WaveSpeedAI Qwen Image Edit LoRA

The thing that turned Qwen-Image-Edit from "another editor" into a movement is its LoRA library - style, pose, and expression adapters trained on the editor itself, which is exactly why they don't work on the base Qwen-Image model. Running that stack locally means a ~20B editor plus a growing pile of adapter files. This node runs the editor on WaveSpeed AI and lets you point LoRAs at it by name, from inside ComfyUI, with no model download at all.

What it is

A hosted Qwen-Image-Edit node that accepts up to two LoRA adapters. You give it a prompt, an image_url, and LoRA references, and it returns the edited image as an IMAGE tensor. The LoRA inputs are worth reading carefully: lora_1_path expects a path, not a file - the tooltip's example is flymy-ai/qwen-image-style-lora, which is the user/model format for a model on the Hub. The API resolves it server-side.

How it works

Same mechanism as the plain Qwen Image Edit node - POST to the Qwen image edit endpoint with your image URL, prompt, seed, and output format - plus the LoRA paths and scales in the payload. Each lora_N_scale runs 0.0 to 2.0 (default 1.0), and 2.0 is aggressive: the style will overwhelm the source image. The returned image URL is fetched back to your machine as an IMAGE tensor.

The inputs that matter

• prompt - the edit instruction, in plain language (Chinese and English both supported).
• image_url - source image as a public URL.
• lora_1_path / lora_2_path - up to two LoRAs in user/model format. This is where the node earns its keep: one box, no manual LoRA file management.
• lora_1_scale / lora_2_scale - influence per LoRA; start at 1.0 and back off if the style fights the edit.
• seed, output_format (jpeg/png/webp), enable_sync_mode (default on).

Output name is output_image, type IMAGE.

Install & setup

cd ComfyUI/custom_nodes
git clone https://github.com/razvanmatei-sf/razv-wavespeed.git
cd razv-wavespeed
pip install -r requirements.txt

or ComfyUI Manager → "razv-wavespeed" → Install, restart. API key from wavespeed.ai in config.ini (from config.ini.tmp), the Client node, or WAVESPEED_API_KEY. Deps: requests, pillow, numpy<2.0.0, torch.

Common issues

• Invalid LoRA path error - the path must be either a full URL or username/model-name. A bare filename won't resolve.
• Style too strong - drop the LoRA scale, don't fight it with the prompt.
• "No output received" - image URL not publicly reachable, or job timed out.
• 401 - key issue.

Honest take: if you already run Qwen-Edit locally with a LoRA folder you've curated, this node is a convenience you may not need. Where it shines is trying out the community LoRA library - you can test five different style LoRAs in five renders without downloading a single adapter. That's a genuinely good way to window-shop before committing to a local setup.