Files

T

ausboss 5ec02ef356 Divisible by: no strip of fill on an edge without padding

Image Crop + Rotate + Pad and both Video Crop + Rotate + Pad nodes rounded
the canvas up to Divisible by and put the pixels left over on the right and
bottom, whatever was padded. A 1916 x 821 picture with 3 px of padding on
the left and Divisible by 24 came to 1920 wide, with 1 px of fill on the
right edge, which had no padding. With Resize on, that strip was smeared
into the picture's last column, and the stitched result kept it as a light
line.

The canvas now rounds one axis at a time by Load Image + Pad's rule
(_pad_helpers._axis_layout): the leftover joins a side that was padded, the
right or bottom one when both are, and an axis nobody padded has its crop
trimmed evenly to the multiple below. A picture smaller than one multiple
still grows on the far side. With no padding on an axis the size therefore
rounds down, not up: 1000 x 750 at 64 gives 960 x 704, was 1024 x 768.
Inputs and outputs are unchanged.

The editor uses the same rule (axisLayout, resolvePadding and keptCrop in
js/shared/transform_geometry.mjs), so the picture on the node, the size
line and the editor match the run. The size line names fill added on any
side and pixels trimmed, and the centre buttons count fill that sits on the
left or top. The Divisible by tooltip and help pages describe the rule, and
the Krea 2 Outpaint Model Patch warning no longer says Divisible by rounds
the other axis up.

tests/test_transform_engine.py and tests/transform_geometry.test.mjs share
one fixture table, so the editor and the run are held to the same numbers.

2026-10-11 20:41:52 -05:00

5.5 KiB

Raw Permalink Blame History

Krea 2 Outpaint Model Patch

Tells Krea 2 where a reference image sits on the canvas. Use it when you extend a picture with Krea 2 (outpainting), so the model carries your picture on instead of painting something that only looks similar.

Place it after any LoRA loader and before the sampler.

How it works: a reference normally reaches the model with no position, so the model treats it as a loose style hint. It borrows the look and reinvents the content. This patch pins the reference to the rectangle the stitcher reports.

Controls

The node's card shows Reference (source rect / whole canvas) and KV cache (every step / once per run). Both accept links and retain their saved values when the workflow is reloaded.

model: A Krea 2 model. Patch last, so a LoRA loaded afterwards does not replace the patched forward pass.
stitcher: From Load Image + Pad 🆎 or a Crop + Rotate + Pad 🆎 node. It tells the patch where your picture sits on the canvas. With a stitcher from Crop For Inpaint 🆎, set Reference to whole canvas: on source rect the reference is pinned to where the crop sat in the full picture, not spread over the crop.
KV cache (kv_cache, once per run by default): Compute the reference's keys and values once per run instead of once per step. The reference does not change while sampling, so this is free speed. Set it to every step only to rule the cache out when debugging.
Reference (placement): Where the reference tokens go. source rect (source rectangle, the default) pins them to the rectangle the stitcher reports. whole canvas spreads them over the full frame. Which one is right depends on the LoRA — see below.

Output

model: The model with reference tokens registered into the canvas grid.

Two LoRAs, two placements

The base Krea 2 weights were never trained on registered references; what makes this patch work is a LoRA trained for it, and the two that exist want different placements.

Every side at once — whole canvas + AnyPaint. yijunwang2's Krea 2 AnyPaint (krea2_anypaint_rank32.safetensors) was trained with the padded canvas itself as the reference, spread over the whole frame, and the known pixels held in place by the sampler. In its training the new area of that reference was always a flat patch of the picture's own median colour. Wire it like the Krea 2 Outpaint example:

A pad node with any padding you like: Image Crop + Rotate + Pad 🆎, as in the example, or Load Image + Pad 🆎. Left, right, top and bottom together are fine. Flat colour fill.
Krea 2 Encode 🆎 with the pad node's image output (the padded canvas, not Load Image + Pad's unpadded reference output) as the reference, vlm_reference on, and the pad node's mask wired into its mask input. The mask makes the encode refill the new area with the picture's median colour, as in training. Without it a dark or very colourful picture can come back with the gray padding painted in as a gray frame.
This patch on whole canvas.
VAE Encode the padded canvas, Set Latent Noise Mask with the pad mask, and sample from that latent at full denoise. The mask is what keeps the source pixels; the reference is what tells the model what they are.

To repaint part of a picture instead of extending it, use the Krea 2 Inpaint Masked example: Crop For Inpaint 🆎 cuts out the area you painted, and this patch runs on whole canvas with the same LoRA. Load Image + Pad 🆎 itself does not take a painted mask.

One axis per pass — source rectangle + Registered Outpaint. The same author's Registered Outpaint (krea2_outpaint_rank32.safetensors) is the LoRA this placement was built for: the unpadded source goes in as the reference (Load Image + Pad 🆎's reference output, VLM reference off) and is pinned to its rectangle. It was trained on a source spanning one whole canvas axis — pad left/right and the source must span the full height, pad top/bottom and it must span the full width. Padding both axes in one pass is outside its training and the new region can break up; the node says so in the console when it sees it, and the cure is two passes: pad one axis, run, load the result, pad the other. A thin strip counts. More than 8 px of padding on the other axis, such as a few pixels left over from a drag, means the source no longer spans it. Divisible by on the Crop + Rotate + Pad nodes and Load Image + Pad's Multiple never add a strip on a side you did not pad. The warning reports the spare pixels on each axis.

Without either LoRA the patch still places the reference, but the bare Turbo model treats it loosely and results are hit and miss.

Notes and limitations

Nothing happens without reference latents. If the conditioning carries none, the patched forward pass calls straight through to the original. Wire Krea 2 Encode 🆎 with a VAE and a reference image.
It patches comfy internals. The placement reaches into the flux attention layers, so a ComfyUI release that moves them can break this node specifically. It imports those internals when you run it rather than at startup, so a break surfaces as an error on this node instead of the node disappearing from the menu.
Reference tokens cost attention. The reference is shrunk so its longest side is 384 px before encoding for that reason — see Krea 2 Encode 🆎.

5.5 KiB Raw Permalink Blame History