How does this Pipeline work?

Stage 1 — Image preparation (optional T2I generation):
If no input_image is provided, generates one from the prompt using a selected text-to-image model —
black-forest-labs/flux-schnell (default), flux-dev, stability-ai/sdxl, or ideogram-ai/ideogram-v2(-turbo).
Resizes the image so both dimensions are divisible by 32.

Stage 2 — Offset transformation:
Shifts the image by 50% horizontally and vertically (offset_by_half), wrapping around edges so seams are centered.

Stage 3 — Mask creation (model-aware):
Builds a seam mask at the image center with a width based on seam_percentage (min 32px), tapering edges for smooth blending.
Mask color scheme depends on inpaint backend — ideogram-ai/ideogram-v2 uses white-keep/black-paint,
black-forest-labs/flux-fill-pro uses black-keep/white-paint.

Stage 4 — Inpainting seams:
Runs the chosen inpaint model (ideogram-ai/ideogram-v2 or black-forest-labs/flux-fill-pro)
with the prompt, the offset image, and the seam mask to fill and blend the seams.

Stage 5 — Offset restoration:
Reverses the initial 50% shift (offset_back) to return the image to its original orientation,
now with seamless edges.

Stage 6 — Save output:
Saves the final tileable image as tileable_image.png, ready for use as a seamless texture or background.