Kling Image Omni

Kling Image Omni is the pack's image-conditioned generation node: you hand it a reference image plus a prompt, and you get a new image that incorporates both - subject reimagined, style transferred, composition edited - all driven by how you reference the image inside the prompt. It's Kling's omni-image model (the kling-image-o1 engine) exposed in one node, and it's the closest thing in this pack to the "put my character in this scene" pattern people actually want from reference-based generation.

The mechanism is what makes it interesting: the reference isn't a separate strength slider, it's referenced by name. You write your prompt using @image1 to point at your input image - "the woman from @image1, now standing in a neon-lit Tokyo alley at night." The model reads the prompt, pulls the reference, and generates accordingly. It's a different mental model than ControlNet-style reference handling, and once you've got it, it's actually quite flexible.

How it works

Inputs:

• auth - the KLING_AUTH object.
• prompt - multiline. Use @image1 to reference the input image. This is the key skill; the tooltip spells it out: "Use @image1 to reference input."
• image_1 - the reference image (an IMAGE tensor). One reference slot in this version; the pack's video-side Omni node uses the same @image1, @video1 referencing syntax for its multi-input version.
• aspect_ratio - one of the seven Kling ratios, default 1:1.
• resolution - 1k or 2k.

Outputs: image (IMAGE), url, and task_id - the standard image trio.

How to actually prompt it

• Always reference the image. If your prompt doesn't mention @image1, you're basically doing plain text-to-image with an unused input sitting there. The reference is invoked by naming it.
• Describe the transformation, not the whole scene. "The character from @image1 in a cyberpunk market, rain-slicked streets, cinematic lighting" tells the model what changes (setting, mood) while the reference anchors what stays (the character).
• Keep the reference clean. A focused subject on a simple background transfers far better than a cluttered group shot. Garbage in, garbage out - the reference constrains identity, and a messy one constrains it messily.
• It's one reference, not a collage. For multi-image consistency, Kling's Multi-Shot node is the better tool; Image Omni is about transforming a single input.

Where it fits

This is your identity-preservation-ish node in the pack's image line: keep a character's look while changing everything around them. It pairs naturally with Image Gen (generate the character, then Omni it into scenes) and feeds Image-to-Video for the animated follow-through. If you're building "consistent character, many settings" work, this is the node you'll reach for before the more specialist ones.

Common issues

• "Image ignored / result looks like plain T2I" - you forgot @image1 in the prompt. It's the single most common miss, and the fix is one token.
• Weak resemblance - the reference is a cluttered or heavily filtered photo. Clean it up, or accept that strong style-transfer prompts will push the identity further away.
• One-reference limit surprises you - that's this node; the multi-reference omni workflows (with @image1, @image2, @video1) live on the video side of the pack.

Install

Part of ComfyUI-Kling-Direct:

cd ComfyUI/custom_nodes
git clone https://github.com/IxMxAMAR/ComfyUI-Kling-Direct

or ComfyUI Manager → search "Kling Direct" → install → restart. No models, no GPU, no extra dependencies. The learning curve is one sentence long - remember the @image1 - and then it's a very capable reference node.