ComfyUI Workflow: IPAdapter Plus (Simple)

IPAdapter Plus is the workhorse of image-to-image prompting in ComfyUI. Instead of describing a style or composition in text, you provide a reference image and IPAdapter transfers the visual properties of that image into your generation. The Plus version (cubiq's maintained fork) is faster, more flexible, and better-documented than the original IPAdapter.

This workflow is the official simple example from the IPAdapter Plus repo. It uses the IPAdapterUnifiedLoader node, which combines the IPAdapter model loader and the vision encoder into a single block. The result is a minimal config: one reference image in, one prompt, one output.

Key benefits

• Unified loader node, no need to chain separate model and vision encoder nodes.
• Style, composition, or face transfer, all from a single reference image.
• Compatible with SD 1.5, SDXL, FLUX, and most current base models.
• Official example from the IPAdapter Plus maintainer, not a third-party tutorial.

How it works

1. Install the IPAdapter Plus custom node via ComfyUI Manager or by cloning the repo into your custom_nodes folder and running pip install -r requirements.txt.
2. Download the model files: an IPAdapter model (the ip-adapter-plus_sdxl_vit-h.safetensors family is a good starting point) goes in models/ipadapter, and the CLIP vision encoder goes in models/clip_vision.
3. Load the workflow JSON from the linked GitHub URL.
4. Drop a reference image into the LoadImage node. This is the style or composition source.
5. Write a text prompt. IPAdapter handles the visual properties, the prompt handles the content.
6. Queue the prompt. Generation runs at base-model speed plus the IPAdapter overhead, typically 5-15%.

The simple workflow is a starting point. For more advanced use cases (face-only transfer, weight type experimentation, regional conditioning), the IPAdapter Plus examples folder has 20+ official workflows covering every common pattern.