Conditioning Ideogram4 Rebalance

Same idea as Conditioning Krea2 Rebalance, pointed at the other model. If you're running Ideogram 4 locally and want the pack's IP-Adapter-ish reference control, this is the layer EQ tuned for Ideogram's conditioning space instead of Krea 2's. The mechanism is identical; only the default weights differ, because the two models encode differently and the author dialed in separate presets.

Ideogram 4 is worth a sentence of context. It's the closed benchmark that opened its weights in June 2026 - the model everyone reached for when they needed legible text baked into an image, and it turned out to have unusually strong character and IP recall too. It shipped non-commercial, quantized-only, with a refusal filter trained in. Point being: it's a capable renderer with no native edit model, which is exactly the gap this pack tries to paper over by manipulating conditioning.

How it works

You feed it a conditioning and it re-weights each layer's contribution, then applies a global scale. Two inputs:

• per_layer_weights - a comma-separated list of floats, one per layer. The Ideogram default is 1.0,1.0,1.0,1.0,1.0,0.0,2.25,0.0,2.25,0.5,1.0,1.0,1.0 - thirteen values here, and notice it's a different shape from the Krea preset. A couple of layers are zeroed out entirely (the 0.0 entries), two are boosted to 2.25, and one is halved. That's the author saying "these layers hurt, mute them; these help, lean on them" for Ideogram specifically. Don't copy the Krea numbers over - they were tuned against a different encoder.
• multiplier - default 4. Overall gain on top of the per-layer weights. Higher is more aggressive.

Output is a single conditioning that goes into your sampler's positive slot like a normal encoded prompt.

You don't need to know which layer is which to get value here. Run the defaults, and if the rebalance is too heavy-handed, pull the multiplier down; too subtle, push it up or raise the boosted layers.

Installing it

ComfyUI Manager: search ComfyUI-Conditioning-Rebalance, install, restart. Or by hand:

cd ComfyUI/custom_nodes
git clone https://github.com/nova452/ComfyUI-ConditioningKrea2Rebalance.git

Restart afterward. No extra pip packages, no model files bundled - but you do need a functioning Ideogram 4 graph for these nodes to have something to reshape. The nodes are just tensor math; the model is the heavy part.

Common issues

The README barely documents anything - it's screenshots plus a git clone line and a few workflow.json links. Pull an example workflow from the repo and study the wiring; it'll teach you more than the text does.

This node only makes sense on Ideogram 4. Its layer weights assume Ideogram's conditioning; run Krea 2 through it and you want the ConditioningKrea2Rebalance node instead, which carries its own preset. Cross-wiring the two models' rebalance nodes is the most likely way to waste an afternoon.

And set expectations: this is a community workaround, not an official Ideogram editing pipeline. Ideogram 4 settled into a graphic-design and text-layout niche fairly fast while general attention drifted to Krea 2, so there's less community mileage on this particular node - the GSC numbers for the page are basically nil. Treat the defaults as a starting point someone found by feel, change one weight at a time on a fixed seed, and compare outputs rather than trusting any single value.