YouMind
登录

How I got 3M views with AI slideshows and Fable 5.2

@Inomsxbt
英语2026年11月17日
128K
545
20
32
630

TL;DR

A comprehensive guide on creating viral UGC slideshows using Claude Fable 5.2 for analysis and scripting, combined with AI image generation and Postiz for automated scheduling.

By the end of this guide you will have one finished 6-slide AI UGC slideshow, five variations of it and a full week of posts scheduled. It is a method you can repeat on any app or product, on any platform that supports slideshows. What you get is a repeatable way to make good slideshows fast.

The slideshow comes out of one simple loop and the loop is what this article is about:

  1. Find a slideshow format that is already working for someone else.
  2. Have Claude Fable 5.2 take it apart and explain why it works.
  3. Rebuild it around my app, slide by slide.
  4. Turn the one that works into five or six variations.
  5. Schedule the batch in Postiz and read the numbers a week later.

Here is what you need to follow along:

A Claude account with Claude Fable 5.2 selected, which is the current top model An image generator (use the best one you have access to)

Canva or CapCut for the text overlay. 5 to 10 screenshots of the app or product you are promoting.

A free Postiz account at https://postiz.pro/inomsxbt

Everything below follows one example from start to finish, with the exact prompts, so you can copy them and get the same kind of result.

  1. Pick one slideshow format that already works

Don't start from a blank page. A format that already works for someone else is a safer bet than an idea you made up at 11pm.

Here is how to find one:

  1. Open TikTok or Instagram and search "your niche + slideshow", plus two or three close variants.
  2. Filter to the last 30 days
  1. Open the posts with the most views, then check the account. If it has under 50k followers and the post still did well, the format did the work, not the audience.
  1. Save 10 of these. Screenshot every slide of each one into a folder, one subfolder per post.

Then cut the list down with four questions:

  • Can you rebuild it without filming a person?
  • Is it 5 to 8 slides?
  • Is the hook on slide 1 plain text on an image?
  • Does the app only show up around slide 4 or 5, not slide 1? If a post fails any of those, skip it. Everything left is a candidate.

The running example for the rest of this guide is a made-up sleep tracking app called Nightly, and the format is "i tracked my sleep for 30 days, here is what changed". Swap in your own app and format as you go.

  1. Teach Fable 5.2 your app: the app brief

If you skip this step, every slideshow will sound like it was written for a generic app. So write one document about your app, once and let Claude read it every time.

In Claude, create a Project called "UGC Slideshows" and paste the brief into the project instructions. Every new chat inside that project starts with the brief already loaded. Select Claude Fable 5.2 as the model.

The brief should be short and blunt. Use this template:

APP: [AppName]

ONE LINE: [what it does in plain words, no marketing language] WHO IT IS FOR: [one specific person, e.g. "night-shift nurses who can't switch off"]

THE MOMENT: [the single moment a new user says "oh, that's cool"] WHAT USERS COMPLAIN ABOUT BEFORE THEY FIND US:

  • [complaint 1, in their words] 
  • [complaint 2] 
  • [complaint 3]  VOICE: [e.g. "talks like a friend texting, lowercase is fine, never says 'game-changer'"] NEVER SAY: [words and claims you will not use] MY 5 BEST POSTS EVER (hook only):
  • [hook] 
  • [hook]  ... ASSETS: 8 screenshots in the folder /app-screens, named 01-home, 02-streak, etc.

The "never say" line matters more than it looks. Models like certain phrases and your audience can smell them. Add to that list every time a draft annoys you.

The fastest way to fill it in is to talk. Dictate five minutes about your app into your phone, paste the transcript into Claude, and ask Fable 5.2 to turn it into the template above. Then add one more instruction: "List everything you had to guess, as questions." Answer those questions and the "THE MOMENT" line usually gets sharper.

  1. Tear the reference slideshow apart

Now give Fable 5.2 the reference slideshow from step 2. Open a new chat inside the project and attach every slide as an image, in order.

What you ask for matters most. A summary of the story is useless here. You want the format: the hook, the pacing, where the text sits, and where the app comes in. Use this prompt:

I'm attaching every slide of a slideshow I found on [TikTok or Instagram]. The slides are in order.

Do a teardown. Don't summarise the story. I need the reusable format.

For each slide give me:

  1. The job of the slide (hook, setup, proof, turn, payoff, CTA) 2. The exact on-screen text, word for word
  2. Where the text sits (top, middle, bottom; left or centred),its size relative to the image, and whether it has an outline or a box
  3. What the image shows and how it is composed (subject, angle, lighting, crop)
  4. The emotion the viewer is meant to feel Then answer these:
  5. What is the hook formula, as a fill-in-the-blank sentence? - What is the open loop that makes someone swipe?
  6. On which slide does a product appear and how is it introduced? - Which three things must stay identical when I copy this, and which can change?
  7. What would make this format stop working? Return the slide-by-slide part as a table and the rest as short bullets.

The last question is the one people skip. When Claude tells you what would break the format (too polished, a logo on slide 1, a sales line before slide.

4) you get a list of things to check before you post.

Here is the kind of table that comes back, shown with the Nightly example. It illustrates the shape, so your real output will differ:

Inoms - inline image

Read the whole thing before moving on. If it got something wrong, for example it called slide 3 "proof" when it is really the turn, say so in the chat and it will fix the table. That correction is what makes the next steps accurate.

  1. Write the slide script for your app

With the teardown and the app brief in the same chat, ask for the script. Use this prompt:

Using the teardown above and the app brief, write a slideshow for [AppName] in the same format.

Rules:

  • Same number of slides as the reference, and the same job for each slide. - Keep the hook formula. Change the topic to: [your topic]. 
  • Slide 1 text: under 10 words. Every other slide: under 18 words. - The app appears on slide [5], introduced the way the reference introduced its product. No sales line. 
  • Follow VOICE and NEVER SAY from the brief. 
  • If a number is needed, write [NUMBER] and I will fill it with a real one. Do not invent statistics.  For each slide return:
  • slide number 
  • on-screen text 
  • an image prompt (vertical 9:16, phone-photo look, no text inside the image, same lighting and subject across all slides) 
  • which of my app screenshots to use, if any  Then give me:
  • a caption under 150 characters and 3 hashtags 
  • 3 alternative slide 1 hooks, ranked by how closely they match the hook formula

Two lines in that prompt do a lot of work.

"No text inside the image" keeps the image model from drawing broken letters. We add the text ourselves in step 6.

"Same lighting and subject across all slides" is what makes six separate images look like one person's camera roll instead of six stock photos.

Never post the first draft. Read every line out loud and rewrite anything you would not say to a friend. Give the hook the most attention: pick one of the three alternatives, then change a word or two so it sounds like you.

  1. Generate the slide images

Take the image prompts from the script and run them in your image generator. A few settings and habits keep the result looking like one camera roll.

Aspect ratio: 9:16, and 1080 x 1920 if the model lets you pick a size.

Slide 1 first. Generate it, pick the best of four, then attach it as a reference image when you generate slides 2 to 5. That is what keeps the person, room and lighting the same.

One style block, reused. Paste the same sentence at the end of every image prompt, for example: "candid phone photo, natural window light, slightly imperfect framing, no filters, no text, no logos."

Never generate the app screen. Image models draw fake interfaces that look right from far away and wrong up close. For the slide where the app appears, use a real screenshot from your app screens folder.

Generate 3 to 4 options per slide and keep the one that looks least like AI. Perfect skin and perfect symmetry are the giveaway.

This is the shape of one image prompt:

A [age]-year-old [person description] sitting on a [location], looking at their phone with a tired but relieved expression. Candid phone photo, natural window light, slightly imperfect framing, no filters, no text, no logos. Vertical 9:16.

Before moving on, zoom in on every image and check hands, teeth, background text and anything that should be symmetrical. Expect to regenerate several slides. That is normal.

Inoms - inline image

Six AI-generated images in one album. Image 5 is a stand-in screen from my phone Clock app.

  1. Add the text overlay

The text is added after the images, so it is always sharp and always spelled right. You can do this in Canva or CapCut. Both are free to start with and no code is needed.

  1. Create a custom design of 1080 x 1920 px (9:16).
  2. Upload your six images and put one on each page, scaled to fill the page.
  3. Add a text box to each page with the text from your script. Use a bold sans-serif font, centred.
  4. Set the size and position from the settings below.
  5. Style the first page's text once, then copy that style to the other pages so every slide matches.
  6. Export all pages as JPG or PNG and check them in order.

Save the finished design as a template. For the variations in step 7, swap the images and the text and leave everything else alone.

Inoms - inline image
Inoms - inline image

The finished template in Canva: six pages, one text box per slide, with a shadow so the text stays readable.

The settings used here, and why:

Hook slide: 84px, centred. It has to be readable on a phone held at arm's length. Other slides: 56 to 64px.

White text with a black outline of about size/14. It stays readable on bright and dark images alike.

Top margin 260px, bottom margin around 540px. The platform puts its own buttons and caption over the edges of the slide. Check these values against the in-app preview on your platform, since the layout changes now and then.

Max 18 characters per line on the hook, 24 on the rest. Short lines look native. Wide paragraphs look like an ad.

  1. Turn one winner into variations

Once a slideshow is posted and doing well, don't start over. Go back to the same chat and ask for variations.

The prompt: This slideshow worked. Make 5 variations that keep the same mechanism and change the surface.

Keep: the hook formula, slide count, the job of each slide, the text length limits, and the app appearing on slide 5.

Change: the scenario, the person's situation, the specific story, and the emotional angle. Use a different angle each time, from this list: relief, embarrassment, curiosity, frustration, pride.

Each variation must be different enough that someone who saw the first one would not feel they are watching a repeat.

Return the same format as before, plus a one-line "angle" label per variation so I can track which angle performs.

The angle label is the useful part. After a week you can see whether "embarrassment" posts beat "relief" posts, and that tells you what to ask for next.

Then repeat steps 5 and 6 for each variation. The first slideshow takes the longest. The five after it go fast because the brief, the teardown and the script are already in the chat.

  1. Schedule everything in Postiz

By now you have a set of finished slideshows in folders. This is where most people stall, because uploading, captioning and timing each one by hand eats the evening. Postiz handles that part.

The routine:

  1. Connect the accounts. Add each channel once in Postiz.
  2. Create a post per slideshow. Paste the caption and hashtags from the script, and pick the channel. Check in the editor that your platform's slideshow or carousel format is offered for that channel.
  1. Pick the slot. Drag the post onto the calendar instead of typing times. One slideshow per account per day, and never two of the same format on the same day.
  2. Spread the variations. Put the five variations from step 7 on different days, ideally on different accounts, so the same hook is not in front of the same people twice in a row.
  3. Review the week. Open the calendar view and look for gaps, doubled-up days and any post with the wrong caption attached.

Here is an example:

Inoms - inline image
Inoms - inline image
Inoms - inline image
Inoms - inline image
Inoms - inline image
Inoms - inline image

Schedule the whole batch in one sitting, so for the rest of the week the only job left is replying to comments.

  1. Results: What the posts actually got

Scheduling the content was only half the experiment. The next step was seeing how the slideshows performed once they went live.

Here are the results from the six-post batch, including views and engagement across the posts.

Inoms - inline image

The numbers give you something concrete to work with. Instead of guessing which hook or story worked, you can compare the posts and see which ones earned more views, saves and shares.

One batch is not enough to establish a winning formula, but it gives you a starting point for the next one.

  1. Read the numbers and feed them back

A week after posting, pull the analytics for every post and give them to Fable 5.2, in the same project, so it still has the brief and the teardown in context. Paste the numbers as a table (or attach screenshots) and use this prompt:

Here are the stats for last week's [N] slideshows: views, likes, saves, shares, comments, and link clicks or installs where I have them. Each row also has its hook and its angle label.

Sort them into four groups:

  1. Winners: high views and high saves or shares
  2. Hooked but didn't convert: high views, low saves and clicks 3. Quiet but loved: low views, high saves
  3. Flops

For each group, tell me what the hooks and angles have in common. Then write a failure log: for every flop, one sentence on the most likely reason, based only on the data I gave you. If the data can't explain it, say so.

Finish with the brief for next week's batch: which format, which angles, and which 3 hooks to test.

Two habits make this work:

Saves and shares matter more than likes. A save means somebody wants to come back to it.

Keep the failure log as a file and paste it into the project instructions. The next batch gets written knowing what already flopped.

Then the loop starts again at step 2 with the winning format.

What you should still do yourself

The model does the typing. These parts should stay with you:

The app brief. Nobody knows the moment your app clicks better than you do. Reading every slide before it is scheduled. Wrong text, a seventh finger, a screenshot from the wrong screen. A quick review catches these, and they are expensive after posting.

Fact-checking any claim. If a slide states a number, the number has to be true. Fill any [NUMBER] placeholders yourself.

The hook. Claude gives you three options. Pick one and change the words until you would say it out loud.

Labeling AI content. Several platforms require you to mark AI-generated images. Turn that on every time and don't try to hide it.

Replying to comments. Real replies from a real person.

The kill and scale calls. Fable 5.2 sorts the posts. You decide what gets cut and what gets five more variations.

It also does not always work. Some batches will do almost nothing. When that happens it is usually a weak hook or a format that has been copied too much, and the failure log is how you find out which.

The whole system in one recap

  1. Save 10 slideshows that already worked in your niche and pick one format.
  2. Write an app brief once and keep it in a Claude Project with Fable 5.2.
  3. Have Fable 5.2 tear the reference apart: hook formula, text placement, where the app appears.
  4. Have it write your script in that format, with a hook you rewrite yourself.
  5. Generate the images (9:16, no text inside, reuse slide 1 as the reference), and use real app screenshots.
  6. Add the text in Canva or CapCut with one saved template so every slide is consistent.
  7. Ask for five variations that keep the mechanism and change the angle. 8. Schedule the batch on a calendar in Postiz.
  8. Pull the numbers a week later, sort them, write the failure log, and start again

The scheduling step is the one most people drop, and it is the one that turns a good afternoon of work into a full week of posts. That is what Postiz is for.

Try it: sign up at https://postiz.pro/inomsxbt, connect your accounts and schedule your first batch of slideshows.

二次创作

使用 YouMind 创作爆款文章

收集素材、拆解爆点、生成视觉资产、撰写内容,并在一个 AI 工作空间里完成分发。

了解 YouMind
写给创作者

把你的 Markdown 变成干净的 𝕏 文章

图片上传、表格、代码块,往 𝕏 上手动重排太痛苦。YouMind 把整篇 Markdown 一键转成干净、可直接发布的 𝕏 文章草稿。

试试 Markdown 转 𝕏

更多可拆解样本

近期爆款文章

探索更多爆款文章