AI on the Lot 2026_Qwen-Image-Edit-2511
A bonus workflow for AI on the Lot 2026, made available for ComfyUI 101 workshop attendees to open and follow along if time allows. An image editing workflow powered by Qwen-Image-Edit-2511.
About
This ComfyUI tutorial workflow shows how to perform fast, instruction-based image edits using Qwen-Image-Edit-2511. It wires a simple I/O path—LoadImage (or VHS_LoadVideo for frames) into a compact Subgraph node (ID cdb2cf24-c432-439b-b5c8-5f69838580c9)—and ends with SaveImage. Inside the subgraph, the edit is driven by the Qwen 2.5 VL text encoder (qwen_2.5_vl_7b_fp8_scaled.safetensors) and a lightweight LoRA tuned for speed (Qwen-Image-Edit-2511-Lightning-4steps-V1.0-bf16.safetensors). The subgraph encapsulates the model plumbing so you can focus on a plain-language instruction like “make the sky overcast” or “remove the background,” then preview results quickly.
What you provide
- Reading the workflow
Runs in ComfyUI. This workflow can't run on this page. Open it in Comfy Cloud with everything loaded, or download it for your own ComfyUI.
Try in CloudQuestions & answers
Core nodes (LoadImage, SaveImage, MarkdownNote) are built-in. The workflow uses a Subgraph node (ID cdb2cf24-…) and optionally VHS_LoadVideo for video frames. If you don’t see VHS_LoadVideo, install the VideoHelperSuite (VHS) extension. Then place the provided model files in models/text_encoders and models/loras as noted.
Place qwen_2.5_vl_7b_fp8_scaled.safetensors in ComfyUI/models/text_encoders and Qwen-Image-Edit-2511-Lightning-4steps-V1.0-bf16.safetensors in ComfyUI/models/loras. Restart ComfyUI so they appear in any model/LoRA selectors inside the subgraph.
Use high-quality input images, write precise instructions (mention target objects, colors, and context), and keep the edit scope focused. If results are too strong or weak, adjust the subgraph’s exposed edit controls (e.g., strength/intensity if available) or try slightly different phrasings. Running at moderate resolutions also helps stability.
This template focuses on instruction-based global edits and does not include masking nodes. You can often localize changes by writing specific instructions (e.g., “only change the jacket color to navy”), or you can extend the workflow by adding a masking path if you need precise region control.
This workflow saves frames via SaveImage. After processing, import the numbered image sequence into a video editor or use a command-line tool (e.g., ffmpeg) to encode a video. If you prefer an in-graph export, add a video writer node and connect it after the edit stage.