Keep a prompt log: your own record of how each AI image was made
By The NeutronField editorsPublished Updated
Short answer
A prompt log is a simple table you keep yourself: the date, the tool and model version, the full prompt, the settings and seed, which frame you kept, and a SHA-256 hash of the original file. The file itself rarely stores your prompt, so the log is often the only record of how an image was made.
What to record
Write down what you would need to make the image again, and to show someone how it was made. That means the date, the tool and how you reached it (app, website or API), the model and its version, the full prompt, any settings such as aspect ratio, seed and guidance, how many frames you made, and which one you kept.
The tools will not do this for you. None of the files we have tested from Google or OpenAI stores the prompt, and the chat models we used gave us no seed to record. A content credential notes the tool and the steps, not your words. Our four layers page shows exactly what three real files held.
| Field | Value |
|---|---|
| Date and time | 2026-11-02, 04:28 UTC |
| Tool and route | Gemini API, through our own media library |
| Model and version | gemini-3.1-flash-image (Nano Banana 2) |
| Prompt | Overhead photograph of a ceramicist's hands shaping a wet clay bowl on a pottery wheel... (full text kept) |
| Settings | Aspect 3:2; no seed offered |
| Frames made, frame kept | 11, 11A, 12, 12A; kept 12 |
| File kept | JPEG, 1264 by 848, 821,978 bytes |
| SHA-256 of the file | bf647d6c...9211c79 (64 hex characters in full) |
| Records in the file | Google-signed content credential; no prompt |
| Edits after | Resized to WebP for the web; original kept unchanged |
Hash the file
A hash is a short fingerprint of a file. SHA-256, defined in the US federal standard FIPS 180-4, turns any file into 64 hexadecimal characters, and changing a single byte of the file changes the result completely (NIST FIPS 180-4). Recording the hash of the original download lets you prove later that a file is the one you logged, even after copies have been resized or re-saved.
You do not need special software. On a Mac or Linux terminal run shasum -a 256 yourfile.jpg; in Windows PowerShell run Get-FileHash yourfile.jpg, which uses SHA-256 by default. Paste the result into your log next to the prompt.
This is the same idea a content credential uses: the C2PA standard stores a hash of the image inside the signed record, so a viewer can tell whether the pixels changed after signing (C2PA specification). Your log is the unsigned, private version of that record, and it covers the parts a credential leaves out.
Where to keep it
A spreadsheet works. Keep one row per kept image, store the original files in one folder named by date, and never overwrite an original: export copies for the web instead. Back the folder up somewhere other than the computer you generate on.
If you work with clients, a shared sheet with read-only access gives them the same record you have. If you sell images, keep the log for as long as the image is on sale.
When a log helps
A log answers the questions people actually ask: how was this made, can you make another like it, and is this the file you delivered. Stock sites and marketplaces increasingly ask you to declare AI use (see selling AI images), and a dated log makes that declaration easy to back up.
It also helps you. Writing down the prompt and the frame you kept is the fastest way to learn which words changed the picture, the same method we use for every example on this site, described on our About page. More guides on labels and records are in Made with AI.
A missing label does not mean an image is human-made.
Sources
Questions
What is a prompt log?
A record you keep yourself of how each AI image was made: date, tool, model version, prompt, settings, the frame you kept, and a hash of the original file.
Why log prompts if the file has a content credential?
Because the credentials we have read record the tool and the steps, not the prompt, and they disappear when a file is re-saved. Your log keeps both.
How do I get a file hash?
Run shasum -a 256 on a Mac or Linux, or Get-FileHash in Windows PowerShell. Both give a 64-character SHA-256 value to paste into your log.
Should I log rejected frames?
Log how many you made and which one you kept. Keeping the rejects is optional, but noting them shows how often a prompt works.
Related reading
- Made with AIWhat your AI image carries: credentials, watermarks, metadata and the prompt itselfAn AI-generated image file can carry four records: a signed content credential, an invisible watermark in the pixels, ordinary metadata fields, and the prompt saved as text.
- Made with AISelling AI images: what stock sites and marketplaces require you to declareSome marketplaces take AI images and some refuse them.
- PromptsHow AI image models turn a prompt into a picture (and why the picture carries traces)An image model turns your prompt into numbers, starts from random noise, and removes that noise step by step, steering each step towards the prompt.
- PromptsPrompts for ChatGPT and Gemini images: talking to a chat-based image modelWrite to ChatGPT and Gemini in plain sentences, not tag lists: say what the picture shows, where, in what light and medium, and the one detail that must be right.