Prompt Machine — visual model arena

one prompt → 2–8 models → 2–8 HTML renders, side by side · powered by aimlapi.com · v32

AIMLAPI key

Stored only in this browser (localStorage). Requests go directly from your browser to api.aimlapi.com.

Object / scene — a word or a description

Inserted into the shared prompt as {{OBJECT}} for every model.

Mode

Scene: static world, drone-orbit camera. Directed: one model writes a precise brief first, then every model builds that same brief.

Models 0 / 8 selected (min 2)

Loading model list…

Photo reference (optional)

Drop an image here or click to choose.
Attached to every model that supports vision; text-only models get the prompt without it (their card says so).

Advanced — shared prompt template & generation settings

Build prompt (pass 1, identical for every model; {{OBJECT}} is replaced with your object) Review prompt (passes 2+; {{BRIEF}}, {{HTML}} and {{RUNTIME}} are filled in automatically)

passes per model

1 = build only. 2+ = build, then review & correct. Each pass re-renders and re-bills.

max_tokens

Thinking models (Gemini, GPT-5.x) spend part of this on hidden reasoning — keep it high or output truncates.

temperature

parallel requests

AIMLAPI docs: 429 = rate/concurrency limit — queue requests.

reasoning effort (if supported)

Claude deep thinking (effort: high)
off = effort:low, the fastest setting — measured: sonnet-5 ~65 s, fable-5 ~2 min, opus-5 ~5 min (it thinks ~3 min before the first HTML). On = effort:high, considerably slower.