ComfyUI Workflow: Qwen Image Edit 2511

Qwen Image Edit 2511 is the newest point release of Alibaba's Qwen vision-language image editing model. Feed it up to three input images and a natural-language instruction. It produces edits that stay consistent across the whole frame: same character, same product, same scene, only the parts the prompt asks to change.

This ComfyUI workflow wraps the 2511 checkpoint using the GGUF loader. The bf16 weights need 24GB VRAM. The Q4_K_M GGUF runs on 12GB. The TextEncodeQwenImageEditPlus node accepts the prompt plus up to three image inputs, so a multi-reference edit (face + clothing + background) goes through one conditioning block instead of three stacked VAE Encode passes.

Key benefits

• Multi-image conditioning in a single node, no need to chain three separate VAE encode passes.
• GGUF Q4_K_M quantization runs on 12GB VRAM (RTX 3060 and up).
• Higher consistency on character and product edits than the 2509 release, per community reports.
• Native 1328x1328 resolution support without resampling artifacts.
• Uses a ComfyUI-Org-blessed loader pattern, so updates to the model release slot in cleanly.

How it works

1. Download the model files: the Q4_K_M GGUF for the 2511 edit model and a Qwen 2.5 VL 7B GGUF for the text encoder. Drop both into your ComfyUI models/diffusion_models and models/clip folders.
2. Load the workflow JSON from the GitHub repo. The three LoadImage nodes (left, center, right) take your three reference images.
3. Write the instruction in the TextEncodeQwenImageEditPlus node. "Transform image 1 into real life" is the example prompt. The model interprets image 1 as the target and images 2 and 3 as style or context references.
4. Queue the prompt. Generation takes around 48 minutes on a 12GB card at 1024x768, faster on higher VRAM.
5. Review and iterate. If the result drifts from the reference, lower the CFG slightly and re-queue.

Where the 2511 improves on 2509: character consistency across edits and handling of small text inside images. If you are doing product swaps, scene restyling, or character redraws and the 2509 result felt slightly off, the 2511 is the upgrade.