How It Works & Measured Data

How every tool works

The text tools (X profile diagnosis, SNS post generator, landing-page and ad copy, text summarizer/proofreader) work in this order.

1. 01

   Input

   What you enter in the browser is sent to our server.

2. 02

   Generated by AI

   The server sends it to an outside AI service (OpenAI or Google Gemini) to produce the result. The default is a fast model; paid plans can choose a higher-end model.

3. 03

   Shown to you

   The result is displayed. What you entered isn't stored in our own database.

4. 04

   When AI isn't available

   If the AI service can't be used, we make a basic result with fixed rules that don't use AI and label it "Simple mode".

5. 05

   Usage limits

   The free plan allows up to 3 uses per tool per day. Besides the count in your browser, the server also checks a limit per connection.

Image compression and conversion are the only exception. Images are processed entirely in your browser and are never sent to our server or to AI.

How the X profile score is calculated

A diagnosis can take one of two paths, and the score is decided differently on each.

Normal: scored by AI

We give the AI your bio and recent posts and have it score the account's follower-growth and engagement potential out of 100. The instructions to the AI include these rules.

• Make the score consistent with the number and severity of the weaknesses listed (don't hand out extreme highs or lows lightly).
• Never invent follower counts, achievements or third-party opinions — base the score only on the wording and structure of the text.
• Internally, think in the order current state → blunt feedback → improvements, and output only the finished result.

Because the AI does the scoring, the same input won't always give exactly the same score.

Simple mode: a fixed formula

When AI isn't available, the score is calculated like this: start from a base of 50, add or subtract points for five checks, and keep the result within 0–100.

The highest this formula can give is 85; the lowest is 0. It only looks at surface features such as length and symbols, so it can't judge how good the content is — treat simple-mode scores as a rough guide. Note that "👇" counts both as an emoji and as a call-to-action word.

Score bands

Neither score is an official X metric, and neither guarantees that your follower count will grow.

Measured image-compression data

Image compression and conversion run in the browser (Canvas). We measured how much smaller WebP and JPEG actually get, using 24 standard test photographs.

How we measured

• Images: the Kodak Lossless True Color Image Suite (24 PNG photos, 768×512 or 512×768, 14.7 MB in total), a standard set widely used in compression experiments. We don't redistribute it.
• Processing: the same steps as the tool (load → draw to a canvas → encode at the chosen quality), at the original size with no downscaling.
• Environment: Chromium 152.0.7977.130 (Electron desktop shell; macOS), 2026-10-12. Other browsers or versions encode differently and will give different numbers.
• Quality settings: six steps from 0.5 to 0.95. The tool's default is WebP at 80% (0.8).
• Quality metric: PSNR (dB) — higher means closer to the original. It is not a measure of how it looks to a person.
• Verification: converting kodim01 with the actual tool gave 93,010 bytes (WebP) and 104,936 bytes (JPEG), identical to the measured values. We ran the whole measurement twice and all 288 results were identical.

Result 1: by quality setting

Averages over 24 images. The average original PNG is 626.4 KB. "% of PNG" is the converted size as a share of the original PNG.

Result 2: compared at the same image quality

A quality setting such as 0.8 means different things for each format, so comparing the same number isn't fair. Instead we compared sizes at the same PSNR.

At every quality level WebP was about 66–69% the size of JPEG. Rows with fewer than 24 images leave out images whose measured range didn't reach that quality level (we don't extrapolate).

What this shows

• At the default setting (WebP, 80%) files came out at about 9.1% of the original PNG (58.8 KB on average). JPEG at 80% gave about 11.7% (74.8 KB).
• At the same image quality (PSNR), WebP was roughly a third smaller than JPEG.
• Raising WebP quality from 80% to 95% grew the average size from 58.8 KB to 140.9 KB (2.4×). Size climbs steeply as quality goes up.

Limits of this measurement

• Only photographs (24 Kodak images) were measured. Screenshots, logos and illustrations — images with large flat areas — behave very differently.
• Images that are already JPEG or WebP, and PNGs that aren't photographs, can get larger when converted. The tool says so when that happens.
• AVIF was not measured: the browser used can't encode AVIF through Canvas, and the tool doesn't offer AVIF in such a browser either.
• PSNR is a rough guide, not a judgment of how it looks. Only one browser and one version were measured.

Bugs found along the way

While measuring, we found and fixed three bugs in the image tool: transparent areas turned black when a PNG was converted to JPEG; a file that got larger still showed "0% smaller"; and switching the format afterwards changed the extension of an already-converted file's download name.