Image Prompt Loader 🖼️ (TensorVizion)

Image prompting - "make it look like this reference" - needs two model files, not one: a CLIP vision encoder that reads the image, and a style model that injects what it learned into your conditioning. ComfyUI core has separate loaders for each, and because they're always used together, this node loads both in one shot. It's the setup half of OmniNodes' image-conditioning pair; Image Prompt Apply is the half that does the actual work.

What it loads and where from

Two dropdowns, two standard ComfyUI model folders:

• clip_vision_name - a CLIP vision encoder from ComfyUI/models/clip_vision/.
• style_model_name - a style model from ComfyUI/models/style_models/.

Both are enum dropdowns populated from those folders, so you pick by filename. The node is a thin wrapper around core's CLIPVisionLoader and StyleModelLoader - the load behavior is identical to stock, it just does both in one place and hands you both outputs. The summary output confirms what you loaded.

Outputs are clip_vision (CLIP_VISION) and style_model (STYLE_MODEL) - wire both into an Image Prompt Apply node alongside your reference image, plus the summary string.

What a "style model" even is

This is ComfyUI's native, dependency-free route to the same idea IP-Adapter popularized: keep the text path intact, add a separate image path so a reference image can steer style or content without retraining anything. The style model files you drop in style_models/ are usually IP-Adapter-style models - on the SD1.5/SDXL era, which is where this mechanism lives, that's the well-trodden ground. No extra custom-node pack, no InsightFace dependency: just two files and this loader. (Modern Flux-era models route the same jobs through in-context editors like Kontext or Redux instead, so expect this pair to be an SD/SDXL tool.)

Install and gotchas

Same pack install as everything else in OmniNodes:

cd ComfyUI/custom_nodes/
git clone https://github.com/TensorVizion/OmniNodes

Restart ComfyUI, or install "OmniNodes" via ComfyUI Manager. No extra pip packages and no model downloads from the node itself - the model files are yours to supply.

The gotcha is the obvious one: the dropdowns only show files in ComfyUI's registered clip_vision/ and style_models/ folders. If your style model doesn't appear, that's where it needs to live - ComfyUI won't see it elsewhere. And remember, this node only loads the pair; without an Image Prompt Apply node downstream to encode your reference image and push it onto conditioning, the two outputs do nothing on their own. Loader → Apply → KSampler is the shape of a working graph.