Qwen-Image-2.1 vs SenseNova U1.5 Lite: 5 same-prompt image edit tests show an instruction-following gap

Wild_Artichoke7648 · reddit · 2026-09-24

The author ran Qwen-Image-2.1 and SenseNova U1.5 Lite on five editing tasks with identical inputs and prompts.

Key takeaway: Qwen's weakness is instruction adherence, not image quality—it often misses explicit requirements in complex prompts. SenseNova is less dramatic visually but more consistent with multi-part constraints. Parameter counts aren't directly comparable: Qwen's 7B is the visual generation component (paired with a Qwen3-VL 8B encoder), while SenseNova U1.5 Lite uses a unified encoder-free, VAE-free 8B-MoT architecture.

Original post →

More from Models

Models channel →