Add scene-harmonisation final pass: match the photo's grain + colour cast over the swapped head so the GAN-smooth face/hair stop looking synthetic/pasted and read as photographed with the scene

Hair swap: run Laplacian blend BEFORE hair (it was overwriting transferred hair with target's); opaque core + soft edge + recolour residual old hair for a seamless blend, both genders

Skin tone: single uniform LAB offset over whole skin (no gate/per-region) -> no patches; verified clean across 10 tones light-to-dark, strength 0.92

Skin tone: recolour face-skin + body to SOURCE complexion (sampled from clean parsed skin, lips/eyes excluded, relative luminance gate) — clean on normal faces

Move skin-tone match to final pipeline step so the Laplacian blend can't undo it (output skin now matches source across whole body); strength 0.9

Fix close-up selfie detection: pad margin so face-fills-frame photos are found (0 faces -> crude paste was the no-hair/skin bug)

Fix hair composite: widen parse crop + region clamp so long hair transfers into scene (re-enable hair by default)

Hair swap back to opt-in: in-scene hair composite loses the hair (verified)

Enable source hair+neck transfer by default (HairFastGAN on GPU, BiSeNet fallback)

Tighten skin-tone match 0.75 -> 0.85 (closer to source; glasses transfer already on)

Default to the clean face swap; make head-swap & hair-swap opt-in

Auto de-roll tilted source faces (eye-line alignment) before swap

Wire HairFastGAN (GPU) into the swap pipeline for realistic hair transfer

GPU: 1536px working res on CUDA + GPU deploy files (Dockerfile.gpu, requirements-gpu.txt)

Swap only the main subject + always run face swap (fixes the mess)

Revert swap-quality regressions: skin 0.92->0.75, glasses face-swap-path only

Custom per-location result messages + unique defaults; remove em-dashes & "4K"

Role permissions + tablet-width 2-col layout

Redesign Generate page (centered single-column + header), stronger skin-tone match

Fix core swap: auto-download InsightFace buffalo_l + glasses in both paths

Always sync locations: no-store on fetch + backend, lighter cards

Set the 3 real Control Panel owner emails (Aditya, Nishith, C D Parmar)

Control Panel: Google sign-in restricted to a 3-email allowlist

Store target images in Supabase (table + Storage) with local FS fallback

New flow: user picks gender + location instead of uploading target

Fix NoneType error: exif_transpose can return None in some Pillow versions

Fix dependency conflict, Laplacian blend arg order, pyrightconfig

Improve face-neck-hair pipeline per 7-step checklist

Fix all listed technical challenges across detection, landmarks, segmentation

fix: apply EXIF orientation on upload (explicit, consistent across builds)

feat: head swap — keep the SOURCE's face shape (not just identity)

refactor: don't store swap outputs on the server — download is client-side

feat: swapped face takes the SOURCE skin tone + previews show the whole face

fix: download works on the HuggingFace site (inline data-URI, not file round-trip)

fix: natural 4K upscale + hair/glasses on by default (toggles hidden)

fix: fair naturalness (face-region) + make hair/glasses toggles visible

feat: make source-glasses transfer an opt-in toggle (default off)

fix: stop double-GFPGAN making faces look plastic/unnatural

fix: hair paste-back + transfer source glasses onto the swap

feat: enhance uploads before the pipeline + real alignment metric

fix: match swapped face+neck to TARGET tone, not source (no washed-out face)

feat: high-quality hair transfer via hosted HairFastGAN Space (opt-in)

fix: match face AND neck to source tone (kill the pasted-jaw seam)

feat: experimental hair/head swap (opt-in)

feat: match swapped face complexion to the source

fix: GFPGAN face restoration removes swap blur; simplify pipeline; GPU 4K

fix: sharp output, source skin tone preservation, 4K upscaling pipeline

fix: remove match_skin_tone (was pushing toward target tone), source skin tone preserved by InsightFace

feat: restore hair-to-neck blend pipeline on top of InsightFace swap

perf: strip heavy post-processing pipeline, trust InsightFace output directly