Descriptions:
Nerdy Rodent walks through a fully reorganized ComfyUI workflow for Qwen Image Edit 2511, Alibaba’s late-2025 image editing model, demonstrating a range of upgrades over earlier Qwen edit pipelines. The tutorial introduces a custom split-sigma sampler group, restructured latent and image upscaling options, and LLM-powered prompt enhancement via Ollama — with Gemma 3 and Qwen 3VL highlighted as lightweight vision-capable options.
Three new LoRAs are covered in detail: a deblur/restore upscaling LoRA, an in-context editing LoRA, and a pose-transfer LoRA that eliminates the need for a separate pose preprocessor. The presenter demonstrates masked vs. unmasked editing side by side, showing how masking controls color drift during head or object compositing, and compares latent upscaling against standard image upscaling across multiple examples. Depth, Canny, and pose preprocessors are also shown as optional conditioning inputs.
The video targets intermediate-to-advanced ComfyUI users and serves as a practical reference for building flexible, production-quality local image editing pipelines. With support for 5K output resolution and multiple workflow modes — plain text-to-image, multi-reference editing, and pose-guided generation — it provides enough technical specificity to be directly reproducible at home.
📺 Source: Nerdy Rodent · Published December 31, 2025
🏷️ Format: Tutorial Demo
