Final scope · images only · 4K · Higgsfield Ultimate · 2026-06-13
The entire 1,681-SKU catalog, elevated at 4K, costs ~$600–1,000 in compute.
At a measured 4 credits / 4K image (preflighted today on both engines), even the widest scope (~19,300 credits with a re-roll buffer) is covered by your two months of Ultimate (6,000 cr across June + July rollover) plus one $750 / 15k credit pack. Credits are not the bottleneck — human QA review is. Plan the project around review hours, not dollars.
*Features ≠ a separate burned-in image. Caraway-style callouts are best done as a theme overlay: AI generates the clean macro/detail visual (touchscreen, whistle spout, enamel rim — like the treatment pilot), and the banner text sits in an editable, bilingual rs- section on top. So "features" = ~160 macro generations for the items that actually have features (appliances + mechanism gadgets), not 949. Cheaper, and the copy stays editable in EN/FR.
Three scenarios — all at 4 cr / 4K, ×1.35 re-roll buffer
Lean
product-all + kitchen lifestyle
Product shots (all 1,681)1,681
Lifestyle (kitchen 639)639
Features macro (~110)110
Buffered credits~13,120
~$615all-in compute
Recommended
★ product-all + kitchen&home lifestyle
Product shots (all 1,681)1,681
Lifestyle (kitchen+home 942)942
Features macro (~160)160
Buffered credits~15,030
~$710–1,010all-in compute
Maximal
literal 3× on all Tier 1
Tier 1 × 3 (949)2,847
Tier 2 × 1 (739)739
—
Buffered credits~19,250
~$920–1,010all-in compute
All three converge near ~$1,000 because one 15k credit pack ($750) covers the gap in every case — the pack is the natural buying unit. À-la-carte top-ups (also $0.05/cr) land the lean/recommended cases at ~$615–710 if you'd rather not overbuy.
Credit supply plan
What actually gates this (the real cost)
QA review labor — the true cost. ~2,400–3,600 base generations each need a human keep/re-roll call. Even at 60–90 sec each that's ~40–90 hours of review, plus the re-roll loop. This is where the project time and the client price live — not the ~$1k of credits. Our pilots ran near-100% first-take, which would shrink this, but full-catalog uncurated SKUs (multipacks, odd shapes, bath textiles) will reject more.
Monthly cap forces packs. 3,000 cr/mo = only 750 4K imgs/mo from the subscription. Two months = 1,500 imgs. Full scope is ~2,800–3,600 → packs are required, exactly as you flagged.
Throughput is fine. 8 concurrent image jobs → roughly 30–50 hours of generation wall-clock, trivially absorbed in overnight batches across the build. Aligns with mid-to-late July delivery if QA review starts in the next week or two.
Base-photo dependency persists. The 33 Cook & Fresca specials + ~31 net-new request SKUs still have no packshot — reference-locked AI can't touch them until Jessar supplies a base photo. They're a hole in "entire catalog" regardless of budget.
Recommendation
Go Recommended scope: a redone color-background product shot for every SKU (kills the grey-on-grey problem you fixed by hand on the JS Gourmet box — and we redo for quality while we're in there), lifestyle for all kitchen + home, and feature macros for the ~160 items that have features — banded in-theme, not burned in. Buy one 15k pack ($750) on top of the two months you're already paying → ~$1,000 all-in compute, ceiling.
For the client quote: compute is ~$600–1,000, so the Phase-1 price is almost entirely workflow + QA labor + margin — exactly the value-anchored stance. This confirms the earlier ~$14–20k full-catalog two-tier figure, comfortably under the $35k migration. Video stays Phase 2, quoted later per-style. Charge for the judgment and the pipeline, not the pixels.