Get ConDelta from Prompt

The whole point of a ConDelta is that it's the difference between two prompts - subtract "a city" from "an art deco city" and you're left with a vector that encodes "art deco"-ness. Building one by hand means picking a sensible baseline prompt yourself and wiring up a subtract. GetConDeltaFromPrompt skips that step: give it one prompt, and it diffs it against a built-in baseline for you.

How it works

You give the node a prompt and a clip model, and it encodes your text the normal way. The difference from a plain CLIP Text Encode is prompt_type: it picks which baseline pool your prompt gets compared against before the delta is produced. The choices are misc, anime, custom, and none. Pick misc or anime and the node uses its own built-in generic-prompt set for that domain as the "before" side of the subtraction - so a portrait prompt run through the anime baseline should isolate what's actually distinctive about your prompt relative to typical anime prompts, rather than relative to nothing. Pick custom and you supply your own baseline list via custom_prompts, one prompt per line, which is the honest way to read that field's own tooltip. none presumably skips the baseline subtraction, in which case what comes out is closer to a plain encoded prompt than a genuine delta.

The inputs and outputs that matter

• prompt - the text you want to extract as a delta.
• clip - the CLIP model to encode with.
• prompt_type - default misc, the other choices being anime, custom, none. This is the setting that decides what your prompt gets compared against.
• custom_prompts - multiline, only meaningful when prompt_type is custom; the tooltip is explicit that it's "one per line."

Output is a single CONDITIONING - a delta, ready to feed into ConditioningAddConDelta (or ApplyConDeltaAutoScale's sibling for live deltas) to apply it to an actual generation, or into SaveConditioningDelta to keep it for later.

How to install it

Via ComfyUI Manager: search "ComfyUI-ConDelta", install, restart. Or:

cd ComfyUI/custom_nodes
git clone https://github.com/envy-ai/ComfyUI-ConDelta

then restart. No extra models or dependencies to fetch for this node specifically.

Common issues & troubleshooting

Match the baseline to your domain. If your prompt is an anime character but you leave prompt_type on misc, the subtraction is measuring your prompt against a baseline that doesn't share its vocabulary, and the resulting delta is more likely to carry noise from that mismatch than a clean concept. Switch to anime when that's the domain you're actually working in.

custom is the escape hatch when neither built-in baseline fits. If you're working in a niche outside both misc and anime - a specific art style, a technical illustration domain, whatever - write a handful of representative baseline prompts into custom_prompts yourself rather than fighting the built-ins.

This produces a delta, not a finished conditioning. Unlike QuickConDelta, which blends a concept straight into an existing conditioning in one step, this node's output still needs to go somewhere - chain it into ConditioningAddConDelta or ApplyConDelta's file-based sibling before you'll see any effect on a generation.