Image to URL

Ever needed to hand an image you just generated to something that only accepts a URL - an image API, a chat model, a webhook, a browser? Image to URL converts an IMAGE tensor into a URL string. One output, and depending on the mode you pick, that string is either a self-contained data URL or a link to an uploaded copy. It's the glue between "I have pixels" and "this API wants a URL."

The three modes

• data - embeds the image as a base64 data: URL. Self-contained, no network, works offline, and the string is enormous (a 1024px PNG is a megabyte-plus of base64). Fine for local tools and chat models that accept data URLs; miserable for anything with URL length limits.
• kefan - uploads the image to ai.kefan.cn's upload endpoint and returns the hosted URL. Short, shareable, requires network. It's a third-party Chinese image-hosting service - the pack author's own choice of upload backend, so this is the mode that leaves your image on someone else's server.
• auto (default) - tries kefan first, and silently falls back to a data URL if the upload fails. Convenient, but it means your workflow's behavior depends on network availability and a third-party API.

The inputs

• image - the tensor to convert. Batches produce one URL per image, joined by newlines.
• mode - one of the three above.
• timeout - upload timeout in seconds, 5-300, default 30. Only matters in kefan and auto modes.

Output: url, a STRING.

When you'd reach for it

The genuinely useful pattern is API feeding: generate an image, convert it, paste the URL into a vision-capable model call or a remote service that wants to see what you made. In data mode it's also a decent way to stash a preview into a text field or a log without touching disk. But read that third-party caveat before you point it at anything sensitive - kefan and auto both upload your pixels off-machine.

Installing it

ComfyUI Manager, search "ComfyUI 1hewNodes", install, restart. Or:

cd ComfyUI/custom_nodes
git clone https://github.com/1hew/ComfyUI-1hewNodes

Pack note: this node needs requests (in the repo's requirements), but the install's heavier stuff - rembg, ultralytics, transformers - is irrelevant to it.

Gotchas

The data-URL size is the one that bites people: paste a data mode output into a URL field and you'll often find it truncated or rejected. And in auto mode, an offline workflow that "worked yesterday" will silently switch to data URLs - the node won't tell you which mode it used, so if a downstream API suddenly rejects your output, check the URL prefix. data: means it fell back.