Back to Model
Q

Qwen Image 2.1 Turbo EXL3

Model Oct 18, 2026

Community 3-, 4- and 5-bit EXL3 diffusion Transformer checkpoints for a custom AMD ROCm image-generation runtime.

Visit primary link 2 sources attached
Author-generated Qwen Image Turbo samples compare BF16, 3-bit, 4-bit and 5-bit EXL3 across text, composition and two-reference cases.
Contributor-published generated samples with identical prompts and seeds, using independent generation trajectories. Full comparison preserved. This is a three-case author screen, not an AITrending reproduction or a general quality guarantee. Source: jyohukuchan / Qwen Image Turbo EXL3
multimodalself-hosted

What happened lately

Qwen Image 2.1 Turbo EXL3 timeline

This third-party conversion packages Qwen-Image-2.1-Turbo’s diffusion Transformer in three EXL3 bit widths. The published Transformer files occupy approximately 2.89, 3.77 and 4.64 decimal GB; the BF16 text encoder and VAE are separate. These figures are storage sizes, not complete-pipeline VRAM requirements.

The checkpoints require the custom rocm_exl3_forme engine and its image CLI. They are not GGUF or ordinary Diffusers checkpoints. The maintainer reports validation on one Radeon AI PRO R9700 32GB, with other GPUs and CUDA unvalidated for this release. The Qwen Research License limits use to noncommercial research or evaluation unless separately licensed; the engine’s MIT license does not override those model terms.