Gigachad Resolution Picker

Gigachad Resolution Picker is the pack's take on EmptyLatentImage with a bonus: besides a blank latent, it also hands you the WIDTH and HEIGHT as plain integers. That third trick is the whole reason it exists - a surprising number of workflows need the dimensions as numbers to feed into other nodes, and the stock empty-latent node doesn't give them to you.

What you set

Only three widgets, and two of them are the obvious ones:

• width (default 1024, step 8) - image width. The step of 8 keeps you aligned with the latent-space grid, which matters for SD1.5/SDXL.
• height (default 1024, step 8) - image height, same idea.
• batch_size (default 1) - how many empty latents to create, i.e. how many images to generate in one queue.

The tooltip on width/height adds something genuinely useful: "set by the picker UI or connect from outside for subgraph override." The width and height are optional inputs as well as widgets, which means you can wire another node's output into them and drive the resolution programmatically - handy when you're building subgraphs or chaining picks.

The three outputs

• WIDTH (INT) and HEIGHT (INT) - the numbers you chose, for any node that wants them (an upscale-by-factor calc, a prompt that mentions resolution, a conditioning setup).
• LATENT - a torch.zeros tensor shaped [batch, 16, height/8, width/8]. The 16 channels and the 8x spatial compression are the SD1.5/SDXL VAE's native latent format, so this plugs straight into a KSampler's latent_image with no conversion.

The pack README's minimal workflow wires all three straight into the KSampler, and that's the intended loop: pick resolution → get latent → sample → decode.

Install

cd ComfyUI/custom_nodes
git clone https://github.com/Winnougan/comfyui-gigachad.git

or search comfyui-gigachad in ComfyUI Manager and restart. No extra dependencies - pure ComfyUI.

Honest verdict

The cosmetic part is loud - the node renders with a gold "GIGACHAD" glow and animated sparkles around it, which is charming for about thirty seconds. The functional part is genuinely reasonable: if you've ever built a workflow where you needed both an empty latent and the dimensions as numbers, this saves you a GetImageSize-style detour or a math node. If all you want is an empty latent, the stock EmptyLatentImage does the same job with less theater. And remember - this one is for SD1.5/SDXL. For LTX video work, the same pack ships a dedicated LTX picker with the right latent format for that model family, and that's the one you want.