Deepseek R1 KlingAI Text/Image to Video Node

Here's a node that reads like it does a lot and quietly does something much smaller. DeepseekR1KlingAINode won't generate a video for you, and despite the display name it won't even send an image anywhere. What it actually does: logs into your KlingAI account with your email and password, opens the site's built-in DeepSeek assistant, asks it to write or refine a prompt, and streams the answer back. Think of it as "prompt-writing through your Kling account," not "image to video."

Why bother? The selling point is the free tier. Kling's DeepSeek chat rides on the account and doesn't burn the credits you'd spend on actual generations - and people have been motion-prompting Kling with DeepSeek for a while now, by hand in the browser. This node just automates the browser part and drops the result into your workflow as a string.

How it works

Mechanically it's a web client, not a proper API client. It POSTs your email and password to id.klingai.com, pulls three session tokens (portal_st, portal_ph, user_id), stashes them in environment variables, then opens a chat session at api-app-global.klingai.com with modelType=deepseek and streams the SSE reply. It also queries your point balance and returns it as a second output.

Inputs and outputs

Four inputs, all required:

• email and password - your KlingAI login. The password field is masked (display: password), but it's still your real credential sitting in a node.
• prompt - the text you want DeepSeek to work from. The default is un auto de color rojo; the author's Spanish shows up all over this pack.
• model - a dropdown: Text to Image / Text to Video / Image To Video.

Two outputs: generated_text (STRING, the DeepSeek reply) and user_points (STRING, your Kling balance). The honest read on that dropdown: the "model" isn't you choosing a renderer - it's just the referPageCode (t2i / t2v / i2v_tail) that Kling's DeepSeek assistant uses as context. It flavors the generated prompt; it doesn't pick a video model.

The part you need to hear before typing a password in

This is an unofficial, reverse-engineered call to a web endpoint. The code ships with a hardcoded __NS_hxfalcon token and browser fingerprint headers that were valid when the pack shipped - a single commit from May 2025 - and when Kling changes their site, this node breaks with no maintainer to fix it. And it's asking for your actual KlingAI password, which is a real credential to hand to a random custom node. If you use it, the sane move is a throwaway account, or changing the password afterward. The tokens at least stay in the process's environment rather than a file.

Installing it

Same story as the rest of the pack - ComfyUI Manager, search "ComfyUI-GeminiImageToPrompt", or:

cd ComfyUI/custom_nodes
git clone https://github.com/santiagosamuel3455/ComfyUI-GeminiImageToPrompt

Then restart. No models to download; the requirements are requests, google-generativeai, and Pillow. Where it fits: wire it after a Gemini prompt node or ShowGeneratedText so you get a polished, edited prompt without leaving the graph. Just don't expect a video generator - it's a free prompt assistant that happens to live in Kling's building.