Stitch Inpaint takes stitchers from five nodes, not one. Load Image + Pad's reference is always built. Seed has no convert-to-input and no "↺" button any more. LoRA templates live in the browser, and the absorb section listed only some of the loaders it reads. Video Crop + Rotate + Pad → Frame has four outputs and no resize. Frame Interpolate names the RAFT file and folder, and the extras install with ComfyUI's Python.
3.2 KiB
Load Image + Pad
Loads an image and builds an outpaint canvas around it in one node. The canvas drawn on the node is the control: drag any edge of the dashed final rect to grow that side's padding — the whole edge is the handle, and corners grab the nearer edge. The second output is a mask covering exactly the padding, ready for an inpainter.
The widget card holds Source and Upload, fill mode, color or backdrop blur, Feather, Multiple, and Budget. Exact padding opens a separate scrub row for each side. Every row can take a link; linked controls dim, and their values and links save with the workflow. The canvas remains available below the card for dragging the borders. Linked sides are controlled upstream; dragging their borders leaves the saved padding value unchanged.
The on-node canvas
- Each padded side shows its "+N px" count on the band; when the band is too thin to read, the label hops inside the image onto a pill.
- The badge in the corner is the truth: the final output size after the
canvas-multiple and megapixel math, exactly what the
width/heightoutputs will say. - Clicks on empty canvas space fall through, so the node still drags.
- The hidden
pad_left/top/right/bottomwidgets hold the real values — the canvas is their remote control, so undo, save/load, and the API format all see plain INT widgets.
Guarantees
- The original pixels land bit-identical at their position (resized only when a megapixel target is set — and then resized before padding, so the mask seam stays one crisp pixel wide).
- The mask is
1.0over every padded pixel and0.0over the source, ramped only where feather says so.
Controls
- image: Choose or upload from ComfyUI's input folder.
- mode / fill_color / backdrop_blur: Four fills —
color,edge,edge pixel,pillarbox blur. - feather: Ramps the mask inward across the image edge on each padded
side (ramp width capped by the image size), so the sampler blends the
seam.
0keeps the seam hard. The padding itself always stays solid. - canvas_multiple: The final canvas rounds up to this multiple; the remainder joins the right and bottom padding.
- target_megapixels:
0= off. Rescales the source so the padded canvas lands on this many megapixels, then re-rounds to the multiple — the way to outpaint a small or huge image at a sampler-friendly size.
Outputs
- image: The padded canvas.
- mask: The outpaint mask (white = padding, feathered per
feather). - width / height: The final canvas size — the badge's numbers as INTs, for wiring into latent nodes.
- stitcher: Hand to Stitch Inpaint 🆎 with the sampled result to keep only the new padding and restore the original pixels bit-identically — whatever the sampler did inside the source area is discarded. It also records where the source sits on the canvas, which Krea 2 Outpaint Model Patch 🆎 reads to place the reference.
- reference: The source alone, no padding, fitted to a small multiple of 16 — the reference image for Krea 2 Encode 🆎 and other reference conditioning. It is built on every run; leave it unconnected when nothing needs it.