Make a browser game with AI and publish it on CrazyGames and Playgama

Both portals tell you what they measure. CrazyGames moves a game from its test phase to a full release on four numbers: average playtime, day-1 retention, conversion to gameplay and, for games that start Basic Launch on or after 23 November 2026, thumbnail click-through rate [CrazyGames]. Playgama's own guidance asks for clear goals, early wins and "at least 10 minutes of distinct gameplay before the game starts repeating itself" [Playgama].

That rules a lot out before you write a line. A game that needs a lobby of other players, a story that takes five minutes to start, or a big world to explore loses the first minute, and the first minute is what the numbers see. A game with one verb you learn in ten seconds and master over ten minutes keeps it.

What both portals will not take

• Copies and brands. Playgama refuses games that copy another game in its catalogue, duplicate one of your own games, or use other people's names, logos, music or characters [Playgama]. CrazyGames asks that names, assets and content "exhibit originality" [CrazyGames].
• Real money. No gambling, lotteries or real-world prizes [Playgama]. CrazyGames requires PEGI 12 content [CrazyGames].
• Many small variations. CrazyGames lets you resubmit only "if you've made clear improvements" [CrazyGames], and a game whose test metrics fall short can only come back as a new game with significant improvements [CrazyGames]. Sending ten quick reskins is how an account gets a reputation, not a hit.

The same files are on GitHub, so a coding assistant can fetch them by URL; the list with sizes is in agent.md. Repos: characters, animals, toy cars. If your game is a shop or a room, the walkable store guide uses six free store props.

Bake them before you ship. We ran the command from the prompt on the free knight and zombie: 410 KB became 171 KB and 344 KB became 161 KB, and all ten clips still played in three.js r170. The clip at the top of this page is that test.

A note on AI and art. Playgama accepts AI-generated assets but not games built entirely from them, and says AI-generated games currently perform "more than 2x worse on metrics" than titles made without AI graphics [Playgama]. That is the platform's own statement, not our measurement. Hand-made models with a consistent style are one way to stay clear of it; your own art direction on top is the other.

Claude Code (or any coding assistant) is good at the plumbing of a three.js game: the loader, the animation mixer, input for keyboard and touch, collisions on a flat arena, a wave timer. It is not good at knowing whether the game is fun. Play every build yourself, on a phone too, and give it short, concrete feedback ("zombies reach me before I can see them", not "make it better").

Three rules that save a rewrite

• Bundle three.js. Playgama requires that the game "loads no external resources" and ships everything in the archive [Playgama]. A CDN import map works on your laptop and fails review. Install three from npm and build with Vite. (The two portal SDK scripts are the exception: the portals tell you to load those from their own URLs [CrazyGames] [Playgama].)
• Relative paths. CrazyGames allows relative paths only [CrazyGames]; set base: './' in the Vite config.
• Real sound. Playgama does not accept sound synthesized in code, "such as procedural Web Audio tones generated by an AI coding agent"; it wants recorded or produced audio files, music and a mute button [Playgama]. Plan an afternoon for sound.

What we measured. three.js r170 with GLTFLoader, the meshopt decoder and SkeletonUtils, minified with esbuild: 756 KB, 195 KB gzipped, 160 KB with Brotli. Add the two baked characters and the whole first scene is about half a megabyte over the wire.

Both portals require their SDK for a full release. CrazyGames needs it for Full Launch (it is optional, with ads switched off, during Basic Launch) [CrazyGames]; Playgama makes the Bridge "a mandatory condition for uploading a game" [Playgama]. The calls are different, the rules are close:

The ad rules, in short

• No ads before the player has had "a reasonable amount of gameplay", never during play, never chained [CrazyGames]. Playgama: only at logical pauses, and the Bridge "holds back the first one" [Playgama].
• Rewarded ads are opt-in, the button says that an ad will play and what it gives, the "no" option is just as visible, and a revive is offered at most once per attempt [Playgama] [CrazyGames].
• If an ad fails or an ad blocker is on, the game carries on normally; players with ad blockers may not be blocked [CrazyGames].
• During CrazyGames Basic Launch ads are off, so a rewarded button that does nothing must not be shown; portal.canReward() is where that goes [CrazyGames].

portal.js (99 lines; the ad part below)

We ran this file on 127.0.0.1 in three modes on 13 November 2026: without an SDK, with the CrazyGames SDK v3 in its local test mode (break ad, rewarded ad and data module all answered, sound and pause toggled around each ad) and with the Playgama Bridge, which uses a mock platform outside a portal. It has not been through either portal's review; test it in their preview and QA tools before you submit.

  // natural pauses only (level end, game over). The SDKs decide whether an ad really shows.
  // CrazyGames: resolves after the ad or right away when none is served. Playgama: resolves at once;
  // if an ad opens, the state events above pause and mute the game until it closes.
  breakAd() {
    return new Promise((done) => {
      if (kind === 'crazygames') {
        CG().ad.requestAd('midgame', { adStarted: adOpen, adFinished: () => { adShut(); done(); }, adError: () => { adShut(); done(); } });
      } else {
        if (kind === 'playgama') PG().advertisement.showInterstitial();
        done();
      }
    });
  },

  // only show the "Watch an ad: revive" button when this is true
  canReward() { return kind === 'crazygames' || (kind === 'playgama' && PG().advertisement.isRewardedSupported); },
  // resolves true only when the ad was watched to the end: grant the reward then, never on an error
  rewardAd() {
    return new Promise((done) => {
      if (kind === 'crazygames') {
        CG().ad.requestAd('rewarded', { adStarted: adOpen, adFinished: () => { adShut(); done(true); }, adError: () => { adShut(); done(false); } });
      } else if (kind === 'playgama') {
        rewardGot = false; rewardDone = done;
        PG().advertisement.showRewarded();
      } else done(false);
    });
  },

Download the whole portal.js (init, language, loading, saves) or copy it from agent.md.

One Playgama detail that cost us a console error: the Bridge loads playgama-bridge-config.json from next to index.html and logs an error without it. An empty {} was enough for the test; the docs have a config editor for real placements.

Playgama's docs also say a game with the Bridge can be uploaded to CrazyGames directly [Playgama]. We use two builds with the portals' own scripts instead, because each portal's own SDK is what its review checks.

CrazyGames asks that games "land new users in gameplay immediately", with "a maximum of 1 click" if that is not possible [CrazyGames]; its conversion metric is the share of players who play "for at least one minute after starting the game", and top titles convert 80% or more [CrazyGames]. Playgama gives the same advice in more words: one click from the loading screen at most, ideally zero; no name, sign-in or options before playing; teach the core mechanic "in the first 30 to 60 seconds through play" [Playgama].

• Controls where the thumb is. Show the hint next to the player character or on the joystick, for the current device: keys on desktop, a finger on touch. Hide it after the first move.
• An early win. The first wave should be hard to lose. A score pop and the cheer clip within the first 15 seconds tell the player the game is working.
• Language. English is required on both portals; take the player's language from the SDK (portal.language()) if you translate [CrazyGames] [Playgama].
• Save at once. A returning player who lands in a tutorial again is a lost day-1 retention point. Save after every meaningful step through portal.save.

CrazyGames measures load time and size up to the first gameplayStart call [CrazyGames], so whatever the first scene does not need should load after it. For a three.js game that means: bake every GLB (step 2), compress textures to WebP, load later levels and music in the background once play has started, and keep the device pixel ratio at 2 or lower on phones.

Test the slow case, not your desktop: Chrome DevTools with a 4x CPU slowdown and a throttled network is a cheap first check. It is not a real phone; try one before you submit.

CrazyGames

You upload in the developer portal and can test in their preview tool first [CrazyGames]. Every game then starts with Basic Launch: a limited audience for 7 to 21 days, with video ads, banners and in-game purchases switched off. It ends once the game has been live 7 days with at least 500 plays, or after 21 days. If all metrics meet the benchmarks, you are invited to add the full SDK integration and submit for Full Launch: global release with monetization and a revenue share. If some do, you may be invited to improve and try again; if most do not, the game cannot proceed [CrazyGames]. New games may appear in a new-games carousel on the homepage; after that, placement "depends on ongoing performance" [CrazyGames].

Playgama

You upload a ZIP with index.html at the root in the developer account, fill in title, description, how to play, devices, orientation and languages, test in their QA tool and submit [Playgama]. Moderation "usually takes 3-5 business days" on Playgama; partner platforms can take up to 2 weeks [Playgama]. The Bridge targets many platforms at once; the docs list, among others, MSN, Facebook, YouTube, Discord, Huawei, Xiaomi, Samsung and Yandex Games [Playgama], and Playgama says it reaches "450+ million users monthly" [Playgama] (its own figure).

Both at once?

Possible: CrazyGames says publishing elsewhere does not affect revenue share eligibility as long as you hold the distribution rights [CrazyGames], and Playgama accepts games published on other platforms [Playgama]. CrazyGames' 2023 developer terms also describe an optional 50% higher compensation for two months of exclusivity on browser game sites, with the full ad SDK [CrazyGames terms 2023]; check the current terms in the portal before you count on it. A simple order: one portal first, learn from its numbers, then the other.

The benchmarks above are from CrazyGames' own page on Basic Launch metrics, which also gives the definitions: average playtime per session, the share of players who return the day after their first session, the share who play at least one minute, and the share who click the thumbnail after seeing it [CrazyGames]. Their tips, condensed: a satisfying core loop and gradual difficulty for playtime; levels, unlocks and saved progress for retention; a small build and no long intro for conversion; one clear focal point, a bold title, an honest cover and a 15 to 20 second preview video for CTR.

Playgama sends reports on time spent, retention and revenue [Playgama], and offers an analytics module in the Bridge for your own events [Playgama]. Send "level started", "level completed" and "level failed" with the level number, and the drop-off point is usually obvious.

What usually moves each number, in our experience making browser games: playtime moves with a second layer of goals (unlocks, a best score to beat); retention with saved progress and a daily reason to return; conversion with the first ten seconds; CTR with the cover. None of these is a promise, and most first games miss at least one benchmark.

Questions

How much can a web portal game earn?

Nobody can tell you in advance. Earnings depend on plays, playtime and advertiser demand, and portal traffic goes mostly to the games with the best metrics. Most games earn little; a few earn well. Neither portal promises an amount.

Can I publish the same game on CrazyGames and Playgama?

Yes, as far as their docs say: CrazyGames says publishing elsewhere does not affect revenue share eligibility if you hold the distribution rights, and Playgama accepts games already published elsewhere. CrazyGames' 2023 terms describe a higher share for two months of exclusivity, so check the current terms before you decide.

Do portals accept games made with AI?

Playgama accepts AI help for code, story and logic, and AI-generated assets in part, but not games built entirely from generative AI, and it rejects emoji icons and default generated-looking interfaces. CrazyGames asks for original names, assets and content. Your own design and art direction matter more than the tool.

Why are there no ads in my CrazyGames game at first?

Ads, banners and in-game purchases are switched off during Basic Launch. Monetization starts only after a game is selected for Full Launch and has the full SDK integration.

Can I use the free Vuena models in a commercial game?

Yes. The free files are CC0 1.0: commercial use, no credit and no permission needed.

Is Vuena connected to CrazyGames or Playgama?

No. Vuena is not affiliated with, endorsed by or sponsored by CrazyGames or Playgama. Their names are used only to describe where games can be published. Their rules change; their own docs are the final word.

Licence

The free models are CC0 1.0: commercial use fine, no credit, no permission needed. portal.js and the prompts on this page are yours to copy and use, no credit needed. The knight clip in step 3 shows one paid-pack animation (sword slash) and says so. The models are made with our own procedural Blender code, written with AI help; the meshes and textures are not AI-generated. This page was written with AI help and checked against the sources below; the test scene and the sizes were measured by us.