Instructions to use SyntheticMDProductions/Roblox_Oasis_V4_AI_Model with libraries, inference providers, notebooks, and local apps. Follow these links to get started.
- Libraries
- Diffusers
How to use SyntheticMDProductions/Roblox_Oasis_V4_AI_Model with Diffusers:
pip install -U diffusers transformers accelerate
import torch from diffusers import DiffusionPipeline # switch to "mps" for apple devices pipe = DiffusionPipeline.from_pretrained("SyntheticMDProductions/Roblox_Oasis_V4_AI_Model", dtype=torch.bfloat16, device_map="cuda") prompt = "Astronaut in a jungle, cold color palette, muted colors, detailed, 8k" image = pipe(prompt).images[0] - Notebooks
- Google Colab
- Kaggle
import torch
from diffusers import DiffusionPipeline
# switch to "mps" for apple devices
pipe = DiffusionPipeline.from_pretrained("SyntheticMDProductions/Roblox_Oasis_V4_AI_Model", dtype=torch.bfloat16, device_map="cuda")
prompt = "Astronaut in a jungle, cold color palette, muted colors, detailed, 8k"
image = pipe(prompt).images[0]Roblox Oasis V 2.3.7 V1
An experimental action-conditioned, rectified-flow video world model trained on recorded Roblox gameplay. Given a current visual state and supported input actions, it predicts the next simulated game frame.
This is a research/demo model, not a playable Roblox client. It does not connect to Roblox, its servers, or any Roblox accounts.
Game Previews
Model details
- Resolution: 256 x 144 (16:9)
- Architecture: action-conditioned rectified-flow video model
- Inference weights:
unet/diffusion_pytorch_model.safetensors - Supported demonstrated inputs: movement (W/A/S/D), jump, shift,
1, left/right mouse, and relative mouse movement - Training result: 29 completed epochs; best recorded counterfactual accuracy: 85.4% on 48 validation samples
Files
unet/: model architecture configuration and inference weightsaction_flow_model_info.json: input encoding, actions, and training metadatagui_previews/: training preview images
Use
Use this model with the companion trainer/player application: Oasis-Game-Trainer. Copy the downloaded folder into that application's model-output directory, then select it from the player interface.
The project uses Python, PyTorch, Diffusers, OpenCV, Pillow, MSS, and Pynput. See the companion repository for installation and runtime instructions.
Limitations
This model is a learned visual simulation. It may drift, become inconsistent over longer rollouts, or react imperfectly to inputs. It was trained on a limited action distribution and should be evaluated as an experimental project rather than a game replacement.
Attribution and rights
Unofficial research project. “Roblox” is used descriptively; this project is not affiliated with, endorsed by, or sponsored by Roblox Corporation. Only use or redistribute data and footage for which you have the necessary rights.
- Downloads last month
- -
