Black Forest Labs

Listed at https://blackforestlabs.ai

Overall Rank #11 ⭐ Consider
❌ Proxy required (AWS US-East + EU-Frankfurt deployment) | 🌍 International

💰 Token Pricing

TypePriceNote
Input FLUX.2 [pro]: $0.05/MP, FLUX.1.1 [pro]: $0.04/MP, FLUX.1 [schnell]: $0.003/MP, FLUX.1 Fill: $0.05/MP per million tokens
Output Flat per-megapixel billing, no tier-based markup per million tokens
💡 Free Credits:

🤖 Supported Models (8)

FLUX.2 [pro]FLUX.1.1 [pro]FLUX.1 [pro]FLUX.1 [dev]FLUX.1 [schnell]FLUX.1 Fill [pro]FLUX.1 Depth [pro]FLUX.1 Canny [pro]

✨ Pros

  • Original FLUX creators: official source of open-SOTA FLUX.1/FLUX.2 image models
  • FLUX.2 Pro launched May 2026, text rendering + photorealistic portraits at Midjourney level
  • 8 endpoints: text-to-image, image-to-image, inpainting, depth, canny, redux variants
  • FLUX.1 [schnell] is Apache 2.0 open-source (12B params), commercial use OK
  • Pricing $0.003-0.05/MP, 40-60% cheaper than Replicate/Stability
  • Both hosted inference + commercial API: production-safe licensing, no copyright ambiguity
  • EU-Frankfurt hosting node available for GDPR compliance

⚠️ Cons

  • ×Image generation only — no LLM/chat capability
  • ×China access requires proxy (AWS US-East + EU-Frankfurt, no mainland edge nodes)
  • ×Strict rate limits on [pro] endpoint (production workloads need enterprise negotiation)
  • ×No video generation (unlike fal-ai/Replicate which also serve Kling/Wan)
  • ×No Chinese documentation, English prompts give best results
  • ×No formal affiliate program (API link is just UTM source tracking)

🎯 Best For

Official FLUX model inference; copyright-safe commercial use; batch image generation; photorealistic portraits + text rendering; SaaS path alongside [schnell] self-hosting

💰 Pricing & Plans

ModelPriceOutput ResolutionBest For
FLUX.2 [pro]$0.05/MPUp to 2KFlagship photorealism + text rendering (May 2026)
FLUX.1.1 [pro]$0.04/MPUp to 2KFast generation, ~0.3s first token
FLUX.1 [pro]$0.05/MPUp to 2KClassic pro quality for high-res art
FLUX.1 [dev]$0.025/MPUp to 1MP standardHigh-quality open-weights model, commercial-friendly license
FLUX.1 [schnell]$0.003/MPUp to 1MP standardFastest, Apache 2.0 open-source, real-time drafts
FLUX.1 Fill [pro]$0.05/MPVariableInpainting / outpainting

🔧 API & Developer Experience

  • REST + JSON API: Single POST endpoint per model family (text-to-image, image-to-image, inpainting) returning a JSON payload with result URLs and a seed for reproducibility.
  • Per-megapixel billing: Charges by MP of output rather than by step or image — predictable cost scaling for batch pipelines.
  • SDK Availability: Official Python and TypeScript/JS clients plus an OpenAI-style integration; third-party SDKs (e.g. Replicate, fal.ai, ComfyUI custom nodes) cover the FLUX endpoints.
  • Seeds & Determinism: Every call accepts a seed parameter; same seed + same prompt reproduces the same image, which is critical for A/B testing and consistent brand assets.
  • Asynchronous Mode: Long-running jobs (Fill, upscaling) return an ID you poll for the result — no blocking socket ties up your app server.
  • Webhooks: Apply a webhook URL for job completion notifications on async endpoints.
  • Fine-tuning via API: FLUX.1 [dev]/[pro] support LoRA fine-tuning through the BFL API, letting you train a custom style or character on top of the base model.

🎨 FLUX 图像生成专长

Black Forest Labs is the original creator of the FLUX family, the open-SOTA image-generation models that set the reference benchmark for DiT-based text-to-image quality. FLUX.1.1 [pro] already delivered gapless text rendering and fast iteration, and FLUX.2 [pro] (May 2026) pushed photorealism and legible in-image text to Midjourney-class quality while keeping a flat per-megapixel price. The company ships a full endpoint suite — text-to-image, image-to-image, inpainting/outpainting (Fill), depth-and-canny-conditioned generation, and redux variants — so one API covers the entire creative pipeline from ideation to final edit. Because FLUX.1 [schnell] is Apache 2.0 licensed and [dev]/[pro] support LoRA fine-tuning, teams can prototype on the hosted API, then self-host or fine-tune the weights for production. For image-only workloads FLUX is purpose-built where generalist multimodal providers dilute throughput. This is a specialist API — not an LLM/chat endpoint — so pair it with a text model for captioning or prompt-expansion in your stack.

🌐 China Access & Latency

Black Forest Labs hosts its inference in AWS US-East and EU-Frankfurt, with no mainland China edge node. Direct connections from China are unreliable and typically need a stable proxy or an overseas relay, adding significant latency to what is already an image-generation workload that is bandwidth-heavy by nature. For China-based production, teams commonly route through an overseas GPU aggregator (fal.ai, Replicate, or a China-adjacent relay) that caches the FLUX models on regional infrastructure, or self-host the Apache-2.0 [schnell]/[dev] weights on a domestic instance. Expect exports of generated images to be large — MP pricing is cheap, but you should consider CDN or object-storage egress when serving results to mainland users. EU-Frankfurt nodes can serve as a GDPR-friendly option for Europe-focused teams, but are not a China latency solution.