After 8 Oct: FLUX 3 Image without the 50% promo—which layout-box and multi-ref routes to keep in Bubble bulk catalogs via BIAIG

Context

Black Forest Labs shipped FLUX 3 Image on 1 October 2026. One endpoint — POST https://api.bfl.ai/v1/flux-3-image, header x-key — covers text-to-image, instruction edits, layout bounding boxes, and multi-reference composition (1–10 images). The digest records 50% off API through 8 October (last day); after that window, list tiers from BFL release notes apply (768² $0.041/image, 1k $0.048, 2k $0.100, 4k $0.607; API schema also lists 1.5k at $0.070 list in third-party summaries citing BFL).

Secondary sources put the promo cutoff at 8 Oct 2026, 08:00 Pacific (15:00 UTC)—not local midnight. If you ran layout-box and multi-ref experiments under the discount, 9 October is when you pick what stays on production bulk routes at full price.

This is a follow-up to the pre-deadline promo tutorial: post-promo routing and catalog reruns, not how to turn the discount on.

BIAIG in this publication

There is no public Bubble or BFL definition of BIAIG. I use it here as an editorial label for the internal bulk image routing layer in Bubble: which catalog SKU maps to which provider, endpoint, resolution, and QA policy—invoked from backend workflows via the API Connector (not a frontend loop over hundreds of rows). Call it whatever you want in your stack; the decision logic still applies. A closer look: 29 Oct 2026 Nano Banana. A closer look: 29 Oct 2026 Nano Banana.

Keep on FLUX 3 direct (list price)

No official doc tells you exactly what to keep. From BFL capability docs and tier economics, I’d use your promo-era QA with this heuristic:

Keep the BIAIG route on api.bfl.ai when the feature is load-bearing, not “just FLUX”:

1. Layout boxes embedded in prompt — a JSON array of bbox rows (no separate box parameter; 0–1000 grid; Keep/Move/New/Remove). Marketplace grids, fixed slots, pixel-targeted edits where cheaper FLUX plugins cannot match control.
2. Multi-ref (2–10) with ordered images[] and prompt references (“image 1”, “image 2”), including cross-ref placement docs (src_bbox / tgt_bbox). Many fallback resellers lack full box JSON + 10-ref parity and BFL billing rules.
3. Hybrid instruction + boxes where you relied on result.prompt (model-expanded boxes)—reruns without that snapshot can drift visually.

Demote to fallback (Replicate, Fal, legacy Bubble FLUX plugins) when the promo route was plain t2i:

• No images[], no box JSON in prompt, and no measured quality gap in your catalog.
• High volume where doubling to list post–8 Oct does not pay back on accepted outputs.

Before demotion, verify reseller parity (Fal, OpenRouter, etc.): promo end dates and list rows may differ from direct BFL; do not assume identical schemas.

Treat 4k bulk reruns as a separate tier

List 4k (~$0.607) vs 1k (~$0.048) is roughly ~12.6× per image on BFL’s table. I would not rerun an entire thumbnail catalog at 4k by default. Public summaries also note no guaranteed exact replay when re-running at higher resolution—treat it as a new generation pass, not a deterministic upscale.

Snapshot in Bubble before list-price reruns

Even identical-looking requests can diverge visually. Persist metadata before a full catalog rerun:

Request at submit time:

• Full prompt, including appended box JSON
• images[] order and URLs (or your storage ids)
• aspect_ratio, resolution, grounding (schema default true—may affect reproducibility), safety_tolerance, endpoint version

Poll / result:

• Job id, returned polling_url (BFL integration guide: poll that URL)
• cost / output_mp when present
• result.prompt (expanded)
• Final stored asset after fetching result.sample—editing docs: download within 1 hour

Bubble pattern I’ve seen in forum bulk-catalog threads: backend workflow → API Connector → recursive scheduled polling, job state on a Data type. Miss a poll and you can lose paid outputs. A closer look: Mixboard offline on 28.09: exporting. A closer look: Mixboard offline on 28.09: exporting.

BFL billing: rejected/failed/withheld are not charged; accepted outputs are charged even if you dislike them. Post-promo decisions should use $/accepted asset, not attempts alone.

Checklist for 9 October

1. Export promo-era jobs with pass/fail and resolution tier.
2. Tag SKUs: box-heavy, multi-ref, plain t2i.
3. For box/multi-ref with acceptable promo QA—lock BIAIG → BFL direct @ 1k/2k (not 4k unless you have a separate business case).
4. Plain t2i—demote unless you measured a quality gap worth list price.
5. Ensure workflow snapshot fields above exist.
6. Pilot N rows at list price; compare to archived assets and saved result.prompt.
7. Update BIAIG option sets; document promo cutoff (15:00 UTC on 8 Oct) for the team.

Checklist item	Do this	Not for / unless
Export promo-era jobs	Export with pass/fail and resolution tier	—
Tag SKUs	Label box-heavy, multi-ref, plain t2i	—
box-heavy / multi-ref (acceptable promo QA)	Lock BIAIG → BFL direct @ 1k/2k	4k unless you have a separate business case
plain t2i	Demote	Unless you measured a quality gap worth list price
Workflow snapshot fields	Ensure the snapshot fields above exist	—
Pilot & BIAIG updates	Pilot N rows at list price; compare to archived assets and saved result.prompt; update BIAIG option sets; document promo cutoff (15:00 UTC on 8 Oct) for the team	—

Limits

• No public FLUX 3 Image leaderboard at launch (per Kingy.ai)—demotion quality calls need your QA, not external benchmarks.
• Very small boxes and boundary bleed are documented failure modes; layout-heavy SKUs may need human QA at list price.
• Announced open-weight FLUX 3 (weeks after launch) is a separate economic path from API list pricing—I’m not mixing it in here without license facts.

Sources (Scout packet)

BFL release notes, FLUX 3 API reference, layout/bbox/multi-ref guides, integration guidelines, bfl.ai pricing; Bubble API Connector and forum backend bulk patterns; digest: “FLUX 3 Image — launch BFL; −50% API through 8 October (last day)”. A closer look: FLUX 3 Image 50% off. A closer look: FLUX 3 Image 50% off.