Qwen-Image-Edit-2511
An image editing model released by Alibaba that supports multi-reference image fusion. It demonstrates capabilities in texture rendering and lighting simulation, as well as instruction following and spatial understanding.
Timeline (last 90 days)
Qwen-Image-Edit-2511 with the Lightning four-step LoRA produces an image in about 4 seconds on an H100, but it will alter joint angles that were not instructed (e.g., changing a skull crusher back to a straight-arm bench press).
Qwen-Image-Edit-2509 and 2511 now accept up to 3 images per edit.
Unsloth v0.1.906 integrates the official 8-step version of Qwen-Image-2.1, Qwen-Image-2.1-Turbo, with a model size of 4.2 GB in Q4 GGUF, and 8-step sampling increases image generation speed by about 2.5× compared with the regular version.
In community tests, Qwen-Image-Edit-2511 stood out in image texture quality, lighting, and shadow handling.
Selection recommendation: choose Qwen-Image-Edit-2511 for superior lighting and texture quality; for multi-image fusion with complex spatial relationships, SenseNova-U1.5-Lite is worth evaluating.
In the "cat riding a scooter" test group, Qwen-Image-Edit-2511 generated an unreasonable pillar beneath a coffee table, and the cat's front paws were placed in an unnatural position.
On the spatial understanding dimension, SenseNova-U1.5-Lite outperformed Qwen-Image-Edit-2511, with more reasonable spatial relationships between objects and physical contact between limbs.
All Facts (7)
Qwen-Image-Edit-2511 with the Lightning four-step LoRA produces an image in about 4 seconds on an H100, but it will alter joint angles that were not instructed (e.g., changing a skull crusher back to a straight-arm bench press).
50%UnverifiedQwen-Image-Edit-2509 and 2511 now accept up to 3 images per edit.
50%UnverifiedOn the spatial understanding dimension, SenseNova-U1.5-Lite outperformed Qwen-Image-Edit-2511, with more reasonable spatial relationships between objects and physical contact between limbs.
50%UnverifiedIn community tests, Qwen-Image-Edit-2511 stood out in image texture quality, lighting, and shadow handling.
50%UnverifiedSelection recommendation: choose Qwen-Image-Edit-2511 for superior lighting and texture quality; for multi-image fusion with complex spatial relationships, SenseNova-U1.5-Lite is worth evaluating.
50%UnverifiedIn the "cat riding a scooter" test group, Qwen-Image-Edit-2511 generated an unreasonable pillar beneath a coffee table, and the cat's front paws were placed in an unnatural position.
50%UnverifiedUnsloth v0.1.906 integrates the official 8-step version of Qwen-Image-2.1, Qwen-Image-2.1-Turbo, with a model size of 4.2 GB in Q4 GGUF, and 8-step sampling increases image generation speed by about 2.5× compared with the regular version.
50%