EnclaveHost/qwen-image-2.1-sd

Enclave image bundle — Qwen-Image-2.1

Repackaged Qwen-Image-2.1 components for stable-diffusion.cpp, with generic component names used by Enclave model volumes. Files are renamed without changing their contents. This is an upstream model bundle, not a fine-tune.

Components

Local filename	Component	Upstream
diffusion.gguf	Qwen-Image-2.1 Q8_0 transformer	leejet/Qwen-Image-2.1-GGUF
llm.gguf	Qwen3-VL-8B-Instruct Q8_0 text encoder	Qwen/Qwen3-VL-8B-Instruct-GGUF
vae.safetensors	Qwen-Image-2.1 BF16 VAE	Comfy-Org/Qwen-Image-2.1

Exact source revisions, original filenames, byte counts and SHA-256 checksums are in provenance.json. The transformer and text encoder were quantized by their respective upstream publishers. Enclave does not requantize them.

Runtime

Requires a stable-diffusion.cpp revision supporting Qwen Image 2.1 and its Qwen3-VL text encoder. The older Qwen-Image-2512 engine and VAE are incompatible. The Enclave fleet's historical b5d812008eb7082a238fc589444544b3278187ae engine pin predates this support; merely attaching this volume does not upgrade that engine.

Example with a compatible sd-cli:

sd-cli --diffusion-model diffusion.gguf --llm llm.gguf \
  --vae vae.safetensors --cfg-scale 6.0 --sampling-method euler \
  --steps 40 -W 512 -H 512 -p 'a red fox in snow, studio lighting' -o image.png

Use dimensions divisible by 32. This bundle contains the components for text-to-image generation; reference-image editing additionally requires the matching Qwen3-VL vision projector and a runtime exposing that input.

License

Qwen-Image-2.1 uses the Qwen Research License Agreement, included as LICENSE, with required attribution in Notice. It permits research and evaluation use only; commercial use requires a separate license from Qwen. The text encoder is independently distributed under Apache-2.0 (see LICENSE.text-encoder). No commercial license is granted by this repackaging.