Summary
Nerdy Rodent’s latest ComfyUI tutorial puts the newly supported JoyAI Image Edit and Image Edit Plus models through a series of practical tests, comparing their output directly against Qwen Image Edit and the Flux Kline 4B int8 model. The video is structured as a no-frills visual comparison: the same prompts and source images are run through each model side by side, letting viewers judge results themselves rather than relying on marketing claims.
Key findings include JoyAI’s strict resolution preference for 1024×1024 (degrading noticeably at 1920×1080), its slower inference relative to competing models, and mixed spatial reasoning performance — a top-down view test produced an upside-down sea creature and a hallucinated chair. Qwen consistently outperformed JoyAI on spatial and compositional tasks, while the Flux Kline 4B model proved faster with four-step inference but required more detailed prompting to avoid artifacts like ghost hands. JoyAI Image Edit Plus, which accepts up to six reference images, showed stronger results when constrained to its preferred resolution.
The tutorial follows the creator’s “rodent method” of color-coded modular ComfyUI workflow groups, making the setups easy to replicate. It’s a practical reference for anyone integrating local image editing models into ComfyUI pipelines and weighing the tradeoffs among JoyAI, Qwen, and Flux Kline in terms of speed, resolution flexibility, and instruction-following accuracy.
📺 Source: Nerdy Rodent · Published July 26, 2026
🏷️ Format: Comparison
