Qwen-Image-Layered

Qwen-Image-Layered is the image model that made the community joke that Photoshop's days were numbered. Instead of generating a flat JPEG, it outputs an RGBA image decomposed into layers - the foreground subject separated from the background with transparency, ready for compositing. The KB nails the framing: the release "read to the community as the beginning of the end for Photoshop." For anyone doing cutouts, character sheets, or sticker-style assets, a model that just returns a layered result is a workflow revolution compared to running an image through a background-removal node afterward.

The original Qwen-Image-Layered weights are Apache 2.0, so you could run this locally - but the node in front of you is the cloud version, served through Runware. The layered output lands back as a TIFF, which is exactly what you want for RGBA work.

What you set

referenceImages and positivePrompt are both required - this is an image-input model, not pure T2I. You feed it a picture and describe the layered result you want.

• lora - a RUNWARE_LORA socket for community LoRAs (the open family has a real LoRA ecosystem).
• steps / scheduler / CFGScale - gated diffusion controls. scheduler is a monster dropdown (76 entries) because it exposes the underlying diffusers schedulers by name. Leave them at defaults until you know what you're doing.
• settings.trueCFGScale - a separate, "true" CFG scale; the tooltip says higher values improve prompt adherence at the cost of quality. This is the Qwen-Image quirk: it uses a non-standard guidance formulation, hence the "true" CFG slot.
• seed - exposed and reproducible.
• numberResults - up to 20 per run.

Output is image as IMAGE in TIFF format - save as TIFF and you keep the layers. outputFormat is locked to TIFF here (one choice), which is the giveaway that this node is fundamentally about preserving transparency.

Installing

Part of the Runware/ComfyUI-Runware pack.

ComfyUI Manager: search Runware, install, restart.

Manual:

cd ComfyUI/custom_nodes
git clone https://github.com/Runware/ComfyUI-Runware
pip install -r ComfyUI-Runware/requirements.txt

Deps are runware-sdk, pillow, soundfile. Runware API key from runware.ai/api-keys (Settings panel or RUNWARE_API_KEY).

Where people get burned

Two things. First: since this is a cloud call, "send the image to Runware" means your source image leaves your machine - a real consideration for proprietary or sensitive art. Second, the output being TIFF-only catches people: if you wire this into a normal image pipeline that expects JPG/PNG, you may get a surprise, and downstream nodes that can't read multi-layer TIFFs will flatten it anyway. If you want the layers, save the TIFF immediately; if you just wanted a cutout, a background-removal node might be cheaper than a billed layered generation.