner workings revealed in Blender

Make a mechanical heart and show its insides, use blender.

What it is:
Part 6 from camperaigg: a mechanical heart that opens up to show how it is built inside, made in Blender.
Clip:
Video, 30 s. About 5,836 views and 106 likes as of 11 November 2026.
Note:
Pre-release test; the post is labeled "Fable 5.5".
by @camperaiggSource
Hover to preview · click for sound

Blender 3D animation test from a Japanese creator

Prompt not published

No prompt published. A 21-second Blender 3D animation test by a Japanese creator.

Clip:
Video, 21 s. About 104K views and 409 likes as of 11 November 2026.
video by @ICO_AIvideoSource
Hover to preview · click for sound

Blender short finished with zero human feedback

Prompt not published

No prompt published. A Blender short completed in 2 hours 50 minutes with zero human feedback, with its music synthesized in numpy.

Clip:
Video, 24 s. About 583 views and 6 likes as of 11 November 2026.
video by @HyunSeob_Source
Hover to preview · click for sound

Story of Fable's own invention, told as an 18-shot 4K short

Summary

Only the creator's request is public, not a full prompt: "told Fable 5.2 to create a story"

What it is:
Fable was asked to come up with a story of its own, and the result is a 4K short with 18 shots, written in 3,807 lines of code.
Clip:
Video, 59 s. About 113 views and 2 likes as of 11 November 2026.
Note:
The post only contains a summary of the request.
video by @CalebJones800Source
Hover to preview · click for sound

Temple of a Thousand Lamps, compared with Opus on the same prompt

Prompt not published

No prompt published. "Temple of a Thousand Lamps", set against Opus working from the same prompt.

Clip:
Video, 20 s. About 1,498 views and 2 likes as of 11 November 2026.
video by @akshar_aiSource
Hover to preview · click for sound

Airbnb house rebuilt as a 3D walkthrough from 4 screenshots

recreate this house as a 3D video

What it is:
The model was given 4 screenshots of an Airbnb listing and asked to recreate the house as a 3D walkthrough video.
Clip:
Video, 20 s. About 131 views and 2 likes as of 11 November 2026.
by @WalidBou07Source
Hover to preview · click for sound

Glass house through the four seasons, animated in Blender

Prompt not published

No prompt published. A Blender animation of a glass house as the four seasons turn around it, made together with Clico.

Clip:
Video, 25 s. About 5,214 views and 30 likes as of 11 November 2026.
Note:
The author put the details behind an artifact link, which could not be retrieved.
Hover to preview · click for sound

Perpetual calendar clock mechanism animated in Blender

Prompt not published

No prompt published. A mechanical animation of a perpetual calendar clock, made in Blender.

Clip:
Video, 10 s. About 2,486 views and 19 likes as of 11 November 2026.
video by @MakeAI_CEOSource
Hover to preview · click for sound

Low-poly floating island scene made with Blender MCP

create a low poly floating island scene in blender using blender mcp

What it is:
A low-poly floating island scene built with Blender MCP.
Clip:
Video, 29 s. About 923 views and 30 likes as of 11 November 2026.
by @rewind02Source
Hover to preview · click for sound

Blender short from 1,110 lines of Python in 16.4 minutes

Prompt not published

No prompt published. A video made from 1,110 lines of Blender Python, finished in 16.4 minutes.

Clip:
Video, 34 s. About 63 views and 0 likes as of 11 November 2026.

How to write great Claude Fable 5.2 prompts

  1. 1Say what done looks like, then let it take the time. The pelican prompt is three sentences, but it names the deliverable (a finished video file), the bar (believable anatomy, a drivetrain that works, real-world lighting) and gives explicit permission to work long: "Take the time it needs."
  2. 2Pin the design with numbers, not adjectives. The anime locomotive brief gives lengths in metres, a 120-frame transformation in five 24-frame stages and a 144 bpm music grid. The zombie eagle brief sets a triangle budget, texture sizes and a file-size limit. Numbers give the model something to check its own work against.
  3. 3Make every frame a function of time. For browser work, ask for one self-contained page that exposes window.renderAt(t), as the voxel Paris brief does, so a headless browser can capture any frame on demand at a steady frame rate. In Blender the equivalent is a script that sets every state per frame; the anime brief calls it a story clock.
  4. 4Make it check its own renders before the expensive one. The strongest briefs end with checks: render six key frames and review them (the zombie hand), test every frame for parts passing through each other or wheels slipping (the anime short), look at stills from both sides of every cut (the robot night market). Fable 5.2 reads images, so the review loop runs on stills.
  5. 5Put stops and a cost gate in front of anything paid. The robot night-market brief stops twice for approval; the zombie hand brief makes it quote the cloud GPU job, with expected and maximum credits, and wait for a yes. Reported spends on this page run from $5.75 of rented GPU time for a 4K Blender render to about $550 for an 8-hour game build.
  6. 6Give it a reason or a persona when the task is open. The most-liked painting post here had the model invent a painter persona before it painted, and another creator asked it to paint what it feels like to be the model. One-liners can travel too: "invent a new musical form" drew about 132K views.
  7. 7Choose the effort level on purpose. Creators here report medium (an open-source trailer), High (a lighthouse scene) and Extra High (a Bach chorale in LilyPond). On the API Fable 5.2 defaults to high effort; Claude Code and the Claude apps default to medium, per Anthropic's launch post. Raise it for long one-shot builds, keep medium for small fixes.

Prompt formula

Creative coding brief = the deliverable and its format (an MP4 at a set size and frame rate, or one self-contained page) + the design pinned in numbers (sizes, frame counts, tempo, budgets) + a build contract (renderAt(t) or a per-frame state, which renderer) + self-checks on stills before the final render + stops and a cost gate before anything paid + permission to take the time it needs

Example (our template, not a creator's prompt): Make a 20-second Blender shot at 1280x720 and 24 fps: a lighthouse on a rock at dusk, waves breaking, the beam sweeping through the spray, a slow push-in. Build everything in Python from an empty scene and compute every keyframe from the frame number. Render 4 preview frames at 25% size, look at them and list what is wrong before each change. Before the final render, tell me the seconds per frame and the total time, and wait for my OK. Deliver one H.264 MP4. Take the time it needs.

Frequently asked questions

Can Claude Fable 5.2 make a video, a song or a painting by itself?

Not directly. Fable 5.2 takes text and image input and returns text. Every piece on this page was rendered from code it wrote: Blender Python rendered by Blender, Three.js or WebGL pages captured by a headless browser, Web Audio or numpy for sound, canvas stroke engines for paintings, and FFmpeg to encode the result. In a chat window you get code or an HTML page you can open; to get an MP4 you need an agent that can run commands, such as Claude Code.

Did these really come from one prompt?

Some did. The FPS level, the pelican film and the browser Minecraft clone started from one or two sentences. But one prompt rarely means one quick reply: reported run times range from about 20 minutes to 8 hours, one open-world game was built over 7 days, and several creators added follow-up messages. Other cases ran on briefs of about 6,000 to 9,400 characters. Every card says whether the text is the creator's prompt, an excerpt of it, only their one-line request, or unpublished.

How much did these cost?

As the creators reported them: $7.48 and $8.43 for the two pixel-art switches, $21 for the Stonehenge scene, $22.62 for a Swiss lake painting, $34.96 for a 2,000-line Three.js short, and about $550 for the 8-hour FPS level. A 4K Blender render on 10 rented GPUs cost $5.75 in GPU time. This kind of work is output-heavy: the pelican run wrote about 1.1M output tokens, which is about $26 of output on apimodels.app at $24 per 1M, or about $44 at Anthropic's $40 list price.

What do I need installed to make one?

An agent that can run commands (Claude Code or similar) pointed at Fable 5.2; Node.js with Playwright and FFmpeg for anything rendered in a browser; Blender 5.x for the Blender cases, plus the Blender MCP add-on for the cases that used it. A GPU helps for Cycles, but EEVEE runs on a laptop: our own starter script renders 72 frames at 1280x720 in 77 seconds on an Apple M4. Music cases use Web Audio in the browser, numpy or LilyPond.

Which effort level should I use?

Fable 5.2 defaults to high effort on the API; Claude Code and the Claude apps default to medium. Raise it for long one-shot builds where a failed run costs more than the extra tokens, and keep medium for small fixes and iterations. Creators on this page used medium, High and Extra High.

How do I run Claude Fable 5.2 through apimodels.app?

Set ANTHROPIC_BASE_URL=https://api.apimodels.app, your apimodels.app key in ANTHROPIC_AUTH_TOKEN and ANTHROPIC_MODEL=claude-fable-5-2, and Claude Code runs on Fable 5.2; the same key works on /v1/messages and the OpenAI-compatible /v1/chat/completions. If the primary upstream errors, requests switch to a second channel automatically, and failed requests are not billed. It is billed per token: $4.80 per 1M input, $24 per 1M output and $0.12 per 1M cache reads, with no subscription.

Can I reuse these prompts and videos?

The works belong to their creators. Clips from X play straight from X's own servers, every card links to the original post, and we do not host or re-upload any of them. Published prompts are quoted with credit and a link: in full when short, excerpted when long, and never filled in when unpublished. Check the original post before using a prompt commercially. If you made one of these and want the card removed or the credit corrected, email us.

Run these prompts on apimodels.app

Claude Fable 5.2 is live here alongside 160+ other image, video, audio and language models — one API, one key, pay as you go, with $0.10 of trial credit on sign-up.

This library is open source on GitHub

The whole collection is on GitHub as Markdown and JSON: all 145 cases with credits and links to the original posts, our outlines of the long briefs, and two starter scripts we ran ourselves (Blender and Three.js). The videos and images stay with their creators and are linked, not re-hosted. Prompts up to 400 characters are quoted in full; longer ones are excerpted, and unpublished ones are left out. Our own write-ups, outlines and code are licensed CC BY 4.0. Open an issue to add a case or correct a credit.

View on GitHub

Other prompt libraries