Image Translation

Sometimes the fix for a composition isn't regenerating - it's nudging the whole image a hundred pixels left. Image Translation is the node that does exactly that and nothing else: shift the image by a pixel offset, fill the exposed edge with your chosen background, and hand the same-size result to whatever's next.

It's the simplest node in WtlNodes' image-transforms trio (zoom, rotation, translation), and honestly it's one of those "why did I never have this" utilities once you've used it. The gap it fills is real: core ComfyUI has no translate node, and every img2img user eventually needs to reposition content without resizing.

The inputs (all three of them that matter)

• translate_x - horizontal shift in pixels, ±4096. Negative moves the image left (content shifts left, background appears on the right edge).
• translate_y - same for vertical.
• bg_color - black or white, the fill for the strip left behind. Choose to match whatever's downstream: black for dark scenes, white if you're going to outpaint or composite.

Plus apply_type (the pack's live-preview dropdown) and an IMAGE output at the input's exact dimensions.

Where it earns its keep

• Panning for video / animation. Feed a sequence of frames with a small translate_x stepped per frame and you've got a linear pan. Combined with Image Zoom you can fake a dolly that stays consistent frame to frame.
• Repositioning before a sampler pass. Shift a subject toward center, then img2img at low denoise to fill the shifted edge with matching content. The model does the inpaint-like healing; you did the composition.
• Outpainting prep. Shift the image over and send it to a ControlNet Tile or outpaint workflow so the model extends the scene into the background strip instead of trying to invent around the whole border.
• Testing edge behavior. Want to know if your subject survives cropping or where your watermark tolerates a margin? Slide it and look.

Install and gotchas

It ships in WtlNodes:

cd ComfyUI/custom_nodes
git clone https://github.com/Scorpiosis0/ComfyUI-WtlNodes.git

Restart ComfyUI, or install via ComfyUI Manager by searching "WtlNodes". No models, no heavy deps.

The two things that trip people up:

• The background fill is flat, not generative. After a big shift you get a hard-edged slab of black or white. That's the point - healing the gap is a separate img2img/outpaint step. Expect a raw seam, don't be surprised by it.
• Pixel offsets, not percentages. Shifting half the frame width means translate_x = width/2. There's no "center the subject" button; that's a math node away if you want it automated.