Qwen-Image-2.1 (7B) + Qwen-Image-2512 (20B)

Qwen-Image-2.1 is the finally released unified successor to the announced 2.0: one 7B visual generator for text-to-image and instruction editing, native 2K output, multilingual typography, native transparent RGBA generation, masks/paint edits, identity preservation, and up to ten reference images. Its licence is non-commercial research/evaluation only. Keep Qwen-Image-2512 for commercially clean Apache-2.0 text rendering, paired with Edit-2511 when editing is required.

License: Qwen Research License (2.1, non-commercial); Apache 2.0 (2512) · Context: n/a (diffusion); native 2K resolution · Released: September 20, 2026 (Qwen-Image-2.1); December 31, 2025 (Qwen-Image-2512)

The decision in five lines

The call: Buy — for image
Best for: image
Runs on: 11 hardware picks fit (cheapest: Minisforum UM890 Pro · $463 barebone (no RAM / SSD); $991 with 32 GB / 1 TB — Minisforum US store, Sep 15 2026)
Watch out: Commercial deployment without a separate Qwen licence — 2.1 is not Apache 2.0.
Evidence: Estimated · last verified September 2026

7B visual generator (2.1): PARAMETERS
UNIFIED IMAGE GENERATION + EDITING: TYPE
n/a: CONTEXT
~24 GB comfortable with component offload; no official VRAM table: VRAM AT Q4

Where we recommend this

Every tier slot in the planner where this model is a top or alternate pick. Pulled live from planner.js — when the planner refreshes, this table stays current.

IMAGE ·

Qwen-Image-2.1 (7B, research-only)September 20, 2026 unified generation + editing release: native 2K, transparent RGBA, masks, identity preservation and up to ten reference images. The full pipeline benefits from frontier memory. Qwen Research License is non-commercial; keep Qwen-Image-2512 for Apache-2.0 commercial work.

IMAGE · TOP

Qwen-Image-2.1 (7B, Qwen Research License)Fresh unified gen+edit pick: strong multilingual text, native transparency and up to ten references. Research/evaluation only unless Qwen grants a commercial licence; use Qwen-Image-2512 for Apache-2.0 commercial work.

IMAGE · HIGH

Qwen-Image-2.1 (7B, research-only)Unified generation and editing with CPU-offload support, native 2K and transparent RGBA. No official VRAM table: expect component offload below 24 GB. Non-commercial Qwen Research License.

The call

Qwen-Image-2.1 is the finally released unified successor to the announced 2.0: one 7B visual generator for text-to-image and instruction editing, native 2K output, multilingual typography, native transparent RGBA generation, masks/paint edits, identity preservation, and up to ten reference images. Its licence is non-commercial research/evaluation only. Keep Qwen-Image-2512 for commercially clean Apache-2.0 text rendering, paired with Edit-2511 when editing is required.
When not to use: Commercial deployment without a separate Qwen licence — 2.1 is not Apache 2.0. Use Qwen-Image-2512, HiDream-O1 (MIT), Z-Image-Turbo or FLUX.2 klein 4B instead. For speed-sensitive batch generation where text fidelity is secondary, Z-Image-Turbo remains much lighter.

Runner notes

Official Diffusers pipeline supports `enable_model_cpu_offload`; ComfyUI community packaging and quantized variants appeared immediately. The official repository contains roughly 14.2 GB of visual-transformer weights plus a large Qwen3-VL text encoder and VAE (about 39 GB downloaded in total), and Qwen publishes no official VRAM table. Do not treat a community 16 GB run as a guarantee. Qwen-Image-2512 retains mature ComfyUI, GGUF and FP8 paths.

License: Qwen Research License (2.1, non-commercial); Apache 2.0 (2512)
Released: September 20, 2026 (Qwen-Image-2.1); December 31, 2025 (Qwen-Image-2512)
Maker: Alibaba
Model card: huggingface.co/Qwen/Qwen-Image-2.1 →

Hardware that fits

Every hardware pick whose memory fits this model at the quant we recommend. Sorted cheapest-first — the top row is your best-value fit. Click through for the full buyer’s guide.

Next step

Find-by-model — see what hardware runs this→