Game Bot Creator · Settings → Computer Vision Training

Kaggle Training Setup

One token to turn on Kaggle as a training provider, so training a detection model happens on Kaggle's own servers and shows up as a trackable job instead of a Colab tab you have to babysit.

You don't need this at all if Colab works for you. Colab is the default and needs no account or token. Kaggle is the alternative for training runs you'd rather not sit and watch — see the comparison below before setting it up.
1

Generate a Kaggle API token

Go to kaggle.com/settings/api (or click Create account next to Game Bot Creator's Training provider dropdown if you don't have a Kaggle account yet, then come back to this URL once you're signed in).

  1. Find the API section of the page.
  2. Click Create New Token.
  3. Kaggle generates a single token string starting with KGAT_… — copy it.

This is Kaggle's newer single-token credential, not the older kaggle.json file (the one with a separate username and key) some older guides still describe — if a page mentions "Legacy API Credentials" or a downloaded kaggle.json, that's the wrong one for this.

2

Paste it into Game Bot Creator

Don't see the Computer Vision Training settings? It only appears once a config uses AI: click Tools… next to the config selector and tick Find things that move or look different each time.

In the Settings page's Computer Vision Training group, set Training provider to Kaggle — a Kaggle API token field appears below it. Paste the token in. That's the whole setup; no separate username field.

Kaggle vs. Colab

Both train the exact same notebook and produce the exact same model file — the difference is entirely in how hands-on the process is.

Kaggle

scripted · needs a token

  • Fully automated: Game Bot Creator pushes the dataset, trains, and downloads the finished model for you — no manual upload or download step.
  • Tracked as a job in the Jobs page, with live status and the notebook's own training log in one place.
  • Keeps training on Kaggle's servers if you close Game Bot Creator — a still-running job picks back up automatically next time you open it.
  • Cancelling the job from the Jobs page also stops the run on Kaggle, no separate cleanup.
  • One-time setup: a free Kaggle account plus generating the token above.
  • Kaggle's free GPU time is limited on a rolling weekly basis — see your quota at kaggle.com/settings if you train often.
  • Unlike Colab, Kaggle has no paid tier to buy more once you hit that limit — you wait for it to roll over.

Colab

manual · no account needed

  • Nothing to set up — Game Bot Creator just opens the notebook in your browser and you click through it.
  • No Kaggle account or token required at all.
  • A paid Colab subscription gets you a lot more headroom — faster GPUs and much longer uninterrupted runtime than the free tier of either platform.
  • Hands-on: you click a button inside the notebook to upload the dataset, and download the finished model yourself when it's done.
  • Needs the browser tab open for the whole run — closing it, or the tab disconnecting, interrupts training with nothing to resume from.
  • Free-tier GPU access isn't guaranteed and can disconnect after a period of inactivity, a real risk on a run that takes hours.
  • No progress tracking inside Game Bot Creator — everything happens in the separate Colab tab.

Back in Game Bot Creator

With Training provider set to Kaggle and the token pasted in, Start Training on any project in the AI page now hands off to Kaggle instead of Colab — track it from the Jobs page.