EA 3-Way Image Compare
Three-way compare for LoRA weight hunting
- image_1
- image_2
- image_3
- comparison
When you're sweeping a LoRA weight, two panels aren't enough. Base model, weight 0.5, weight 1.0 - that's the whole test, and you want to see all three at once, not remember what the middle one looked like. EA 3-Way Image Compare stitches three captioned images into a single output so the progression is visible in one glance.
It's the middle sibling of the pack's compare family: identical to EA Image Compare but with three image slots, and there's a 4-Way version if you need a weight sweep with one more step.
What you set
The inputs are the same shape as the 2-way node, just repeated:
image_1/image_2/image_3- the three IMAGE tensors. The default captions telegraph the intended workflow:Base Model,LoRA 0.5,LoRA 1.0. In practice you'll change those to whatever you're actually comparing - LoRA weights, CFG values, samplers, seeds.scale(0.1–2.0) resizes all three proportionally. For three panels, dropping below 1.0 is usually the right call unless you're on a big monitor - a full-res trio gets wide fast.font_size,caption_height,font_color,background_color,spacingstyle the caption bars and gutters.font(optional) pulls a TTF fromcomfyui-kjnodes/fontsif you have it, else falls back to PIL's default.
Everything else is the same as the 2-way node, including the rule that bites people: all three images must have identical dimensions or the node raises a ValueError. Normalize sizes upstream before the compare node.
What comes out
A single comparison IMAGE - captioned panels side by side in input order, spaced a few pixels apart. Wire it to PreviewImage to eyeball, or to a save node to keep a permanent record. Because the output is one image, a three-panel test result drops cleanly into a gallery or a comparison post, which is more than you can say for three scattered previews.
Install
It ships in comfyui-ea-nodes: ComfyUI Manager → search comfyui-ea-nodes, or git clone https://github.com/ExoticArts/comfyui-ea-nodes into custom_nodes/, restart, hard-refresh. The pack declares no Python dependencies - requirements.txt is empty - so there's nothing extra to install.
If you only need two panels, use the 2-way version; if you need four, there's a 4-way. This one exists for the classic three-step sweep, and it does exactly that and nothing else - which is the right amount of node for the job.
Inputs (13)
| Name | Type | Default | Description |
|---|---|---|---|
| image_1 | IMAGE | — | |
| caption_1 | STRING | Base Model | — |
| image_2 | IMAGE | — | |
| caption_2 | STRING | LoRA 0.5 | — |
| image_3 | IMAGE | — | |
| caption_3 | STRING | LoRA 1.0 | — |
| scale | FLOAT | 1.000.1–2 | — |
| font_size | INT | 3212–200 | — |
| caption_height | INT | 6020–200 | — |
| font_color | STRING | white | — |
| background_color | STRING | black | — |
| spacing | INT | 40–50 | — |
| fontopt | COMBO | default | 1 options: default |
Outputs (1)
| Name | Type | Description |
|---|---|---|
| comparison | IMAGE | — |