Can I add my own custom style to the dropdown?

You can extend the presets only if the processing node (d4c06663-cb1a-4a41-a36b-f09c93e55e7e) recognizes additional style keys. Update the JSON presets feeding JsonExtractString and add the new option to CustomCombo. If the node doesn’t support the new key, it will fall back or fail.

Does this require a GPU?

It depends on how the custom processing node is implemented on your system. Many stylization nodes prefer a GPU and will run much faster with one, but some builds can run on CPU at reduced speed.

What image size works best?

Portraits at 512–1024 px on the shortest side typically produce clean lines and colors. Crop to the head and shoulders when possible, use even lighting, and avoid heavy motion blur.

Can I batch-process multiple photos?

This graph focuses on single-image input via LoadImage. For batches, duplicate the branch or swap in a batch-capable loader node and feed the same CustomCombo/JSON preset mapping into the processor.

Photo to Cartoon Style Caricature

This ComfyUI workflow turns a regular photo into a cartoon-style caricature using one of 15 preset looks. It starts with LoadImage for your source photo, passes your style choice from a CustomCombo dropdown, extracts the matching preset parameters via JsonExtractString, and assembles any required identifiers with StringConcatenate. The core stylization happens in a custom processing node (UUID d4c06663-cb1a-4a41-a36b-f09c93e55e7e), which takes the source image plus the selected style key and outputs a single stylized frame. You can inspect results live with PreviewAny and save the final image via SaveImage.

Under the hood, the preset system keeps results consistent and repeatable across images. The JSON-driven mapping decouples the user-friendly style names from the actual configuration values the processing node expects, making it easy to swap styles without touching low-level settings. Because the heavy lifting is encapsulated inside the d4c06663-cb1a-4a41-a36b-f09c93e55e7e node, the graph stays simple and fast to use—no additional model loading or sampler tuning required.

FAQ

Frequently Asked Questions