FLUX 3 Image 50% off until 8 Oct: layout boxes and multi-ref for Bubble bulk catalogs via BIAIG

Context

Black Forest Labs shipped FLUX 3 Image on 1 October 2026—one hosted endpoint (POST https://api.bfl.ai/v1/flux-3-image, async with polling). I run Bubble bulk catalogs that generate images in batches through BIAIG. That is not a reason to rewrite the whole router. It is a short promo window: publicly 50% off API use until 8 October 2026, for workflows that need multi-reference composition and layout control. Same topic: gpt-image-2 batch discount: unit economics. Worth reading next: After 8 Oct: FLUX 3.

There is no public spec for BIAIG routing rules, Flare/Sunburst/WaveSpeedAI capacity plans, or plugin thresholds. What follows is vendor-verifiable fact plus labelled editorial inference for catalog use cases.

What FLUX 3 adds for catalogs

Catalog need	FLUX 3 Image (sources)	Typical permanent routes
Up to 10 reference images (256×256–16 MP each)	Yes, via images	Gemini 2.5 Flash Image — max 3 input images on the model card; GPT Image 2.5 uses token pricing with limited overlap on bbox / multi-ref
Repeatable SKU placement	Bounding boxes appended to the prompt: [top, left, bottom, right] on a 0–1000 grid	No equivalent contract in public GPT 2.5 docs for apples-to-apples consistency tests
Resolution	768sq, 1k, 2k, 4k (1.5k on some integrators e.g. OpenRouter)	GPT 2.5 — quality/resolution via tokens; flat per-image comparisons need care
Bulk latency	OpenRouter advertises ~43s P50 — budget queues in Bubble	Depends on your BIAIG slot

Editorial inference: boxes plus many refs matter when packshot + room/scene must stay aligned across hundreds of SKUs. A 5% shift breaks the whole catalog. Plain text-to-image without placement control is a weak case for FLUX 3 after promo ends.

Pricing: promo vs list (per output image)

List (BFL pricing overview): $0.041 (768sq), $0.048 (1k), $0.100 (2k), $0.607 (4k).

~50% launch promo (BFL launch thread, OpenRouter, fal — not BFL legal pricing): e.g. 1k ≈ $0.024, 2k ≈ $0.05, 4k ≈ $0.3035 vs list $0.607. Exact promo end instant on first-party pricing pages was not confirmed in the Scout snapshot—I plan with buffer through 8 Oct.

Resolution (per output image)	Launch promo (~50%)	List price
1k	≈ $0.024	Not stated in section
2k	≈ $0.05	Not stated in section
4k	≈ $0.3035	$0.607
Promo sources	BFL launch thread, OpenRouter, fal	Not BFL legal pricing
Promo end	Plan with buffer through 8 Oct	Exact end instant not confirmed (Scout snapshot)

Post-promo anchors (not BIAIG-specific):

• GPT Image 2.5 Flare/Sunburst share token rates ($5/M text in, $8/M image in, $30/M image out); calculators cite ~$0.013/image at 1024×1024 medium (output tokens only)—often cheaper for simple frames, without the same bbox/10-ref contract.
• Gemini 2.5 Flash Image: ~$0.039 per ~1K-class output; Google docs list retirement 2026-10-02—treat Gemini fallback as time-sensitive until you migrate to a successor.
• 4k FLUX 3 at list price is ~6–12× 1k/2k—justify only for final masters, not bulk drafts at thousands of SKUs.

When a temporary FLUX 3 BIAIG slot makes sense (until ~8 Oct)

1. My workflow already passes multiple reference URLs and layout metadata—or I add a Bubble layer that builds bbox strings per BFL docs.
2. I am piloting consistency at 1k–2k under promo—2k promo ≈ $0.05 vs list $0.10 is a reasonable validation budget on real SKUs.
3. Multi-ref beyond three or explicit boxes—functional gap vs Gemini; I keep GPT 2.5 on text/edit routes, not as a drop-in for this contract.
4. The pipeline handles async polling and the one-hour result download window (BFL docs)—bulk without retries/idempotency will be fragile.

When to keep permanent routes (GPT 2.5 / Gemini successor / fallback)

1. After 8 Oct, homogeneous frames without refs/boxes: I prefer cheaper flat or token routes (e.g. FLUX.2 Klein ~$0.014/image in BFL docs, GPT 2.5 at lower quality tiers—if features match).
2. Capacity and latency SLA—if Flare/Sunburst or WaveSpeedAI already meet catalog QPS, FLUX 3’s ~43s P50 can erase promo savings in queue time.
3. Bulk 4k—even at 50% off, 4k stays premium; I draft at 1k/2k and upscale or finalize elsewhere.
4. Grounding—FLUX 3 defaults grounding: true; I do not turn on production bulk without measuring failures/latency (no public BIAIG bulk benchmarks). A closer look: Flare, Sunburst, and WaveSpeedAI: routing. A closer look: Flare, Sunburst, and WaveSpeedAI: routing.

Bubble methodology checklist

1. I tag SKUs with needs_layout / ref_count in the database—your BIAIG router sends only eligible rows to the FLUX 3 promo slot.
2. Promo budget cap on 1k–2k daily volume; 4k only for whitelisted SKUs.
3. Before 8 Oct: I run a sample set (e.g. 20–50 SKUs)—compare drift vs current GPT/Gemini routes; post-promo decisions should follow features, not price alone.
4. After 8 Oct: I keep FLUX 3 only where ref_count > 3 or needs_layout=true; route the rest through permanent BIAIG paths and a documented Gemini successor.
5. Do not mix tier tables—1.5k appears on OpenRouter/fal but may be absent from the BFL pricing snapshot; parity across direct BFL and resellers is not guaranteed.

Item	What it is for	What it is not for / constraint
SKU tags (needs_layout, ref_count)	BIAIG router sends only eligible rows to the FLUX 3 promo slot	Ineligible rows to that promo slot
Promo budget cap	1k–2k daily volume	4k except for whitelisted SKUs
Before 8 Oct	Sample set (e.g. 20–50 SKUs); compare drift vs current GPT/Gemini routes	Post-promo decisions driven by price alone (follow features)
After 8 Oct	FLUX 3 where ref_count > 3 or needs_layout=true	Other SKUs (route via permanent BIAIG paths and a documented Gemini successor)
Tier tables	—	Do not mix; 1.5k on OpenRouter/fal may be absent from BFL pricing snapshot; parity across direct BFL and resellers is not guaranteed

Risks (brief)

• Model is days old (from 1 Oct 2026)—consistency claims are mostly vendor or anecdotal.
• Third-party integrator schemas may differ from native api.bfl.ai.
• Promo wording often comes from mirrors/resellers—contract planning should use BFL list rates after the deadline.

Sources (Scout-verified, 2026-10-05 UTC)

BFL FLUX 3 Image docs; BFL Pricing Overview; BFL product page; launch thread (mirror); OpenRouter model page; fal Flux 3 Edit; Runware multi-ref guide; OpenAI GPT-Image-2.5 Flare/Sunburst; Google Gemini 2.5 Flash Image model card.

---

Internal methodology piece. Not for external publication. BIAIG routing thresholds are not public—adapt this checklist to your configuration.