Open image model releases

last edited

Open image model releases from January to October were shaped by GLM-Image, FLUX.2 klein, and the Qwen-Image successors. Z.ai released GLM-Image in January with strong text rendering, and Black Forest Labs shipped FLUX.2 klein the same month. Qwen-Image successors added their own eight-step generation by autumn. This page is part of Image Generation Services & Models.

GLM-Image text rendering

Z.ai released GLM-Image in January 2026. It is a hybrid autoregressive and diffusion model for information-dense images such as infographics, slides, and technical diagrams. The open release has 16B parameters and pairs an autoregressive generator with a diffusion decoder.

It scored 0.9116 word accuracy on complex text rendering. Z.ai's documentation ranks it first among open-source models on that measure. Nano Banana Pro, a closed Google model, scored 0.7788. On the OneIG benchmark, Nano Banana 2.0 scored 0.578.

FLUX.2 klein

FLUX.2 klein combines generation and editing in one architecture, which suits workflows built on Character consistency and iteration. The release note is dated 15 January 2026. The 4B model is licensed under Apache 2.0. The 9B model is non-commercial only.

Black Forest Labs targets sub-second inference on consumer hardware for interactive tools. Its FLUX.2 Turbo adapter adds an eight-step path for fast generation and editing. Open weights support local generation.

Qwen-Image successors

Qwen-Image-2512 appeared in December 2025 as an open alternative. Alibaba launched Qwen-Image-2.0 in February 2026. Qwen-Image-2.1 followed in September, with open weights on Hugging Face and ModelScope. Its license is research-only, unlike earlier open Qwen image models.

Qwen-Image-2.1-Turbo arrived in October. It generates and edits images in 8 denoising steps. Regional context is in Regional AI Art Development.

These releases sit within the open-source model landscape.

Sources