Loading blocks of both (fl2va, ref2va) models on the fly
#5
by PrimePolygons - opened
Hi All,
this model is great inspirational work, but it is possible to load parts of each model on the fly in custom ComfyUI node, to save some space and more freely experiment/choose block ranges.
On github there is a node, to do that:
(https://github.com/entropicnoise/MiniMax-H3-Combined-Model-Loader)
PrimePolygons changed discussion title from Loading choosen blocks both models on the fly instead of merge to Loading choosen blocks of both (fl2va, ref2va) models on the fly
PrimePolygons changed discussion title from Loading choosen blocks of both (fl2va, ref2va) models on the fly to Loading blocks of both (fl2va, ref2va) models on the fly
I think I was able to use start image + reference image, and also only reference images in pure FL2VA with above loader plus the (https://github.com/entropicnoise/MiniMax-H3-Combined-Image-And-Reference-to-Video), but I'll make few more tests and analyse the code, before saying that with certainty.