Four free AI app builders: the short answer

All four produced a playable Synth Drift game. Lovable had our preferred visual finish. Base44 added working logins, saved runs and private history without a functional repair. Bolt needed a third prompt after its backend change broke the game scene. Hercules passed the recorded player checks, but its free code view was read-only. A working demo is not proof that an app is safe for real users.

One game, the same two prompts

The first prompt asked for a synthwave endless runner with three lanes, a hovercar, obstacles, collectible orbs, rising speed, score, distance and a restart screen. No backend or login in round one.

The follow-up asked for email/password accounts, saved scores and distances, a global top ten and private player history. It required access rules on the server, not just hidden controls. Every builder received the same requests. Repair prompts are identified separately.

What the first game build showed

Lovable produced the strongest visual finish in our judgment. Bolt and Hercules felt more like arcade games. Hercules had a working score and game-over screen. Base44 ran, but its initial bloom made the road hard to see. A later cosmetic prompt reduced that glow; it was not a repair to broken gameplay.

We do not rank the tools by recorded build time. Waiting for a preview does not tell you whether the result follows the brief.

Lovable: strong finish, an incomplete History check

Lovable added its Cloud database and sign-in. A saved best score survived reload. A second player saw their own best score, but the game-over overlay blocked the History button. That route was not verified.

The recorded route offered repository sync. The free build allowance is small, so leave room for follow-up changes. The linked review separates build credits from Cloud and app-AI allowances.

Base44: working backend, platform lock-in

Base44 added auth, a database and a server-side rule on the runs table. The recorded checks showed registration, a saved run, the global top ten and private history surviving reload. A second player saw both scores on the leaderboard but only their own history. Outside users had to use the published app to log in.

The cosmetic bloom change was separate from the backend request. The October 15-16 recording showed 25 message credits a month with a daily cap of five. Code was readable but not editable on free. Export and GitHub routes were paid. These are dated limits, not a promise that the current plan is unchanged.

Bolt: the backend change needed a repair

After the backend prompt, the start screen appeared but the three-dimensional scene did not. The space bar did nothing. A third prompt brought the game back; saving and the top ten then worked.

The second-user check was not completed on camera, so user isolation remains unverified. The recorded route offered a project ZIP and GitHub connection. The daily token budget limits how much repair work fits in one session.

Hercules: qualitative workflow coverage

The recorded backend checks showed both players on the global leaderboard and each player seeing their own runs. A saved best survived reload. The leaderboard used the generic name Player for every entry.

The free code view was read-only, and we did not find an export in this test. This is an observed portability limit, not proof that Hercules owns the code. The recorded account also showed a larger starting balance than its stated monthly allowance; do not expect that balance on every signup.

Hercules is in the free-plan table, but its recorded times stay out of any timing comparison. Its linked free-plan overview covers personal, noncommercial use and the dated workflow limits.

Free plans: credits are not interchangeable

Lovable uses build credits, Bolt uses tokens, and Base44 has message and integration credits. Hercules showed a test-account balance that was not a standard signup promise. One credit in one tool is not the same amount of work as one credit in another.

The table uses facts from the linked individual reviews, not a normalized cost benchmark. Those reviews were recorded on different dates. Check their dated plan notes and current terms before starting.

Before you invite real users

Check login and logout, saved runs across sessions, the global leaderboard and access to another player history. Test database access rules directly. Check phone controls, error recovery, hosting costs and a route to export or maintain the project.

None of the small recorded checks is a full security audit. Free output and commercial-use rights are separate questions. FreeStack may earn a commission from affiliate links at no extra cost to you.