Lovable prompt generator: your prompt in five steps

    Step 1 of 5

    What are you building?

    No account needed.

    Pick what you are building, the feature it needs, the look and where your first users come from. You get a finished prompt in the same structure as the library, ready to paste into Lovable in Build mode.

    • Starts from a library prompt with a named style, the states, access rules and Done When checks
    • Adds the feature you need and a growth hook for your channel as a second prompt
    • The result opens at once; no account, no waiting on an AI call

    Building something the library does not cover? Describe your own idea to the AI

    Output checked against the prompt standard on

    Your own idea

    Describe your app to the AI generator

    For an app the builder does not cover, Claude writes the full prompt from your description. A free account includes 3 prompts.

    The output

    What does the Lovable prompt generator return?

    The same sections every library prompt carries, in this order.

    1. 01

      # Context

      What the app is, who it is for and the one job it does.

    2. 02

      ## Core Features (Priority Order)

      4 to 6 numbered features, so the most important one exists first.

    3. 03

      ## Visual Style & Design

      A named Lovable style direction, fonts, hex colours, a layout and components, so Lovable builds the look without asking.

    4. 04

      ## States

      Empty, loading and error for every list, form and fetch.

    5. 05

      ## Data & Access

      The tables and who may read and write each, enforced with row-level security.

    6. 06

      ## Done When

      Checks Lovable can verify in the preview after building.

    7. 07

      ## Build Order

      Data and access first, then the screens in feature order.

    8. 08

      ## Safe-Guard Instructions

      What must stay untouched and what never leaves the server.

    9. 09

      ## Growth Features

      One feature for the channel you pick, when you pick one.

    10. 10

      ## Before Building

      Lovable's own question line, so it asks before it writes code.

    Unedited output

    What does a generated Lovable prompt look like?

    One call on 23 October 2026 with the inputs below, pasted here as it came back: 721 words, 1546 output tokens from claude-sonnet-5.

    The inputs given to the generator for the example output
    App ideaA booking app for independent yoga teachers: students see the week's classes, book a spot and get a reminder; the teacher sees who is coming.
    Target userIndependent yoga teachers with 20 to 80 regular students
    ComplexityMedium
    Visual styleModern minimal (an API field the form does not show)
    ChannelOrganic SEO / AI Search

    Generated prompt, 23 October 2026

    # Context
    A booking app for independent yoga teachers with 20 to 80 regular students. Students see the week's classes, reserve a spot and get a reminder; the teacher sees who is coming and manages their schedule. Visitors arrive from search results for specific class times, locations and styles (e.g. "morning vinyasa class in [area]").
    
    ## Core Features (Priority Order)
    1. Teacher creates and edits a weekly class schedule (day, time, style, capacity, location).
    2. Students view the week's classes on a public page and book an open spot without creating an account first (email required to confirm).
    3. Automated email reminder sent ahead of each booked class.
    4. Teacher dashboard shows roster per class: who booked, waitlist if full, and ability to cancel a student's spot.
    5. Student can cancel their own booking via a link in their confirmation email.
    6. Teacher can mark a class as cancelled, which notifies all booked students.
    
    ## Visual Style & Design
    - Minimal and focused: whitespace-driven, neutral tones, quiet motion, subtle shadows, monochrome accents.
    - Heading font: Fraunces. Body font: Inter.
    - Background #FAFAF9, text #1C1C1A, accent #3B3A36 used only for primary actions.
    - Layout: single column for the weekly schedule, card grid for class listings.
    - Use shadcn/ui Card, Button, Dialog, Badge, Calendar, Table, Input, Toast.
    - Headline: "Book your spot on the mat." Subtext: "See this week's classes and reserve in under a minute." CTA: "View this week's schedule." Proof placeholder: "[Taught by Name, Studio or Location]".
    - Save these style keywords to the project knowledge.
    
    ## States
    - Weekly schedule: empty state "No classes scheduled this week", loading skeleton cards, error with retry button.
    - Booking form: loading spinner on submit, error keeps entered details and shows "Try again or contact the teacher directly".
    - Teacher roster: empty state "No bookings yet for this class", loading skeleton rows, error with retry.
    - Cancellation: confirmation step before cancelling, error message if link already used or class has passed.
    
    ## Data & Access
    - **teachers**: one row per teacher account, read/write by the owning teacher only.
    - **classes**: schedule entries (day, time, style, capacity, location, status); teacher can insert, update, delete their own; public can read.
    - **bookings**: student name, email, class reference, status; public can insert (insert-only); teacher can read and delete bookings for their own classes; students can delete their own booking using a unique link token, no login required.
    - **user_roles**: separate table holding teacher role, enforced with row-level security in the built-in backend.
    - Secrets (email sending keys) stay server-side only.
    - Signed-out visitors see the public weekly schedule and can book. Signed-in teachers see their dashboard, roster and schedule editor.
    
    ## Done When
    - A visitor can view the week's classes without logging in and complete a booking.
    - A booked student receives a confirmation and can cancel via their own link.
    - The teacher, signed in, sees the correct roster for each class and can cancel a student.
    - A second teacher account cannot see or edit the first teacher's classes or roster.
    - Cancelling a class triggers a notification to all currently booked students.
    
    ## Build Order
    1. Set up teachers, classes, bookings and user_roles tables with row-level security.
    2. Build teacher authentication and the schedule editor (create, edit, delete classes).
    3. Build the public weekly schedule page and booking flow.
    4. Build email confirmation and reminder sending, plus the student cancellation link.
    5. Build the teacher roster dashboard with cancel and waitlist handling.
    6. Add class cancellation flow with student notifications.
    
    ## Safe-Guard Instructions
    Do not alter the teacher's existing schedule data, booking records or roster logic when adding new features. Row-level security rules on classes, bookings and user_roles must not be loosened. Email sending keys and any server secrets must never be exposed to the client. The student cancellation link must remain accessible without login.
    
    ## Growth Features (Required for Distribution)
    Generate a unique, indexable public page for each class (e.g. /classes/vinyasa-tuesday-7am-[location]) showing style, time, location, teacher name and live spot availability, so search engines and AI answer engines can surface specific class times directly.
    
    ## Before Building
    Ask me any questions you need in order to fully understand what I want from this feature and how I envision it. Save my answers to the project knowledge so later prompts build on them.

    What to change before pasting: the bracketed teacher name and the proof placeholder, which the generator leaves for you because it never invents numbers. The growth feature (a public page per teacher with its own URL) is the part a generic prompt leaves out; the channel guide explains why it belongs in the first build.

    Audited

    Does the generator's output meet the library's prompt standard?

    Every section, yes. Length, not yet: with a growth section added it runs past the 550 words the standard sets for a library prompt.

    Three generator calls on 23 October 2026, each audited with the site's prompt audit script
    CallSystem prompt changeWordsDone When bulletsGaps found
    Call 1Standard system prompt8680Done When numbered, over 550 words
    Call 2Hard word ceiling added6960Done When numbered, over 550 words
    Call 3Done When as bullets, aim 450 words7215Over 550 words

    The audit checks the sections the standard requires: a numbered feature list, a named style direction with fonts, hex colours and a layout term, empty, loading and error states, access rules for stored data, 3 to 5 Done When bullets, a build order, safe-guards and the exact question line. Call 3 passed every one of them; its only gap is length. The prompting guide explains each section.

    Compared

    How does it compare with other Lovable prompt generators?

    The two other generators ranking for the term, read from their pages on 23 October 2026.

    Lovable prompt generators compared by inputs and output sections, read 23 October 2026
    This generatorRapidDevPromptSpace
    Sign-up to generateNone for the builder; free account for the AINoneNone
    InputsIdea, target user, complexity, app type, channelApp type, features, context, step modeApp type, stack, feature, style
    Example output on the pageYes, uneditedOne sampleNone
    Named Lovable style directionYesNoNo
    Empty, loading, error statesYesNoNo
    Access rules per tableYesRLS on every tableNo
    Checks Lovable can verifyDone WhenNoNo
    Lovable's question lineYesPauses between stepsNo
    Growth channelFirst prompt free, then the bundleNoNo

    RapidDev's tool asks for no account and puts schema and row-level security first, a sound choice; it leaves out the look, the states and checks Lovable can run. PromptSpace's form builds a general prompt from four menus. Sources: RapidDev generator, PromptSpace generator.

    In Lovable

    What happens when you paste a generated prompt into Lovable?

    The last section asks Lovable to put its questions first, so its first reply is a set of questions about your app.

    Lovable project chat right after a prompt was sent, showing a question card asking who will use the app and what its screens show, with an answer field, Skip all and Next
    The question line at work on a library prompt in the same structure, 26 September 2026. A generated prompt has not been run yet.

    Paste it into a new project in Build mode. Lovable reads the whole prompt, asks its questions, and saves your answers and the style keywords to the project knowledge, so later prompts keep the same look and rules.

    What it costs: I have run 3 library prompts in this structure, from 1.8 to 10.4 credits each, read from the balance before and after. A generated prompt has not been measured yet.

    After the first build, check each Done When line in the preview. A failed check is a one-line follow-up prompt that names it. The common mistakes guide covers what tends to break.

    Channel first

    Why pick a growth channel before you generate?

    Each channel needs different product features, so the channel changes the prompt itself.

    Growth channels the generator supports and what each asks the product to contain
    Organic SEO / AI SearchDesign your product so it naturally attracts traffic by creating pages that directly answer what users are already searching for and get picked up by Google and AI search engines.
    Paid SearchGet a prompt that optimizes your product for fast perceived value and clear conversion paths so ad clicks have a fighting chance to pay back.
    Organic SocialsShape your product around outputs people actually want to share, not because you ask them to, but because it makes them look smart or helpful.
    Paid SocialsStructure your product so its value is instantly understandable in a scroll, making it viable to acquire users even when attention is expensive.
    Short-Form VideoBuild your product around moments that can be shown, not explained, so it works naturally in short, visual formats like TikTok or Reels.
    Product-led GrowthCreate a product experience where users reach a meaningful 'aha moment' before they ever have to sign up, pay, or talk to sales.
    Community-led GrowthDesign your product so users benefit from contributing, sharing, or improving it together, causing value to compound over time.
    B2B Sales / Outreach / LinkedInGet a product structure that can be explained in one message and clearly ties features to business outcomes decision-makers care about.
    X (Twitter)Give the product a visible artifact worth posting, so building in public produces something to show, with nothing left to announce.
    Email MarketingBuild a reason for the product to send something a user actually wants to open, so the list works as a retention channel and stops being a broadcast list.

    The reasoning: product-channel fit.

    Generated or library

    Should you generate a prompt or start from the library?

    Frequently asked

    What makes this different from asking ChatGPT for a Lovable prompt?

    The structure. The generator writes the sections Lovable needs to build right the first time: a named style direction, the states, the access rules, checks it can verify, a build order and Lovable's own question line. A general chatbot tends to return a feature list.

    Do I need an account to use the Lovable prompt generator?

    Not for the prompt builder at the top: five steps return a finished prompt from the library, shaped by your answers, with no account. The AI generator below it needs a free account, which includes 3 prompts and is how the limit is tracked. Every prompt in the library can be read and copied without one.

    Which model writes the prompt?

    Claude, with a system prompt built from this site's prompt standard, which follows Lovable's prompting best practices and design guidance as read on 23 October 2026.

    What should I put in the app idea field?

    One or two sentences on what the app does and for whom. The target user changes the output more than any other field, because it decides which features come first.

    Why does it ask for a distribution channel?

    Because the channel decides features: an app found through search needs pages with their own URLs, one that spreads by invite needs the invite in the first build. The channel field is open on your first prompt and with the bundle.

    How long is a generated prompt?

    On the three calls audited on 23 October 2026, between about 700 and 870 words with a growth channel picked. That is longer than the 550 words the library standard sets, because the growth section and the full access rules add lines; every required section was present on the last call.

    Can I edit the prompt before pasting it into Lovable?

    Yes, and you should replace the bracketed placeholders: names, the proof line and anything you know better than the generator. Keep the Data & Access section and the Before Building line as they are; those are the parts that stop Lovable from guessing.

    Is there an unlimited version?

    The Complete Lovable Bundle includes unlimited generation, with every Pro prompt and build kit.

    What we did not publish

    • A Lovable run of a generated prompt; the screenshot and credit range come from library prompts in the same structure.
    • The output of call 2, which was not kept; its counts are from the audit log.
    • A comparison of output quality across generators; the table compares what each one asks and returns.

    Sources

    1. The generator's system prompt and three calls on 23 October 2026, audited with the site's prompt audit script
    2. Lovable docs: prompting best practices, read 23 October 2026
    3. RapidDev and PromptSpace generator pages, read 23 October 2026