ImageX / Services / Create from words / create.describe

Describe

Upload an image, get prompts that would recreate it

Engine ready Not in the app yet — the engine route is listed below so you can see exactly what it needs.

Free diver, cathedral of light — made with ImageX — A free diver descending into a deep blue ocean cenote, god rays piercing the water from above, tiny silhouette…
**Z-Image Turbo**

Night market, Taipei — made with ImageX — A busy night market street in Taipei after rain, red lanterns reflecting on wet asphalt, steam rising from foo…
**Z-Image Turbo**

Red door in the dunes — made with ImageX — A single bright red door standing alone in vast pale sand dunes at dawn, long shadows, surreal minimalist phot…
**Z-Image Turbo**

Dancer in red — made with ImageX — A contemporary dancer in a flowing vermilion dress mid-leap in an empty white gallery, motion blur on the fabr…
**Z-Image Turbo**

Chromatic split — made with ImageX — Abstract composition of a beam of white light splitting through a glass prism into red, green and blue bands o…
**FLUX.2 klein**

Harbour in watercolour — made with ImageX — A watercolour painting of a small Cornish harbour with fishing boats, soft washes, textured cold-press paper, …
**Z-Image Turbo**

Isometric workshop — made with ImageX — An isometric miniature diorama of a tiny photography darkroom with red safelight, trays and hanging prints, 3D…
**Z-Image Turbo**

01 — How it works

Three steps.

01Describe it

Image

02We route it

Auto sends it to the engine best at create from words and shows you which one and why.

03Perfect & download

Vary, fill, upscale 4×, remove the background, then download — with the AI label in the file.

02 — What you get

The spec.

DeliverableUpload an image, get prompts that would recreate it

Output4 prompts plus tags

Price4 credits per picture (≈ £0.04)

StatusEngine ready plan: Ready (Image Studio describe)

Engine routeL qwen2.5vl:7b / Qwen3-VL in Ollama; Claude vision

FormatsPNG with the AI label · transparent PNG · SVG where it applies

03 — Questions

About describe.

Q01How long does describe take?

Most pictures develop in 30–60 seconds on our own GPUs; edits that keep your composition take a few minutes. The app shows your place in the queue and a live estimate.

Q02Can I use the result commercially?

On any paid plan, yes, subject to the content policy. Free-plan pictures are public and CC BY-NC 4.0. The engine and its licence are recorded with every picture.

Q03What if I don't like it?

Press Vary for new takes, or Fill with a brush to change just one part. A failed render is refunded automatically.

Q04Which engine makes it?

Auto routing picks it for you; today that's qwen2.5vl:7b / Qwen3-VL in Ollama; Claude vision. You can override it.

Q05Do my uploads leave your servers?

No — the default engines run on our own GPUs. Uploads are deleted after 30 days unless you save them.

04 — Related

You might also need.

→Text to image4 images from a description4 crIn the app →PhotorealPhotographic images indistinguishable from a camera4 crIn the app →Text in imagePosters, signs and slogans with correctly spelled words4 crIn the app →IllustrationFlat, editorial, children's-book or vector-look illustration4 crIn the app