Qwen Image Edit Plus LoRA

Most cloud image APIs are strictly "prompt goes in, picture comes out" - no LoRAs, no personality. Qwen Image Edit Plus LoRA is the exception: it takes up to three LoRA model files alongside your edit instruction and reference images, and applies them server-side. For $0.02 a pass, that's effectively "my trained style, on the cloud, with an instruction editor doing the work."

It's part of the ERPK Collection, the community pack wrapping the WaveSpeed AI API. The Qwen Edit family is the open instruction editor you may already run locally - this is the same lineage, hosted, with LoRA support bolted on.

How it works

You give it reference images, an edit instruction, and one to three LoRA file URLs. The catch that trips everyone up: the LoRAs have to be reachable via URL - a local file path on your machine won't work unless the API can reach it. WaveSpeed fetches the LoRA weights, applies them at the given scale, and runs the edit. The result comes back as a ComfyUI image IMAGE tensor.

The model combo offers Qwen Edit Plus LoRA or Qwen Edit 2511 LoRA - the 2511 variant adds multi-person editing and improved consistency, which is exactly the upgrade you want for group shots or character work.

The inputs that matter

• images - reference images to edit, comma-separated URLs or paths, up to 3. Required.
• prompt - the modification, in Chinese or English. Qwen Edit is genuinely bilingual, like the rest of the family.
• lora_1_path - URL to the first LoRA file. Required. This is the odd one out in the pack: most image nodes don't take LoRAs at all.
• lora_1_scale - 0.0 to 4.0, default 1.0. Treat it like the strength slider you know from local workflows; crank toward 4.0 for maximum influence, keep it near 1.0 for subtle.
• lora_2_path / lora_3_path (plus their scales) - stack more LoRAs. Style LoRA + character LoRA is the classic combo.
• width / height - 256 to 1536, default 1024. seed is real; output_format defaults to jpeg.

Install

cd ComfyUI/custom_nodes
git clone https://github.com/eRepublik-Labs/comfyui-nodes-erpk.git erpk
cd erpk
pip install -r requirements.txt

Or ComfyUI Manager → search erpk → ERPK Custom Nodes. WaveSpeed key with credits via right-click canvas > ERPK Settings (Settings > ERPK > API Keys).

Common issues

• "LoRA file not found" / failed fetch - the path must be a URL WaveSpeed can download. Host the file somewhere public (or otherwise reachable) rather than pointing at a local drive.
• LoRA having no visible effect - check the scale. At 1.0 on a strong edit instruction, the instruction can overwhelm a weak LoRA. Bump the scale before you blame the file.
• Multi-person edits falling apart - switch the model to Qwen Edit 2511 LoRA; that variant is explicitly tuned for it.
• Three images and a complex instruction that ignores one - name each reference's role in the prompt, same discipline as the Max Edit node.

This is the niche node for people who've trained LoRAs and refuse to give them up when they move cloud-side. If that's you, it's basically a superpower. If you've never trained one, the plain Qwen Edit Plus node (no LoRA inputs) is simpler and cheaper to start with.