flux-2-pro

`flux-2-pro` is Black Forest Labs' flagship FLUX.2 text-to-image model — the production tier of the FLUX.2 family for high-fidelity generation and editing workflows (docs.bfl.ai). On CallMissed, call `"model": "flux-2-pro"` on `/v1/images/generations` with a text prompt; the API shape matches OpenAI image generation (prompt, n, size) so existing image clients port easily.

BFL documents FLUX.2 Pro with megapixel-based pricing from about $0.03 per megapixel on their direct API, async poll workflows, and support for multi-reference editing with up to eight reference images on the BFL provider route (docs.bfl.ai/release-notes). CallMissed bills $0.10 per image on our catalog — check current pricing before high-volume batch runs.

FLUX.2 Pro targets marketing visuals, product photography, concept art, and UI mockups where prompt adherence and fine detail matter. It improves text rendering in images versus many earlier diffusion models, though small text still benefits from iterative prompting. Default resolution is commonly 1024×1024; follow prompt engineering guides from BFL for photographic vs illustrative styles.

When to choose flux-2-pro vs `flux-2-dev`: Pro is the premium tier with stronger fidelity; Dev is cheaper experimentation. vs `flux-1.1-pro`: FLUX.2 Pro is the newer generation with improved coherence; 1.1 Pro remains faster/cheaper for bulk production. vs DALL·E/OpenAI image models: FLUX excels at certain photographic and typographic styles — benchmark on your creative briefs.

Operational notes: generation may be asynchronous internally even if CallMissed returns a URL in one HTTP response; handle HTTP 429 with backoff. Respect content policies from BFL. Store returned URLs or base64 promptly — ephemeral URLs may expire depending on provider settings.

Limitations: per-image cost above budget tiers, not optimized for real-time game textures at 60 FPS, and results vary with safety filters. For blog hero images in our admin, `flux-2-pro` is the configured fallback when Gemini image models fail.

Prompt engineering for FLUX.2: specify lens (`85mm portrait`), lighting (`soft window light`), and material (`matte ceramic`) separately. FLUX.2 improves typography but spell-check generated text in post.

Multi-reference editing (BFL direct): advanced editing with up to eight reference images may use BFL-native endpoints — CallMissed `/v1/images/generations` focuses on text-to-image; confirm feature parity in API docs before building edit pipelines.

Brand safety: run outputs through NSFW/classifiers for user-generated prompt products; BFL safety filters may blank some prompts — UX should explain rejections without exposing filter internals.

Resolution strategy: start 1024²; upscale externally if print needed. Higher megapixel directly maps to BFL credit cost on direct API.

A/B vs 1.1 Pro: run blind human eval on 50 prompts representing your catalog — pick winner per SKU category (packshot vs lifestyle).

Storage: persist returned URLs to your bucket immediately; do not hotlink provider URLs in production emails.

Seasonal campaigns: batch-generate variants with different seeds; track seed in metadata for reproducibility.

Black Forest Labs ecosystem: FLUX.2 family includes dev/pro/flex tiers on BFL docs — `flux-2-pro` is the quality anchor. Study BFL prompt guide for photographic vs illustration modes; negative prompts may differ from Stable Diffusion conventions.

Advanced parameters (steps, guidance) may appear in provider-specific docs beyond the minimal OpenAI-compatible wrapper.

Creative agency workflow: mood board → prompt matrix → generate 20 variants → art director picks 3 → manual retouch in Photoshop — AI covers first mile, humans last mile.

E-commerce trust: disclose AI imagery where regulations require (EU AI Act transparency discussions) — consult legal counsel.

CDN delivery: optimize image formats (WebP/AVIF) after download; FLUX PNGs are large.

Failure UX: when generation fails, show actionable message ("try shorter prompt") not raw HTTP errors.

CallMissed blog pipeline: internal blog admin uses `flux-2-pro` as image fallback when Gemini image models fail — same model id you use in production marketing APIs.

Catalog merchandising: tag generated assets with model id `flux-2-pro` in DAM metadata — future audits trace which model produced licensed-looking imagery for insurance claims. Run quarterly prompt library reviews so brand colors and banned concepts stay synchronized with marketing compliance. Creative directors should maintain a living Notion board of winning `flux-2-pro` prompts with thumbnail exports — institutional memory beats rediscovering styles each campaign.