16:24Jay E | RoboNuggets · Tutorial
I tested Claude Fable 5 against Opus and GPT 5.5 — 6 Surprising Uses
Six one-shot build tests settle where the Mythos-tier model's 2x cost premium actually pays off.
June 10thThe argument in one line.
Claude Fable 5's edge over other models is taste and design judgment, not raw code generation, so the highest-leverage use cases are the ones that ask it to audit, direct, or design rather than just build.
Read if. Skip if.
READ IF YOU ARE…
- You already use Claude Code or a similar AI coding assistant and want a short list of concrete prompts to try during a free-access window.
- You run a small agency or consultancy and want ideas for productizing AI work (custom internal tools, ad creative, strategy decks) for clients.
- You're curious what non-technical people are doing with a frontier model beyond writing code.
SKIP IF…
- You're looking for a rigorous technical benchmark or comparison against Opus/Sonnet/GPT — this is anecdotal, sourced from social posts, not tested by the presenter.
- You want depth on any single use case — each one gets under 90 seconds before moving to the next.
The full version, fast.
The presenter used Claude Fable 5 to research how people are using it, then distilled five use cases before Anthropic's promised usage-based pricing kicks in. First, one-prompt 3D world generation, improved by looping the build against a self-graded quality bar. Second, auditing an agentic workspace (files, skills, MCP servers) for redundancy and staleness, scoring it across six areas. Third, interviewing a user like a consultant to build small custom software tools that eliminate a repetitive task. Fourth, reverse-engineering a product's best-performing ads via YouTube data and generating new ad variants with image/video tools like fal.ai or Higgsfield. Fifth, running a strategy session on your own or a client's business by reading existing goal notes, asking clarifying questions, and outputting a designed HTML deliverable with a SWOT, top actions, and a 90-day roadmap. The throughline: Fable 5's advantage is taste and design sense, which is best spent on audits, interviews, and creative direction rather than pure code output.
Free for members
Chat with this breakdown — free.
Sign in and you get 23 free chat messages on us — ask for the hook, quote a framework, find the exact transcript moment, generate a markdown action plan. Bring your own key when you want unlimited.
Create a free account →Where the time goes.
00:00 – 00:38
01 · Cold open: the pricing deadline
States the premise: a free-access window on Fable 5 is closing soon, and the video is the output of using Fable 5 to research its own best use cases across social platforms.
00:38 – 03:52
02 · Use case 1: Build entire 3D worlds
One-prompt generation of a walkable, explorable 3D castle (Hogwarts-style), improved via a self-graded iteration loop that requires each pass to beat the last by roughly 20%.
03:52 – 06:41
03 · Use case 2: Audit your agentic workspace
Prompting the model to act as a context engineer, score a workspace across six areas out of 10, and surface stale files, duplicates, and unused skills.
06:41 – 07:22
04 · Use case 3: Build custom software for your work
Interview-first pattern: the model asks about the user's role and repetitive tasks, then builds a narrow single-HTML-file tool scoped to one user or team.
07:22 – 08:48
05 · Use case 4: Run a content factory
Reverse-engineering a product's top-performing ads via the YouTube data API and vision, then generating new ad variants with image/video tools like fal.ai or Higgsfield.
08:48 – 10:44
06 · Use case 5: Run a strategy session on yourself or a client
The model interviews the user about goals, then outputs a designed HTML deliverable — SWOT, top three actions, and a 90-day roadmap.
10:44 – 11:14
07 · Sign-off
Reiterates the pricing deadline and points to the free prompt PDF.
Lines worth screenshotting.
- The presenter treated the model itself as the research tool, using it to scan YouTube, X, and Reddit for how people were actually using it before writing the video.
- A 3D explorable world built from a single prompt gets meaningfully more polished when the build is looped against a self-graded 'must beat the previous pass by 20%' quality bar.
- The improvement loop works because visual polish is hard to verify with a single pass, so an explicit relative scoring instruction gives the model something concrete to optimize against.
- Auditing an AI-heavy workspace works best framed as a role: 'act as a context engineer' and score defined areas out of 10 rather than asking for open-ended feedback.
- Before running a workspace audit, the model asks what's been annoying day-to-day, what should stay untouched, and how aggressive the cleanup should be — alignment questions precede the actual audit.
- For client-facing custom software, the interview-first pattern (ask 3-8 questions about the task, then build) produces a local single-HTML-file tool rather than a hosted app, which keeps the deliverable simple.
- The suggested scope for custom software is deliberately narrow: solve for a single user or team, not a general-purpose product.
- Reverse-engineering ad creative combines three model strengths at once: pulling a business's current ads via its Google profile, vision-checking the current ad's quality, and directing new ad concepts with image/video tools.
- The ad-rebuild workflow ends in a concrete, mailable deliverable: a before/after postcard sent to the business owner as a pitch.
- For strategy sessions, having the model output a designed HTML file (rather than answering in chat) makes a multi-section deliverable like a SWOT or 90-day plan easier to read and hand off.
- The strategy-session prompt explicitly has the model interview the user for at least three questions before producing output, treating alignment as a required step rather than skippable.
- None of the five use cases are pure code-generation tasks — they're audits, interviews, or creative direction, framed around the claim that taste and design judgment are the model's real differentiator.
The model's edge is judgment, not just code output
WHAT TO LEARN
Across all five use cases, the highest-value work wasn't generating code from scratch — it was auditing, interviewing, and directing, tasks that lean on design judgment rather than raw output.
- When a build's quality is subjective (visual polish, UI, creative direction), give the model a concrete relative bar to hit — 'beat the previous version by X%' — instead of vague 'make it better' feedback.
- Before running any audit (of a workspace, a business, a piece of creative), have the model ask alignment questions first: what's broken, what must stay untouched, how aggressive to be.
- Scope custom tools to the narrowest possible audience — one person or one team — rather than aiming for a general product; it keeps the build simple and shippable in a single file.
- An interview-first pattern (ask several clarifying questions before producing a deliverable) produces more usable output than a single detailed prompt, especially for tasks with real ambiguity.
- A structured, scored deliverable (a scorecard, a SWOT, a roadmap) is easier to act on and hand to a client than open-ended chat feedback — consider asking for output as a standalone document rather than a conversation.
Terms worth knowing.
- Usage-based pricing
- A billing model where cost scales with how much of the AI model you use, as opposed to a flat subscription; referenced here as the pricing change expected to make Fable 5 more expensive to run heavily.
- Agentic workspace
- The set of files, saved skills, memory notes, and connected tools (MCP servers) an AI coding assistant reads from when working on a person's projects — treated here as something that accumulates clutter and needs periodic cleanup.
- MCP server
- A connector that gives an AI model structured access to an external tool or data source (for example, a design app or business database) beyond its built-in capabilities.
- Loop / iteration prompt
- A prompting technique that has the model repeatedly redo a task while scoring each pass against the previous one, used here to progressively improve the visual polish of a generated 3D world.
Things they pointed at.
01:00channelMatt Schumer's 3D Hogwarts-style world demo
07:40toolHiggsfield
08:20toolSeedance 2 (image-to-video model)
Lines you could clip.
00:08
“We've got maybe around a week of playing around with it before usage based pricing kicks in and makes it much more expensive.”
03:55
“If your customer is lost, you've lost them.”
Word for word.
Read-along
Don't just watch it. Burn it in.
See every word as it's spoken — crank it to 2× and still catch all of it. The same dual-channel trick behind Amazon's Kindle + Audible.
story
00:00So the world's most powerful AI model, CloudFable five, is back. The catch is we've got maybe around a week of playing around with it before usage based pricing kicks in and makes it much more expensive. So really the question is, what do you throw at this thing so that you get the most out of it while it's cheap?
00:15And how do you actually max it out while we still have broader access? Well, what I did is use Fable five to run a last thirty days research project on itself so that we can find out how people are actually using this model. And it went through YouTube.
00:27It went through X as well as Reddit. And I found five great capabilities of Fable that you should definitely try in the next few days. Let's dive into it.
00:38Now, if you truly want to feel the power and how good Fable five really is, there's probably no better way versus just having it vibe code a full three d explorable world like what you're seeing here. This is courtesy of Matt Schumer who shared this amazing piece of work that was created by Fable. And you can see if we scrub through this, it created a world so expansive that it even created the Great Hall, a Quidditch pitch, and pretty much everything that you would expect in an explorable Hogwarts castle.
01:05So it's pretty wild. And Matt also deployed and shared the link to this website, which you can just get below. And it is really good.
01:12Like, you can fly around. You can also mount the room in here and, uh, just freely explore this whole castle, which is crazy how detailed this is.
01:21And this is something that I don't think Opus will be really capable of doing. Now for all these use cases, in case you want to try them out, I made this free PDF guide that you can just grab in our network below. And this gives you all the prompts which you can just copy or you can also give this whole PDF to your cloud code in order to try out everything that I will mention in this lesson.
01:38For this three d world prompt, there is a base prompt that you can, of course, just send to Fable. But what actually increases the polish of that build that Fable has is a secondary prompt that uses Claude Codes slash loop feature. Because if you saw my video from a few days ago when Fable five was first introduced to us, I also tried this prompt.
01:56But you can see here, it's not as polished as Matt's because I didn't really run a loop command in order for Fable to double check its work and actually improve and iterate its polish and this three d world in order to make it better. And so if you make use of this loop future that Claude Code has, then what it will do is it will just improve the visuals of the three d world that you make.
02:16And since design and aesthetics is usually hard to verify, what you can do is to prompt it in this way where you are treating the previous iteration as a 100 index and the pass that Fable will loop against must land at one twenty index or better. So that just means that it is trying to get it to 20% better every time. And if you run that multiple times, then you will arrive at a quality that is more similar to what you have here.
02:39Now you may be curious because three d worlds might be impactful to you or relevant to you if you are building games. But if you're offering services to clients, this is actually still maybe irrelevant to you. To give one example, there's this company that we used to work with called PAM dot co, which here in Australia, basically, they manage a lot of venues and event experiences.
02:59And if you have something like Fable five in your hands, then you could just imagine how you can offer this service of three d worlds or custom software for three d worlds for them, which is similar to this smart maps that they already have live on their site. And so if you're into real estate or you're offering services for Airbnb's, that might be a way for you to capitalize on Fable five even if you're not in the gaming industry.
03:20And by the way, if you want to learn how to build and sell AI systems that businesses actually pay for, then that's pretty much all we do over at the Robo Nuggets community, where not only do you get access to the Cloud Living Masterclass, which we update every week and takes you from zero to mastery with the latest on AI, but you also get access to our agents as a service course, which walks you through how to actually get paid for all these AI skills that you are learning.
03:42You also get to be part of a genuinely great community of AI builders. In fact, you can see just some of the recent wins our members are getting from the program right here. So if you want to start earning from AI, then check that just in the pinned comment below.
03:53Now back to the video. Another great use case for Fable five is you can ask it to audit your agentic work space. And this is really effective because out of all the models, it's really Fable five that can parse through this amount of information the smartest way possible and also give you a lot of great recommendations in order to make your setup much better.
04:11And apart from that, if you are offering this as a service where you're setting up second brain systems for companies, because Fable five is really good at taste and design, you can see I just created this sort of other version of Obsidian, which is a bit more customized for my workspace and my operating system where you have your memory artifacts in here, you have your skills here on the left.
04:29And this is just the first version, obviously. But just from here, you can see just how much potential Fable five has with visualizing your workspace, especially if you think that you are already storing a lot more files than you should, then Fable five is a good way for you to audit that. But obviously, this fancy map that we're seeing with a lot of second brain systems, they're really just meant for visualization purposes so that we can communicate what our workspaces look like under the hood.
04:53The real value of this is when you actually do the audit and ask Fable five to do the audit so that your workspace is cleaned up. And to try that, you just send to Fable this prompt where you'll ask it to audit your authentic workspace like a context engineer, and you can ask it to score six areas of your workspace out of 10 with the idea that by the end of it, you'll have this deliverable where you'll have a scorecard per area as well as points for improvement.
05:17So if you send that to Fable, it will ask you a few questions around what's been annoying you day to day when working with your workspace. It will ask you what is sacred, what is the thing that doesn't need to change. It will ask how aggressive it should be, and it will also guide you how it intends to clean up your workspace or your OS.
05:33In this case, it's telling me that it's going to check our conversation logs in order to assess the skill usage and which skills we probably don't need. And once that audit completes, it will give you really good improvement points on how to make your workspace even better. So here you can see it's showing me some stale pointers that I have in my reference files.
05:51Gives me a view of duplicates and conflicts, which skills I use the most and which ones can probably be deprecated, and just overall giving us a well assessed audit of how to make our space, our operating system much better. Now related to that is that you can actually ask Fable five to build custom software for your line of work.
06:09So even if you have never developed anything before, this model is so powerful that it will get you, like, 90 to 95% there, especially if you are just the sole user or it's just your team who's the user of the software that you're developing. One good example I found is this post by Chris where he basically used Claude Fable five in order to find businesses that just started running ads.
06:29He then pulls each one's Google business profile for a mailable address plus real photos. Because Fable five has vision capabilities, it can vision check the current ad. And because it has that level of taste, then it can also understand which particular leads are good to target.
06:43It then rebuilds the ad into something that their industry actually runs, most likely with something like GPT Image two, which I featured a lot before in our channel, and also prints a before after postcard, which now they can email to the owner of that business. So if you're a creative agency and you're starting to offer AI creatives and ads for clients like these, then this is a workflow that you can explore as well using Fable.
07:05But it's not just limited to that line of work, obviously. So this prompt that I'll share, it has placeholders so that if you send it to CloudFable five, it will just ask you for your role. It will ask you for what repetitive tasks you may be doing, and you can ask it to build custom software for you to make this task nearly disappear.
07:23Another use case that CloudFable five really excels in is when it comes to creating content. Now, you don't necessarily need to use Higgs Field for this, but I think this video of theirs actually gives you a good idea of how you can approach this. Because if you have, let's say, a product in hand, what you can just ask Fable five to do is to have it analyze, let's say, the most viewed commercials of this product by connecting to the YouTube data API and just asking it to reverse engineer them and generate 50 ads for your product.
07:49And because Fable five is really good at taste and directing actually, then you can generate ads that are of much higher quality versus if you would try to generate the same thing with something like Sonnet or Opus. Now if you already have a Higgs Field subscription, feel free to use that.
08:02But do remember that Higgs Field is just a wrapper. So if you want a more pay as you go model, then you can actually just use fall.ai, which also just aggregates and centralizes a lot of these models.
08:11And I think they actually have more image and video models in there. And you can just ask Claude to connect to File AI in order to pretty much set up the same thing. And if you want to get started with that, again, the prompt is just available below.
08:22You'll literally just need to provide clear directions to Fable five in order to do this. And for this prompt, you can see that you can either use the Higgs Field Connector or the Fala AI MCP. And what Fable will do is it will ask you some clarifying questions in terms of what you want to do, then it will plan the month for content pieces that you want to create.
08:40Now this use case is pretty underutilized in my view because it is not a fancy build. It is not as explosive as that three d world animation or demo that I showed earlier. But I think if you have your workspace set up and you're sort of using Claude as your second brain or your entry towards your second brain now, running a strategy session on yourself or your business or another client's business is actually a good way to use Fable five.
09:03And in order to do this, one prompt that you can send is to just have Fable act as your strategist, read the goals and strategy notes that you have in your workspace. Obviously, I'll have Fable interview me for at least three questions because I think that alignment factor is always really important when it comes to doing anything with Fable.
09:19But what I like to do usually is that instead of just giving that output to me in a chat window, I actually have Claude build me an HTML file, a throwaway HTML file, so that it's much easier for me to parse through the information that it will present. To give an example, when I sent that prompt to Fable, it gave me this strategy session, which is this nicely designed form where, let's say, I can send this to a client for a prediscovery questionnaire or you can run this for your own business as well, especially if you have your long term goals sent to Claude in your, let's say, claude.md or somewhere in your workspace, then you'll just be able to do some sort of regular cadence with Claude in order for it to be your strategist instead of just being your employee.
10:00And then coming from that session, you can have Fable even generate for you really good beautifully designed PDFs like these, which depending on what you're after, can provide a SWOT analysis, can provide and summarize the top three actions that this business can take, and even a road map for the next ninety days. So this is is scratching the possibilities.
10:18Right? But the idea is if you already have a workspace set up, you can actually ask Fable to go through that workspace, understand your goals, and also give you some clarity in terms of strategy that is much more well considered versus, again, if you use something like Sonnet or Opus. So there you go.
10:33Five great use cases for you to fully test out Fable five. At least from what Entropic is telling us, it's only available up until the July 7 when it comes to being part of your normal subscription. So definitely what I would do if I were you is I would test this out to no end so that you can really see how far you can go with a model that is as strong as this.
10:53And again, if you want all of those prompts and for you to just test this out, it's all just available below and you can find it for free. That's it for this one. And as always, thanks for watching until the end, and I'll see you all next time.
11:02Thanks.
The Hook
The bait, then the rug-pull.
The pitch is urgency: a free-access window on Anthropic's newest model is closing, so the presenter ran the model on itself to mine social platforms for how people are actually using it, then narrowed the results to five use cases worth trying immediately.
Named ideas worth stealing.
01:50concept
The 20%-better iteration loop
Treat the previous build as a baseline index of 100 and instruct the model that each new pass must score 120 or better, to progressively improve visual polish that's hard to verify in one shot.
Steal forAny generative build where quality is subjective (UI polish, video edits, design drafts) — gives the model a concrete relative bar instead of vague 'make it better' feedback.
04:40list
Six-area workspace audit scorecard
- memory files
- skills
- MCP servers rule quality
- duplicate pointers
- context-eating servers
- overall organization
A structured audit prompt that has the model score a workspace out of 10 per area and produce a scorecard plus improvement points, after first asking the user alignment questions.
Steal forPeriodic cleanup of any AI tool's memory/config sprawl, or as a consulting deliverable for a client's own AI setup.
06:50concept
Interview-then-build custom software pattern
Instead of specifying a software build directly, have the model interview the user (3-8 questions) about their role and repetitive task, then build the narrowest possible tool for a single user or small team.
Steal forFreelancers/agencies building quick internal tools for clients without a full discovery process.
How they asked for the click.
VERBAL ASK
00:00link
“Get the Fable Prompts PDF Guide for free... search for n63”
CTA is stated up front in the hook rather than saved for the end, then repeated at the close; the actual pitch (RoboNuggets community / Agents-as-a-Service course) is a soft mid-roll mention around the 3:15 mark, not a hard sell.
MENTIONED ON CAMERA
FROM THE DESCRIPTION
PRIMARY CTAWhere the creator wants you to go next.
AFFILIATECommission earned if you click.
Visual structure at a glance.
