Measure the generative quality of converted FLUX.2 Klein models

Measure the GenEval, DPG, and FID performance of the converted FLUX.2 Klein pixel-space models to determine how closely their generation quality matches that of the latent FLUX.2 Klein parent.

Background

The converted FLUX.2 Klein models are used in the downstream depth-estimation and restoration experiments, but the paper does not report standard text-to-image generation metrics for them. Consequently, the quality gap, if any, between the converted pixel-space models and their latent parent remains unresolved.

References

We have not measured GenEval, DPG or FID for the converted models, so we cannot say how close they come to their latent parent.

— Iris-3B: Going Beyond the Latent with Pixel-Space Diffusion Training, Conversion and Fine-Tuning  (2610.09450 - Cai et al., 7 Oct 2026) in Section 4.3, “Production conversions”