POWERED BY JEV (TYPESAFE/JEV) · FREE · NO SIGN-IN
Jev AI Prompt Evaluator: Score a Video Prompt Before You Spend Credits
A weak prompt is the most expensive mistake in AI video: you pay full credits for a clip that misses the shot. This Jev AI prompt evaluator checks your prompt first. It runs on Jev, TypeSafe's structured evaluation model, which answers typed questions about your text with calibrated probabilities instead of guesswork.
Paste a prompt, get four readings in about a second — usability, main focus, detail level and camera direction — plus concrete fixes for what is missing. Free, no account, unlimited iteration while you sharpen the prompt.
Free · no sign-in · scores in about a second
How Jev evaluates your prompt
Jev answers three kinds of typed questions — Noul, Choice and Score — and returns probabilities and confidence for each. The evaluator above asks it four questions about your prompt.
Noul: yes/no with a number
Instead of a plain true or false, Jev returns a calibrated probability between 0 and 1. This evaluator uses it to answer “is this text usable as a video prompt?” and “does it include camera direction?”.
Choice: classification with probabilities
Jev picks the best answer from the options you define and returns the probability of every option. Here it classifies what your prompt mainly describes — a subject, an action, a scene or a style.
Score: graded on a scale
Jev scores the prompt against a legend — Sparse, Adequate, Rich — and shows how confident it is in that grade, so you know when to trust the number.
Built for evaluation, not chat
Jev (provider model jev-latest) is a dedicated structured evaluation model with a 32,000-token context. It does one thing: answer typed questions about a state with calibrated confidence instead of free-form prose.
The same integration as our 22 video models
FlowAIVideo runs every third-party model through one Cloudflare Workers AI binding — Hailuo 2.3, Veo 3.1, Seedance 2.5, and now Jev. The API surface is identical: we callenv.AI.run("typesafe/jev", { state, questions })with the prompt as the state and our four typed questions. No API token, no separate vendor account — the same pattern that powers the video generators on this site. When your prompt scores well, take it straight to the AI video generator or the AI dance generator and render it.
Frequently asked questions
What is the Jev AI prompt evaluator?
A free tool that scores your video prompt before you spend generation credits on it. Paste a prompt and the Jev AI prompt evaluator returns four readings: whether the text works as a video prompt, what it mainly describes, how detailed it is, and whether it includes camera direction — each with Jev's calibrated confidence.
Which model powers this evaluator?
Jev, by TypeSafe — the structured evaluation model exposed on Cloudflare Workers AI as typesafe/jev (provider model jev-latest). We call it through the same Workers AI binding we use for every third-party video model on this site; only the model ID and the typed questions change.
What are Noul, Choice and Score answers?
They are Jev's three answer types. Noul returns a probability between 0 and 1. Choice picks one option from a list you define and returns the probability of each. Score grades the state against a legend, such as Sparse / Adequate / Rich. All three come with confidence values, so weak judgments are visible as weak.
Is the Jev AI prompt evaluator free to use?
Yes — no sign-in and no credits. Jev evaluates a prompt for a fraction of a cent in input tokens, so the tool is free to run as often as you like while you iterate on a prompt (with a light rate limit to stop abuse).
Does it work for prompts other than video prompts?
The questions on this page are tuned for text-to-video prompts, but Jev itself evaluates any state against any typed questions — support triage, refund reviews, risk scoring. If you can phrase it as typed questions, Jev can answer it.
Last updated September 20, 2026