nightknocker/recurrent-qwen3-z-image-turbo

Recurrent Qwen

Qwen breaks its chains: "At last... I run natively on CPU, squire!"

See other models from the retrofitted series.

Inference

qwen_path = 'nightknocker/recurrent-qwen3-z-image-turbo'
text_encoder = RecurrentDecoderModel.from_pretrained(qwen_path).to(torch.bfloat16)
pipeline = ZImagePipeline.from_pretrained(
    'Tongyi-MAI/Z-Image-Turbo',
    text_encoder=text_encoder,
    torch_dtype=torch.bfloat16
)

References

• 2511.07384

Datasets

• artbench-pd-256x256
• anime-art-multicaptions (multicharacter interactions)
• laion
• spatial-caption
• spright-coco
• benchmarks from the Qwen-Image Technical Report