Load Ideogram4 V9 Trigger Embedding

A trigger activation on the community Ideogram4 models isn't one file - it's two artifacts, and this node loads the first half. Gen2_LoadIdeogram4V9TriggerEmbedding pulls a strict V9 [4,H] trigger embedding out of your ComfyUI/models/gen2/ folder and hands you a GEN2_IDEOGRAM4_V9_TRIGGER_EMBEDDING object. It's the "semantic" side of the trigger: the trained vector that makes the model understand your concept when the trigger token appears in the prompt.

Why it's two nodes and not one

These V9 triggers were trained to fire hard - one literal token in your prompt should pull the whole trained concept. To get there, the pack splits the mechanism: an embedding for the semantic meaning, and a separate text-encoder module-LoRA (loaded by Gen2_LoadIdeogram4V9TriggerTEAdapter) that nudges the Qwen3-VL encoder itself. This node is deliberately single-purpose: it validates that the file you point it at is really a strict [4,H] V9 embedding, and it loads nothing else. The strictness is the point - legacy single-token embeddings and shared-residual artifacts are rejected, because they don't match what the rest of the V9 chain expects.

The one input

• artifact_name - a dropdown populated from files in ComfyUI/models/gen2/. If your folder is empty, the dropdown is empty, and that's usually the whole problem: people install the pack, grab a LoRA file, and wonder where the node gets its inputs. It only reads this one folder, so your trigger artifacts have to live there.

The single output, embedding, goes straight into Gen2_ComposeIdeogram4V9TriggerActivator - you don't use it directly in a text encoder. That's a mental shift if you're used to the old embedding workflow where you just dropped a .pt file onto the prompt: here, loader → composer → text encoder is the mandatory chain.

Install

ComfyUI Manager: search ComfyUI-gen2. Or:

cd ComfyUI/custom_nodes
git clone https://github.com/petmycat/ComfyUI-gen2.git
cd ComfyUI-gen2
pip install -r requirements.txt

Restart, then put your V9 embedding artifact in ComfyUI/models/gen2/. The V9 nodes need a real Ideogram4 + Qwen3-VL backend (the open-weights Ideogram 4 fp8/nf4 checkpoints) and refuse to run on a plain Qwen3 CLIP - remember the Ideogram 4 weights ship under a restrictive non-commercial license, so check what you can legally do with what you generate.

The honest caveat

This whole V9 trigger path is young. The pack author says real artifact parity and visual smoke testing remain a release gate - the contract and lifecycle tests are in place, but golden-tensor validation is pending matching artifacts. So if a trigger underperforms, don't immediately blame your weights; the implementation itself is still being proven.