NS WaveSpeed Qwen Edit Plus LoRA

If Qwen Edit Plus is the workhorse of the Qwen edit family, this is the full harness: three reference images and three server-side LoRA slots, all in one node. It's the node you build a workflow around when an edit needs identity from references, style from a LoRA, and exact dimensions - in a single hosted call.

It's part of the Symbiotica pack's neuralsins/WaveSpeed family, and it's the densest input surface of the whole Qwen set. Same mental model as the other LoRA nodes: the LoRA paths are model identifiers on WaveSpeed's side (the tooltip example is flymy-ai/qwen-image-realism-lora), not local files. You're composing style adapters server-side.

How it works

prompt plus up to three reference image_url_1–image_url_3, plus up to three LoRA paths with per-LoRA strength, and the edited result returns through output_image (IMAGE) via WaveSpeed's /api/v3/wavespeed-ai/qwen-image/edit-plus.

The inputs that matter:

• prompt - edit instructions, Chinese or English.
• image_url_1–image_url_3 - references; max 3 total, the first is the one that matters.
• lora_1_path / lora_2_path / lora_3_path - the style adapters, by path. The second and third are optional.
• lora_1_scale–lora_3_scale - influence, 0.0 to 4.0 (note: the Plus LoRA tier widens the range from the plain Qwen Edit LoRA's 2.0 ceiling). Default 1.0 each.
• size - 7 presets from 1328x1328 (1:1) down to 2:3, default 1:1.
• custom_size - width*height string (e.g. 1920*1080) that overrides the dropdown.
• seed, output_format (jpeg/png/webp), enable_base64_output - the usuals.

enable_sync_mode defaults on.

Installing

cd ComfyUI/custom_nodes
git clone https://github.com/symbiotica-ai/comfyui-nodes.git symbiotica
pip install -r symbiotica/requirements.txt

Restart or install via ComfyUI Manager (search Symbiotica), and wire an NS WaveSpeed Client with your WAVESPEED_API_KEY into client.

Common issues

• Three LoRAs at 1.0 is a fight. Each adapter pulls the result its own way; stacked at full strength they can wash the image out or cancel each other. Start the extras at 0.3–0.5.
• References + LoRAs blur together fast. More inputs isn't automatically better composition. If identity drifts, cut references before you cut prompts.
• Wrong LoRA path = call-time error. There's no local browser for these - the API rejects a path it doesn't have.
• custom_size overrides the dropdown. Set both and custom wins; clear it to use presets.
• URLs, not tensors. The standing rule for every image input.

The honest take: this is the node you graduate to when one Qwen edit isn't enough and you'd rather compose references, style, and size in one paid call than chain three nodes. Just budget the prompt time - with this many dials, garbage-in is expensive garbage-out.