🎨 PromptFix Output (To Photoshop)

PromptFix Input drags your Photoshop canvas into ComfyUI; this node is the return leg. You run your workflow, and the finished render gets dropped back into Photoshop as a brand-new layer - no file dialog, no drag-and-drop, it just appears in your layer stack. If you've ever done a round-trip of "select in Photoshop → edit in ComfyUI → import the result," you know the hand-off is where the friction lives. This node kills it.

It's also worth being clear about what it is: a thin subclass of ComfyUI's built-in SaveImage that does one extra job - phoning home to Photoshop when the save is done. That's the whole trick, and it's why the node is so unassuming.

How it works

Under the hood it's local and boring, in a good way. The node saves your image as a PNG (with a _promptfix_ prefix, into ComfyUI's temp folder), then makes a synchronous HTTP call to the pack's own backend route - http://127.0.0.1:{port}/promptfix/renderdone?filename=.... The backend reads that file, base64-encodes it, and pushes it over the same WebSocket the input side uses to the Photoshop plugin, which imports it as a layer. Nothing leaves your machine, no API key, no cloud round-trip. Adobe sells you Firefly for that exact workflow; here you're running open models locally and still getting the Photoshop canvas experience.

The input that matters

Just one visible input, and it's the only one you'll ever touch:

• output_image (IMAGE) - the final rendered image. Wire in whatever ends your editing chain: the decoded output of a KSampler for plain img2img, or the result of an image-edit model like Qwen Image Edit or Flux Kontext (the README's own suggestions for what runs between the two nodes).

It's an output node, so it's terminal - you don't route its output anywhere. The images output you see in the schema is just the standard save-result passthrough from the SaveImage it inherits from; ignore it. If you want the same image saved to your regular output folder and sent to Photoshop, put a plain Save Image node before it in the chain rather than expecting this one to do double duty.

Installing it

Same pack as the input node, so if you already have PromptFix Input you're done - install once. ComfyUI Manager search "Promptfix", or:

cd ComfyUI/custom_nodes
git clone https://github.com/shaaraa/ComfyUI_Promptfix.git

Restart ComfyUI. No dependencies beyond what ComfyUI ships, no models to download. The one thing install can't give you is the paid Photoshop plugin (NanoBanana Seedream AI Plugin) - the bridge's other end, and it's not free.

Gotchas

• No Photoshop connected? It still "works." The node saves the PNG and prints [PromptFix] Error notifying backend... to the console - non-fatal by design. But it saves to the temp folder, which ComfyUI wipes on restart. Don't rely on it as your only save point unless the plugin is actually receiving.
• The notify is synchronous. It's a fast local call, so you'll never notice, but it does block the queue for a moment after each save.
• This is a very new, very small pack (April 2026, zero stars, nobody talking about it on reddit). The Photoshop half is a commercial Gumroad product, so if you hit a bug, you're leaning on the author's GitHub Issues, not a community - temper expectations.
• If the layer doesn't appear, check the plugin side first: the WebSocket connection has to be alive in Photoshop, not just in ComfyUI.