PhotoLex · Subscription help
Which plan fits your workload?
The answer depends on one count, not on your archive. Here is why, what we measured, and a gauge that sizes a month for you.
Tokens do not depend on your files. They depend on how many photographs you send, and for what.
Whatever the original is — a 100 MB Fujifilm RAF, a 4 MB JPEG, a 6000 × 4000 DNG or a scan — PhotoLex sends the AI provider the same thing: a JPEG resized to 2048 pixels on the long edge, the embedded metadata as a short list, and the same instructions every time. The image is worth a fixed number of tokens at that size, and the instructions never change. So file format, sensor size, folder structure and gigabytes on disk make no difference. Your workload is a count of photographs per month, split by what you ask PhotoLex to do with each.
We measured this across real shoots — RAW and DNG from two cameras — and the tokens per photograph varied by less than 8%. The only thing that moved was the length of the metadata.
Size a month
Count your photographs in two bins and switch off any workflow you do not use. Most photographs are keyworded and never captioned, so a keywording-only month is the common case. Everything else PhotoLex does — cataloguing, measuring, grouping, culling, sidecars, metadata, vocabulary, claim review, approval, delivery — happens on your Mac and uses no tokens at all.
What one photograph costs
Keywording — one request per photograph
PhotoLex sends its keywording instructions, your approved vocabulary, the photograph's metadata and the image, and gets back a title, a description, alt text and 5–20 keywords from your vocabulary. A large custom vocabulary adds roughly 6 tokens per term.
| Provider and model | Tokens in | Tokens out |
|---|---|---|
| OpenAI gpt-5.6-terra (default) | ~3,000 | ~300 |
| OpenAI gpt-5.6-sol | ~3,600 | ~200 |
| Anthropic claude-sonnet-4-5 | ~3,700 | ~300 |
| Google gemini-3.5-flash | ~3,000 | ~300 |
Call it 4,000 tokens per photograph keyworded, with any provider.
Captioning — one request per photograph, sometimes two
PhotoLex sends its editorial instructions, the metadata, the context you typed about the photograph, and the image, and gets back a publication-ready caption with a claim-by-claim audit: every claim, its source, what could not be established and the questions worth asking you. Two settings change the bill by a large factor:
- Web research, on by default. The provider's search results come back as part of the request — commonly 3,000–10,000 tokens per search, and up to five searches. Expect 10,000–25,000 tokens in for a researched caption.
- Reasoning effort, High by default with OpenAI. The provider's reasoning is billed as output and never appears in the caption; on an editorial task it commonly runs 2,000–8,000 tokens.
| Caption | Tokens in | Tokens out |
|---|---|---|
| No research, no reasoning (Gemini and Anthropic defaults) | ~4,500 | ~1,000 |
| No research, OpenAI High reasoning | ~4,500 | ~3,000–9,000 |
| With research, any provider | ~10,000–25,000 | as above, plus the search itself |
Call it 6,000 tokens for a plain caption and 25,000 for a researched one, and plan on the second, because the defaults research.
Three sizes of month
Priced at gpt-5.6-sol with research and High reasoning on, so these are upper estimates. Turning research off cuts a caption's tokens by three-quarters; keywording with terra or Gemini halves that line.
| Workload | Keyworded | Captioned | Tokens | List-rate cost |
|---|---|---|---|---|
| A shoot a month — keyword the keepers, caption the few that go out | 100 | 20 | 0.9M | ~$8 |
| Working photographer — several shoots, a caption for every delivery | 500 | 150 | 5.8M | ~$50 |
| Archive — keywording the backlog while delivering weekly | 2,000 | 400 | 18M | ~$160 |
A plan is described in the two counts — photographs keyworded and photographs captioned per month — because those are what you can count. The token allowance is the small print behind them.
Keyword figures are measured. Caption figures are estimates from the exact requests PhotoLex sends and each provider's published token accounting; PhotoLex will show you your own numbers as you use it.