Which image base model you used in pyramidflow-miniflux

[guyuchao]

The transformer config of pyramidflow-miniflux is different from flux.1. How you adapt the flux.1 to this smaller version.

[feifeiobama]

we reduced the width & #layers in FLUX.1 and trained it from scratch.

[guyuchao]

Do you mean you only adopt the flux structure but not its pretrained weights?

[feifeiobama]

yes :)

[guyuchao]

have you pre-trained a t2i model yourself, or directly train a t2v model from scratch?

[feifeiobama]

regarding the released checkpoint, we first pre-trained a t2i model and then perform image-video joint training for the t2v model.