Image Switch
Flip between two images mid-workflow
- Image_1
- Image_2
- Image
- Selection
Image Switch is the two-slot router for IMAGE tensors. One selection, two image inputs, one image out. It's the sort of node that sounds too boring to matter until you've got a workflow where you want to swap between two source images without deleting wires - at which point it becomes the thing you use every day.
Where you'll actually use it
The obvious cases: img2img where you want to toggle between two reference images, ControlNet where you're testing two different pose or edge inputs, or any A/B comparison of two source photos. A really common pattern is loading one image into the img2img slot and a second into a ControlNet reference, then flipping to see which combo works. Since the node's selection is an INT, you can even automate the flip - a random seed feeding Selection gives you a "random image each run" effect without any scripting.
How it works
Same DNA as every node in this pack's switch family - a match on the selection:
case 1: return (Image_1, Selection,)
case 2: return (Image_2, Selection,)
Inputs:
- Selection - 1 or 2. Feed it an integer from anywhere, or just click between values.
- Image_1 and Image_2 - IMAGE tensors from Load Image, a VAE decode, an upscaler, whatever produces pixels.
Outputs:
- Image - the selected tensor. Drop it into your sampler's latent encoder, ControlNet application, or preview.
- Selection - the choice, echoed out so downstream logic can know which source won.
Where people get burned
Same two gotchas as the other switches in this pack. Image_2 is optional - leave it empty, pick 2, and you'll hand a None to whatever's downstream, which usually surfaces as a cryptic error in the node you connected it to. Wire both inputs before you start flipping.
And remember that selecting ≠ skipping: ComfyUI computes both connected branches regardless, so Image Switch won't save you from running a slow upscaler on both images. It picks the output; it doesn't mute the loser.
One practical note on batches: each input is a full IMAGE tensor, which may contain multiple frames or a batch. The switch passes the whole thing through untouched, so batching behavior is entirely up to whatever produced the images.
Installing it
ComfyUI Manager → search "HavocsCall" → install → restart. Manual install:
cd ComfyUI/custom_nodes
git clone https://github.com/HavocsCall/comfyui_HavocsCall_Custom_Nodes
Restart and it'll be under HavocsCall/Switches. No requirements.txt, no model downloads - the pack only uses what ComfyUI already ships, so this installs in seconds.
Inputs (3)
| Name | Type | Default | Description |
|---|---|---|---|
| Selection | INT | 11–2 | — |
| Image_1 | IMAGE | — | |
| Image_2opt | IMAGE | — |
Outputs (2)
| Name | Type | Description |
|---|---|---|
| Image | IMAGE | — |
| Selection | INT | — |