MiniMax-H3 x Z-Image Hybrid

A custom hybrid version of MiniMax-H3 combining the FL2VA and REF2VA branches of the model.

The model uses different branches for different transformer blocks:

B00-B24  -> FL2VA
B25-B49  -> REF2VA
B50+     -> FL2VA

This hybrid layout was constructed at the raw safetensor payload level. No tensor conversion was performed when constructing the BF16 hybrid.

Files

BF16

minimax_h3_hybrid_fl2va_ref2va_zs05_b25-49_bf16.safetensors

Approximately 37.46 GiB.

This is the full-precision BF16 hybrid model.

The B25-B49 tensors were copied byte-for-byte from the REF2VA ZS05 BF16 model, while the remaining tensors come from the FL2VA ZS05 BF16 model.

INT8

The repository also contains an INT8 version of the same hybrid model.

The INT8 model follows the same conceptual hybrid layout:

B00-B24  -> FL2VA
B25-B49  -> REF2VA
B50+     -> FL2VA

Verification

The BF16 hybrid was verified at tensor level.

Results:

Tensor count                 : 532
Expected source matches      : 532/532
Payload failures             : 0
Wrong-source tensors         : 0

B00-B24 : FL2VA
B25-B49 : REF2VA
B50+    : FL2VA

Tensor set       : PASS
dtype            : PASS
shape            : PASS
payload sizes    : PASS
Payload hashes   : PASS

All B25-B49 payloads were verified to be byte-exact copies from the REF2VA source.

No tensor conversion was performed during hybrid construction.

Why this hybrid exists

The purpose of this model is to experiment with a mixed FL2VA/REF2VA architecture where only the middle transformer blocks use REF2VA.

The selected layout is:

FL2VA
  |
  +-- B00-B24
  |
REF2VA
  |
  +-- B25-B49
  |
FL2VA
  |
  +-- B50+

This allows the model to be tested as a single hybrid model without requiring separate model files for the two branches.

Related model

For the original/native Z-Image implementation and the REF2VA BF16 model, see:

https://huggingface.co/joeygambino/MiniMax-H3-x-Z-Image-native

Please refer to the original repository for the native model structure, configuration and usage information.

GGUF

A GGUF conversion of the BF16 hybrid is provided separately:

https://huggingface.co/hoidhxd/MiniMax-H3-x-Z-Image-hybrid-GGUF

The GGUF version is intended for use with GGUF-compatible ComfyUI workflows and loaders.

Disclaimer

This is an experimental custom hybrid model.

It is not an official MiniMax release.

The hybrid block selection was created for experimentation and testing.