Qwen-Image-2.1 (unmodified mirror)

An unmodified, byte-identical mirror of Qwen/Qwen-Image-2.1 at upstream revision b3179ad355be050328e483a9dfdd9e60cd62adfa (2026-09-20). Every file was checked against the upstream LFS SHA-256 / git blob hash before upload. Upstream commits after that revision change only the README and images, not the weights.

It is the weight source for qwen-image21-swift, a Swift/MLX port of the model for Apple Silicon. The port reads this diffusers layout directly; no conversion is applied.

Licence: research / evaluation only

Qwen-Image-2.1 is licensed under the Qwen RESEARCH LICENSE AGREEMENT (included as LICENSE). Use is limited to non-commercial research or evaluation; commercial use requires a separate licence from Qwen (model-business@notice.qwencloud.com). This mirror is redistributed under §3 of that agreement, with the agreement and the attribution notice (NOTICE) included. Nothing here grants any right beyond that agreement.

Qwen is licensed under the Qwen RESEARCH LICENSE AGREEMENT, Copyright (c) 2026 Hangzhou Tongyi Laboratory Technology Co., Ltd. All Rights Reserved.

Contents

Path	What
transformer/	7B single-stream block-causal DiT (bf16 safetensors, 2 shards)
vae/	64-channel 16× RGBA VAE
processor/	Qwen3-VL processor / tokenizer files
scheduler/	FlowMatchEulerDiscreteScheduler config
model_index.json, LICENSE, NOTICE	pipeline index, the agreement, attribution

Not included: text_encoder/. Upstream's text encoder is byte-identical to the Apache-2.0 Qwen/Qwen3-VL-8B-Instruct (750/750 tensors). Load it from there. model_index.json still names a text_encoder component, so diffusers users must pass one explicitly, e.g. QwenImage21Pipeline.from_pretrained("xocialize/Qwen-Image-2.1", text_encoder=Qwen3VLForConditionalGeneration.from_pretrained("Qwen/Qwen3-VL-8B-Instruct", torch_dtype=torch.bfloat16), torch_dtype=torch.bfloat16).

For the model itself (capabilities, prompting, RGBA transparency, editing with up to 10 reference images), see the upstream model card.