Ranked guide

Local Image Generation — Pixels on Your Own Machine

Some of the most capable image generators now fit on a single home graphics card. No cloud account, no monthly fee — just your hardware turning text into images in seconds. But open weights don't always mean free for business: check each model's licence before you sell what it makes.

Decision first

Our ranking

Start with the winner, then compare the trade-offs that might change the answer for you.

#1 Local Image Generation

Qwen-Image-2.1

Alibaba (Qwen Team)

Qwen-Image-2.1 is a small, clever image model you can download and run yourself. With a 7-billion-parameter image generator, it creates regular and transparent images at 2K, edits photos, and blends up to ten reference images into one scene. The catch is legal, not technical: unlike earlier Qwen image models, it isn't licensed for commercial use.

Why It Wins

Compact 7B visual generator (32 single-stream DiT layers), much smaller than the 20B Qwen-Image family before it. Generates transparent (RGBA) images natively, edits transparent layers, and cuts subjects out of photos. Accepts up to 10 reference images and local edits marked with circles, scribbles, or masks. Native 2048×2048 output with widescreen and portrait sizes. Day-one support in ComfyUI and Diffusers.

The Catch

It is released under the Qwen Research License, not Apache 2.0 like earlier Qwen image models. Personal and research use is fine; selling images or building a paid product on it requires a separate agreement with Alibaba. Community add-ons (LoRAs, fine-tunes) will grow more slowly as a result.

8.8 Editorial score
Read review
Best for

Qwen-Image-2.1 is a small, clever image model you can download and run yourself. With a 7-billion-parameter image generator, it creates regular and transparent images at 2K, edits photos, and blends up to ten reference images into one scene. The catch is legal, not technical: unlike earlier Qwen image models, it isn't licensed for commercial use.

Why It Wins

Compact 7B visual generator (32 single-stream DiT layers), much smaller than the 20B Qwen-Image family before it. Generates transparent (RGBA) images natively, edits transparent layers, and cuts subjects out of photos. Accepts up to 10 reference images and local edits marked with circles, scribbles, or masks. Native 2048×2048 output with widescreen and portrait sizes. Day-one support in ComfyUI and Diffusers.

Watch out

It is released under the Qwen Research License, not Apache 2.0 like earlier Qwen image models. Personal and research use is fine; selling images or building a paid product on it requires a separate agreement with Alibaba. Community add-ons (LoRAs, fine-tunes) will grow more slowly as a result.

#2

FLUX.2 Klein

Black Forest Labs

The people's image generator. Built by the same team that created Stable Diffusion, FLUX.2 Klein packs FLUX-lineage photorealism into models small enough to run on a mid-range gaming laptop. The 4B variant needs just 8GB of VRAM — meaning the RTX 4060 in your college laptop can now produce studio-quality images. Apache 2.0 licensed.

8.6 Editorial score
Read review
#3

Z-Image

Alibaba Tongyi

The speed demon of local image generation. A 6-billion-parameter model that generates images in 8 inference steps — often under one second — on hardware so modest it makes other AI models jealous. Runs on 6GB of VRAM with quantization. Apache 2.0 licensed. If FLUX.2 Klein democratized quality, Z-Image democratized *speed*.

8.5 Editorial score
Read review
Questions, answered

Frequently Asked Questions