Open source
Week of 28 September 2026
Local and quantized image generation dominates this week, with Qwen-Image-2.1 appearing in multiple optimized forms.
- 1Qwen Image 2.1 Uncensored GGUFHF Models
abenzerps· 2,203 likes
GGUF quantizations of Qwen/Qwen-Image-2.1 packaged for local text-to-image generation, with variants ranging from a full BF16 file at 14.23 GB down to a 4-bit MLX build at 4.00 GB. Builders who want to run Qwen-Image-2.1 locally through ComfyUI can pick a quantization tier that fits their available VRAM without sourcing or converting weights themselves.
ggufqwenimage-generationcomfyuicomfyui-gguftext-to-image - 2Qwen Image 2.1 Viggle TurboHF Models
Viggle· 370 likes
Viggle's distilled variant of Qwen-Image-2.1 handles both text-to-image generation and instruction-driven image editing with up to three reference images, completing inference in six transformer passes rather than the base model's forty, with no classifier-free guidance required. Builders facing latency constraints will find the roughly 5x end-to-end speedup compelling, and the published side-by-side comparisons suggest quality parity on most prompts outside of small, dense text.
diffuserssafetensorsloratext-to-imageimage-to-imageimage-editing - 3OpenCreatorGitHub
krillinai· 12,479 stars
OpenCreator is an open-source desktop workspace that consolidates AI-powered video generation, image creation, voice synthesis, dubbing, and video translation into a single agent-driven interface built on Codex. Builders working on localization or dubbing pipelines may find the integrated TTS and video translation support useful as a reference or starting point for their own tooling.
agent-skillscodex-clideskop-appdubbingimage-generationlocalization - 4Ming Image 0.1 DesignHF Models
inclusionAI· 319 likes
A 6B text-to-image model from inclusionAI built specifically for text-heavy visual formats like UI layouts, infographics, and posters, with native RGBA output for transparent-background compositions. Builders working on design automation pipelines may find the transparent background support and text rendering focus useful for generating layered assets without post-processing cutouts.
customdiffuserssafetensorstext-to-imageimage-generationgraphic-design - 5Qwen Image 2.1 GGUFHF Models
unsloth· 280 likes
A GGUF-quantized text-to-image model built on Qwen-Image-2.1, applying Unsloth's Dynamic 2.0 per-tensor precision upcasting to preserve quality at reduced size. Builders targeting local or CPU-adjacent inference may find it useful, as it runs via stable-diffusion.cpp alongside a separately loaded VAE and Qwen3-VL text encoder.
ggufquantizedunslothqwenimage-generationtext-to-image