voxprint-z-image-turbo-bf16

VOXPrint's own BF16 dtype conversion of Tongyi-MAI/Z-Image-Turbo @ f332072aa78be7aecdf3ee76d5c247082da564a6, Apache-2.0.

Modified by VOXPrint: dtype conversion from f332072. The transformer weights were cast from the author's FP32 shards to BF16 (521 tensors, 12,309,817,472 bytes of tensor data). No fine-tuning, no ablation, no calibration data. The transformer config is byte-identical to the author's transformer/config.json. Only the transformer is here; the text encoder, tokenizer, VAE and scheduler are unmodified author files, available upstream and in the mirror Mitroshenkov87/voxprint-mirror-z-image-turbo.

• Original repository: https://huggingface.co/Tongyi-MAI/Z-Image-Turbo
• Original commit: f332072aa78be7aecdf3ee76d5c247082da564a6
• Licence: Apache-2.0, same as the original. See https://www.apache.org/licenses/LICENSE-2.0
• Credit: all rights to the model belong to Tongyi-MAI / Alibaba; this repo is not affiliated with them.
• Provenance: the shards were previously hosted in Mitroshenkov87/voxprint-mirror-z-image-turbo (revision 00205098d8ebb0643f4ec72c1379dc8bf798f9c6) under the names in the last column, and were copied here byte for byte.

Files

Made for VOXPrint AI Media Suite

VOXPrint AI Canvas loads this transformer as its default 16 GB / 24 GB BF16 tier, with the remaining pipeline components from the author's original files.