Qwen-Image-2.1-Turbo

01 / Recipe parameters

Fixed resources and deployment boundary

Qwen-Image-2.1-Turbo is a new single-device fast image generation and local editing recipe in MegaCube Cluster Manager 1.2.0. It fixes BF16, Turbo eight-step inference, and the platform runtime revision. This page records download, resource, and deployment boundaries rather than presenting a new performance benchmark.

Model / version

Qwen-Image-2.1 · Turbo recipe

Recipe variant

BF16 · Turbo 8 steps · Single device

Input / output

Text or image input; image output for fast generation and local editing

Model repository

Qwen/Qwen-Image-2.1 · revision fc13ba1986fcb905854dc8de2f52eee395f49b5b

Runtime revision

qwen-image-21-turbo-eightstep-cfg1-kvcache-v1

Resource requirement

At least 110 GiB unified memory; at least 60 GiB disk headroom

Service image

megacube/qwen-image-21-turbo:da1d382-bf16-v1

Runtime

ARM64; Docker 24+; NVIDIA Container Toolkit; platform-fixed runtime image

Deployment limit

One device per recipe; one replica maximum

1. Choose the fixed recipe

   Select Qwen-Image-2.1-Turbo in Cluster Manager and confirm its single-device requirements and recipe version.

2. Prepare resources

   Use the online or offline preparation tool to download model and runtime files, then complete space and integrity checks.

3. Deploy and validate

   Assign one device, deploy through the platform, and validate with a fast generation or local editing task. Concurrency and queue behavior follow model details and actual platform configuration.

A fixed eight-step recipe creates a consistent fast-generation workflow; production projects should still validate image sizes, editing tasks, and business concurrency. The original Qwen-Image-2.1 entry and report remain available.