self-hosted/ai
§01·model · /models

Z-Image Turbo

imageactiveApache-2.0

6B step-distilled variant of Tongyi-MAI Z-Image — 8 inference steps at fp16. Sibling of Z-Image-Base (slower, full quality) and Z-Image-Distilled (further compressed). Fits comfortably on 16 GB consumer GPUs per HF card.

§02·GPUs that run this model
16 total
GPUVRAMSeriesBest speedMin VRAMWorksBenchmarksRecipe
RTX 309024GB309s13GB1recipecheck ↗
RTX 409024GB402.3s13GB1recipecheck ↗
RTX 4060 Ti 16GB16GB401recipecheck ↗
Apple M2 Max64GBapple~0recipecheck ↗
Apple M2 Pro16GBapple~0recipecheck ↗
Apple M3 Max48GBapple~0recipecheck ↗
Apple M4 Max48GBapple~0recipecheck ↗
RTX 3090 Ti24GB30~0recipecheck ↗
RTX 4070 Ti Super16GB40~0recipecheck ↗
RTX 408016GB40~0recipecheck ↗
RTX 4080 Super16GB40~0recipecheck ↗
RTX 5060 Ti16GB50~0recipecheck ↗
RTX 5070 Ti16GB50~0recipecheck ↗
RTX 508016GB50~0recipecheck ↗
RX 7800 XT16GBamd~0recipecheck ↗
RX 7900 XTX24GBamd~0recipecheck ↗

benchmarked·~ runs via recipe (not benchmarked)· untested·doesn't fit