Qwen-Image 2.1 Launches with ComfyUI Support, Single 7B Model for Generation and Editing
Lexius2129 · reddit · 2026-09-21
Qwen-Image 2.1 received ComfyUI support on launch day, with open weights (research-only license). Key points:
- A single compact 7B model unifies image generation and editing, supporting up to 10 input images
- Texture quality is noticeably improved over the previous Qwen-Image-Edit, with strong text rendering for infographics, posters, and Chinese calligraphy, native 2K output, and support for alpha transparency
- Runs on consumer hardware: on a 5080 laptop (16GB VRAM), an int8 diffusion model + w4a8 text encoder workflow averages 42 seconds, dropping to about 30 seconds with comfy kitchen attention enabled
- Kijai produced the w4a8 quantized text encoder; ComfyUI already offers text-to-image and image editing workflow templates, with weights available on Hugging Face
More from Multimodal
- A Tree Through Four Seasons in a Bottle, Built With Three.js + TSL and Suno — techartist_ · 2026-09-21
- MiniMax H3 License Doesn't Apply in US, EU, UK, or South Korea — Read Before Installing — LeoLeg76 · 2026-09-21
- Picteus: open-source local indexer gives every Stable Diffusion output a searchable full recipe — shivam_dewan · 2026-09-21
- Nano Banana 2.5 image model expected next week, with thinking levels and up to 4K output — koltregaskes · 2026-09-21
- AI art community share: underrated AI artists and OG Tezos crypto-artists to follow — Merzmensch · 2026-09-21
- Early Qwen-Image 2.1 tests: art styles OK, realism 'so bad' tester regretted it — No_Daikon3851 · 2026-09-21