Local Image Generation — Pixels Without Permission

The heavyweight champion of open-source image generation. A 27-billion-parameter architecture that fuses a diffusion transformer with a vision-language model, producing photorealistic humans and bilingual text rendering that rivals cloud-only services — all under Apache 2.0, meaning you own every pixel it generates.

Highest-ranked Apache 2.0 open-weight model on Arena.ai (Elo ~1,130). Photorealistic human faces without the uncanny valley. Bilingual text rendering in English and Chinese. Full commercial rights with zero restrictions.

27 billion parameters is a lot of neural network to run at home. You'll need an RTX 4090 with INT4 quantization to squeeze it in at ~14GB VRAM, and even then you're pushing the hardware. Documentation skews heavily Chinese-first.

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.

Most accessible high-quality local model available — the 4B variant runs on 8GB VRAM. Apache 2.0 license with zero commercial restrictions. Inherits FLUX photorealism lineage. Best-in-class text-in-image rendering for its size class. Massive ComfyUI and LoRA ecosystem.

Klein is the consumer tier — Black Forest Labs keeps the best quality for their proprietary Pro and Max models. The FLUX.2 Dev variant exists but is non-commercial. Klein-specific LoRAs are still growing compared to the enormous FLUX.1 library.

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*.

Sub-second image generation in 8 inference steps. Runs on as little as 6GB VRAM quantized — the most accessible local model period. Apache 2.0 with full commercial rights. Multiple specialized variants (Turbo, Edit, Omni-Base) for different workflows. Bilingual text rendering in English and Chinese.

Newest of the three with the smallest community ecosystem. Quality at maximum settings slightly trails Qwen-Image and FLUX at their best. The LoRA library is still nascent compared to FLUX's years-deep collection.