The biggest headache when making posters and packaging is that models can’t render text clearly and you have zero control over layout. Ideogram 4.0 just dropped with weighted prompts, a 9.3B model trained from scratch for text-to-image, and ComfyUI had native support on day one. I’ve been messing with its structured JSON prompts the past couple days, and it’s way better than plain text.
Three things plain text can’t do: first, a color palette—up to 16 hex colors per image, 5 per element; second, using normalized coordinates from 0–1000 to pin down element positions; third, you can separately specify the text content to render and its style description, writing them as two separate things.
In practice, it really rewards you for “clearly defining relationships”—the more specific you get with positions, colors, and individual style descriptions for each element, the closer the output matches. That’s because it was trained on heavily detailed scene descriptions; the more verbose you are, the more obedient it gets. One thing to watch: the safety filter is baked into the model weights, not added by ComfyUI. If you get blocked, tweak your prompt—don’t report it as a bug. Upgrade to ComfyUI 0.24.0 and you can run the workflows from the template library.