I’ve spent over a month testing Fable 5.2. It’s the most capable model I’ve used. It’s also the easiest to use badly.
The reason is simple. Fable is extremely literal. It does exactly what you say, not what you meant.
So think of it like a genie. If you didn’t get what you wanted, the wish (your prompt) was wrong.
Here’s how to make better wishes.
1. Say exactly what you want.
You don’t need a long, clever prompt. You need a clear one. Say who it’s for, what it should do, and what good looks like.
Bad: “Put together an analysis of our competitors.”
Better: “Put together an analysis of our five biggest competitors for our leadership offsite. Cover pricing, who they sell to, and what they’ve launched this year. End with the three moves you think we should make. Keep it to five pages.”
2. Show it, don’t describe it.
I asked Fable to redesign my newsletter’s homepage. The first try was awful. Then I gave it a handful of sites I loved and said: find what these have in common, and make mine like that. The new versions weren’t even close to the first one.
Words only go so far. So give it examples: a report you admired, or some past pitches that worked. You can even give it an email in the tone you like. Anything like this helps a ton.
3. When you don’t like it, say why.
This is the one I’m guiltiest of. Often when the model gives me something I don’t like, I’ll literally just say “nah” or “I don’t like it.” A lot of models actually do pretty well with that. With Fable, it doesn’t work. It just gives you more of the same.
You need to say exactly what’s wrong. For example: “This is way too formal. Cut the length in half and lead with the price.”
4. Catch it early.
On a big, long-running job, if you see it heading the wrong way after your first prompt, don’t just let it keep going. Steer it.
Be very clear about what it’s doing right and what it’s not, and tell it how you want it to adjust from here. That way you’re not wasting tons of time and money on a result that isn’t what you want.
For example: “This is good, except it’s way too long, and you’ve only looked at the March numbers. I need the whole previous calendar year laid out as one table with a short summary on top.”
5. Do one piece first.
Asking for everything at once gets you a mediocre job everywhere. So have it get one piece perfect first.
Say you need a deck. Don’t ask for all 20 slides. Have it make one slide, and work on that one until it’s right: the design, the layout, the tone. Once that slide is exactly what you want, tell it to build the rest of the deck the same way. It’s much easier to fix one slide than twenty.
6. Ask for proof, not “done.”
Don’t just say “do this and tell me when it’s done.” Say “do this, then check it against [the original numbers, the brief, the example I gave you] and tell me what doesn’t match.”
7. Tell it it’s in charge.
Fable is careful. It can stop to double-check things you never asked it to check. Adding these two lines to your prompt helps a ton:
- “You own this.”
- “Don’t add extra steps I didn’t ask for.”
Following these tips will get you 80% of the way there. If you want to go even deeper, including how I’m using it to build a photorealistic New York City game, I made a 17-minute video going through all of this in more detail.
Hit reply and tell me what you want me to write about next.
— Matt
What should I write about next? Vote or suggest a topic →