AI Credits vs API Credits: What You're Actually Paying For (and Where Free Credits Really Come From)

Lucas Hayes — Tue, 27 Oct 2026 17:15:14 +0000

"Credits" is one of the most overloaded words in AI pricing. The same word gets used for a ChatGPT-style app allowance, a prepaid developer balance, a cloud promotion, and a startup grant. Mixing them up is how people end up buying the wrong plan or chasing "free credit" offers that don't exist.

Here is a plain breakdown of the main kinds, how they're metered, and how to find legitimate free ones.

1. App credits (consumer and prosumer tools)

Many image, video, and chat apps sell a monthly plan that includes a fixed number of credits. Each generation, upscale, or long-context request costs some of them. Key points:

They are tied to one product and usually can't be used anywhere else.
They often reset monthly and may not roll over.
The "cost per credit" varies by action, so compare plans by what you actually do (e.g. images per month), not by raw credit count.

2. API credits (developer billing)

API credits are a prepaid or granted dollar balance on a developer platform. Every API call is metered, usually per million input and output tokens for text models, and the cost is deducted from that balance.

This is completely separate from a chat subscription. Paying for a chat app does not give you API credits, and API credits don't unlock the chat app.

To estimate cost, you need two numbers per model: the input price and the output price per 1M tokens. A quick sanity check: a workload of 1M input tokens plus 250K output tokens costs input_price + 0.25 × output_price. Output tokens are often several times more expensive than input, so verbose responses add up fast.

3. Official vs third-party channels

The same model is often available both from the original provider and from third-party gateways and routers. Prices, rate limits, and terms can differ. Before you switch to a cheaper channel, check:

whether the listed price is per token at the same unit,
data handling and retention terms,
whether features you rely on (tools, caching, batch) are supported.

4. Where real free credits come from

Legitimate free credits generally come from a short list of sources:

Free tiers on consumer apps (limited daily or monthly usage).
Free or trial API usage on some developer platforms, which can change or end without much notice.
Cloud promotional credits for new accounts.
Startup programs, typically via accelerators, VCs, or cloud partner programs, with eligibility rules.

Be skeptical of anything promising a fixed signup bonus from an unofficial site. Always apply on the provider's own pages.

If you'd rather not track all of this yourself, directories such as AI Credits collect free tiers, startup credit programs, and official-versus-third-party API prices in one place and link out to the official pages.

5. A simple decision checklist

Are you using a product (chat, image, video app) or building with an API? That decides app credits vs API credits.
Estimate monthly usage in the unit that's actually billed (images, seconds, or tokens).
Compare at least the official price and one alternative channel.
Check free tiers and startup programs before paying, but verify every offer at the source.

Understanding which kind of credit you're dealing with usually saves more money than any discount code.

How to Get a Table Out of a Photo or PDF and Into Excel Without Retyping It

Lucas Hayes — Tue, 27 Oct 2026 17:08:06 +0000

If you have ever stared at a photo of a printed price list, a scanned bank statement, or a dashboard screenshot and thought "I just need this in a spreadsheet," you know the usual options are bad. You either retype it cell by cell, or you run it through an OCR tool that hands back one long block of text, and then you spend twenty minutes splitting it into columns.

This post walks through a workflow that actually saves time, and, just as important, how to check the result so a misread number doesn't end up in a report.

1. Start with the best source you can get

OCR quality is decided before you upload anything. A few habits make a big difference:

Prefer the original file. If the table came from a PDF that has selectable text, use the PDF, not a screenshot of it.
Shoot straight on. When photographing paper, hold the phone parallel to the page and avoid shadows across the table. Skewed photos are the most common cause of columns drifting into each other.
Crop to the table. Headers, logos, and handwritten notes around the table can confuse table detection. Crop tightly if you can.
Keep the resolution. Messaging apps often compress images. Send the file to yourself as a document instead of a photo.

2. Use a tool that rebuilds the table, not just the text

Plain OCR answers "what characters are on this page?" What you need is "which characters belong to which row and column?" That second step, table structure detection, is what turns a picture into something you can sort and filter.

Tools built for this, such as ImageToExcel, detect the grid, keep rows and columns together, and export straight to an .xlsx file. They usually accept JPG, PNG, WEBP and PDF input, including multi-page scans, so the same flow works for a phone photo and a 12-page statement.

3. Review before you export

Never trust OCR output blindly, especially with financial data. A quick review routine:

Check the totals. If the source has a total row, add the extracted column yourself with =SUM() and compare. A mismatch tells you immediately that something was misread.
Look for classic swaps. 0 and O, 1 and l, 5 and S, and , versus . as decimal separators are the usual suspects.
Count the rows. Compare the row count against the source. Merged or dropped rows are easier to spot by count than by eye.
Watch merged cells. Headers that span several columns in the original often get flattened. Fix those by hand before you build formulas on top of them.

4. Clean up in Excel

Once the data is in a sheet, a few built-in features finish the job:

Text to Columns (Data tab) to split anything that landed in one cell.
TRIM and CLEAN to remove stray spaces and non-printing characters that break lookups.
Format as Table (Ctrl+T) so filters and structured references work right away.
VALUE() to turn numbers that were read as text back into numbers.

5. Specific cases worth knowing

Bank statements and credit card statements: dates and amounts matter most. Sort by date after export to catch rows that were read out of order.
Invoices and receipts: you usually only need a handful of fields (date, vendor, line items, total). Extract those into a consistent layout so you can stack many documents into one sheet.
Screenshots of web dashboards: these are usually clean, but check that negative numbers kept their minus sign or parentheses.

Wrapping up

Getting a table from an image into Excel is a two-part job: good extraction, then a fast, deliberate check. Spend thirty seconds on the source, let a table-aware converter do the heavy lifting, and always reconcile at least one total before you rely on the numbers.

DEV Community: Lucas Hayes