mean_flow_xlarge_2(pretrained: bool = False, overrides: object = {})Extra-large width, patch 2 — the paper's headline model.
Model Size
Parameters
pretrainedbool= FalseAccepted for signature parity with the rest of the zoo. No
MeanFlow checkpoint ships — the paper releases none, and a
True here would have nothing to load.**overridesobject= {}Optional
MeanFlowConfig field overrides.Returns
MeanFlowModelThe average-velocity network.
Notes
Reference: Geng et al., arXiv:2505.13447, 2025, Tables 2 and 4. Reports 3.43 FID at one function evaluation on ImageNet 256x256 and 2.93 at two — the first figure being the one that closes most of the gap to 250-step diffusion models of the same backbone (DiT-XL/2 at 2.27, SiT-XL/2 at 2.06).
Examples
>>> from lucid.models import mean_flow_xlarge_2
>>> config = mean_flow_xlarge_2().config
>>> config.hidden_size, config.depth, config.num_heads
(1152, 28, 16)