2024-08-27 22:59:40
gamengen.github.io
Data Collection via Agent Play:
Since we cannot collect human gameplay at scale, as a first stage we train an automatic RL-agent to play
the game,
persisting it’s training episodes of actions and observations, which become the training data for our
generative model.
Training the Generative Diffusion Model: We re-purpose a small diffusion model, Stable Diffusion v1.4,
and condition it on a sequence of previous actions and observations (frames). To mitigate
auto-regressive drift
during inference, we corrupt context frames by adding Gaussian noise to encoded frames during training.
This allows the network to correct information sampled in previous frames, and we found it to be
critical
for preserving visual stability over long time periods.
Latent Decoder Fine-Tuning: The pre-trained auto-encoder of Stable Diffusion v1.4, which compresses
8×8
pixel patches into 4 latent channels, results in meaningful artifacts when predicting game frames, which
affect
small details and particularly the bottom bar HUD. To leverage the pre-trained knowledge while improving
image quality, we train just the decoder of the latent auto-encoder using an MSE loss computed
against the target frame pixels.
Support Techcratic
If you find value in our blend of original insights (Techcratic articles and Techs Got To Eat), up-to-date daily curated articles, and the extensive technical work required to keep everything running smoothly, consider supporting Techcratic with Bitcoin. Your support helps me, as a solo operator, continue delivering high-quality content while managing all the technical aspects, from server maintenance to future updates and improvements. I am committed to continually enhancing the site and staying at the forefront of trends to provide the best possible experience. Your generosity and commitment are deeply appreciated. Thank you!
Bitcoin Address:
bc1qlszw7elx2qahjwvaryh0tkgg8y68enw30gpvge
Please verify this address before sending any funds to ensure your donation is directed correctly.
Bitcoin QR Code
Your contribution is vital in supporting my efforts to deliver valuable content and manage the technical aspects of the site. To donate, simply scan the QR code below. Your generosity allows me to keep providing insightful articles and maintaining the server infrastructure that supports them.
Privacy and Security Disclaimer
- No Personal Information Collected: We do not collect any personal information or transaction details when you make a donation via Bitcoin. The Bitcoin address provided is used solely for receiving donations.
- Data Privacy: We do not store or process any personal data related to your Bitcoin transactions. All transactions are processed directly through the Bitcoin network, ensuring your privacy.
- Security Measures: We utilize industry-standard security practices to protect our Bitcoin address and ensure that your donations are received securely. However, we encourage you to exercise caution and verify the address before sending funds.
- Contact Us: If you have any concerns or questions about our donation process, please contact us via the Techcratic Contact form. We are here to assist you.
Disclaimer: As an Amazon Associate, Techcratic may earn from qualifying purchases.