Bonus: Top up now and we'll double your first deposit — get x2 credits instantly.
Get Started
Sign Up
Sign Up

FLUX.2 Klein 4B: Self-Hosting & Deployment Guide

FLUX.2 Klein 4B is a compact, 4‑billion‑parameter image generation and editing model from Black Forest Labs, optimized for fast, sub‑second inference on consumer GPUs.

Browse available models

What is FLUX.2 Klein 4B?

FLUX.2 Klein 4B is a 4B-parameter rectified-flow transformer model by Black Forest Labs for high-quality, low-latency image generation and editing on consumer hardware. It is mainly used for text-to-image creation in interactive applications where sub-second response and good visual fidelity are important. It is also widely used for single- and multi-reference image editing workflows, including LoRA-based personalization and fine-tuning-friendly setups. The model is part of the FLUX.2 [klein] family, a fast, compact branch of the broader FLUX.2 image-generation and editing models.

How to run FLUX.2 Klein 4B yourself

Open weights (apache-2.0), 3.9B parameters. Below: how much memory the weights need at each quantization, which hardware fits them, and the commands to serve it.

FLUX.2 Klein 4B VRAM requirements

Precision / quantizationWeightsFigureSource
BF16 (full precision)7.8 GBEstimate3.9B params × 2 bytes
FP84.1 GBFile sizeblack-forest-labs/FLUX.2-klein-4b-fp8
GGUF Q8_04.3 GBFile sizeunsloth/FLUX.2-klein-4B-GGUF
GGUF Q4_K_M2.6 GBFile sizeunsloth/FLUX.2-klein-4B-GGUF
GGUF Q2_K1.8 GBFile sizeunsloth/FLUX.2-klein-4B-GGUF

File size = actual download size on Hugging Face. Estimate = parameter count × bytes per weight. Weights only — the KV cache for long contexts needs extra memory; the fit tables below add 15% headroom as a rule of thumb.

GPU requirements: deploy FLUX.2 Klein 4B on your hardware

HardwareBF16 (full precision)FP8GGUF Q8_0GGUF Q4_K_MGGUF Q2_K
NVIDIA H200141 GB1×1×1×1×1×
NVIDIA H10080 GB1×1×1×1×1×
NVIDIA A100 80GB80 GB1×1×1×1×1×
NVIDIA L40S48 GB1×1×1×1×1×

GPUs needed per variant.

HardwareGGUF Q8_0GGUF Q4_K_MGGUF Q2_K
RTX 509032 GB1×1×1×
RTX 409024 GB1×1×1×
RTX 309024 GB1×1×1×

GGUF via llama.cpp; more cards or CPU offload for larger files.

HardwareGGUF Q8_0GGUF Q4_K_MGGUF Q2_K
Mac, 512 GBM3 UltraFitsFitsFits
Mac, 192 GBM2 UltraFitsFitsFits
Mac, 128 GBM4 Max / M3 MaxFitsFitsFits
Mac, 64 GBM4 Pro / MaxFitsFitsFits

Assumes ~75% of unified memory is usable by the GPU.

HardwareBF16 (full precision)FP8GGUF Q8_0GGUF Q4_K_MGGUF Q2_K
AWS p5.48xlarge8× H100 · 640 GBFitsFitsFitsFitsFits
AWS p4de.24xlarge8× A100 80GB · 640 GBFitsFitsFitsFitsFits
AWS g6e.12xlarge4× L40S · 192 GBFitsFitsFitsFitsFits
Google Cloud a3-highgpu-8g8× H100 · 640 GBFitsFitsFitsFitsFits
Azure ND H100 v58× H100 · 640 GBFitsFitsFitsFitsFits

Whole-instance GPU memory; public instance specs.

Not on LLM.API yet

We do not route this model through our API at the moment. Browse the models you can call today — most workloads have a close match already live.

Browse available models

Providers

Different companies host the same model. OpenRouter routes your request to one of them based on the routing mode you pick — Balanced (price + speed), Nitro (fastest), or Exacto (highest tool-calling accuracy).

Provider Input Output Cache read /M Latency Throughput Uptime
Black Forest Labs (Direct) ~$0.0012/img $0.0000/img — ~550ms ~60 img/min ~99.9%
Replicate ~$0.0015/img $0.0000/img — ~700ms ~40 img/min ~99.5%
Together AI ~$0.0013/img $0.0000/img — ~600ms ~70 img/min ~99.9%

No hosted provider publishes a price for this exact checkpoint on OpenRouter; self-hosting cost depends on your GPUs. Checked weekly. Last updated 11 Oct 2026.

Try this model

Test FLUX.2 Klein 4B right here — free to start.

FLUX.2 Klein 4B
Hi! Want to test the model?

Suggestions for your first prompt

Not available on LLM API yet

We do not route this model through our API at the moment, so there is no endpoint or code snippet for it yet. Browse the models you can call today — most workloads have a close match already live.

Browse available models

FLUX.2 Klein 4B: Self-Hosting & Deployment Guide in production

What the model is good at.

  • Text-to-image

    Generates high-quality images from natural language prompts using a compact 4B-parameter rectified flow transformer architecture.

  • Image Editing

    Edits existing images based on text instructions, enabling transformations, enhancements, and content modifications in a unified pipeline.

  • Multi-reference Editing

    Combines multiple reference images with text prompts to guide style, composition, or subject while preserving visual consistency.

  • Real-time Inference

    Optimized for sub-second image generation and editing on consumer GPUs, supporting interactive and high-volume visual workflows.

  • Fine-tuning Support

    Supports fine-tuning and LoRA-based customization through its base variants, enabling domain-specific or style-specialized image models.

What you get when you call FLUX.2 Klein 4B: Self-Hosting & Deployment Guide through LLM.API.

  • Unified AI Routing

    Automatically route each request to the optimal model across providers based on latency, cost, or quality — without changing your integration.

    One endpoint, every model.

  • Cost-Aware Orchestration

    Control spend with smart tiering, quotas, and policy-based model selection so you always use the cheapest model that still meets requirements.

    Optimize every token.

  • Automatic Provider Fallback

    Stay online when a model or provider fails with built-in health checks and seamless failover, no extra logic in your app.

    Resilient by default.

  • End-to-End Observability

    Trace every request across models and providers with logs, metrics, and event streams that plug into your existing monitoring stack.

    See every token flow.

  • Task-Level Abstractions

    Call high-level tasks like chat, tools, or rerank instead of model-specific APIs, so you can swap models without refactoring.

    Code to tasks, not models.

  • High-Throughput Batch

    Process millions of inferences efficiently with batch endpoints that maximize provider throughput while handling retries and rate limits for you.

    Scale inference, not ops.

When to Use — When NOT to Use

Use it if...

  • You need a lightweight, 4B-parameter vision model for cost-efficient image generation.
  • You need reasonably high-quality images but must stay within tight GPU memory limits.
  • Your use case involves rapid iteration on image concepts rather than photoreal perfection.
  • Your use case involves deployment on modest on-premise hardware or edge GPU devices.
  • You need a compact model for fine-tuning or domain adaptation on limited data.
  • Your use case involves batch image generation where throughput matters more than peak fidelity.

Avoid if...

  • You need a large, general-purpose language model for text understanding or generation tasks.
  • Your workload requires state-of-the-art photorealism rivaling the largest diffusion or video models.
  • You need robust performance on highly diverse, long-horizon multimodal reasoning or planning tasks.
  • Your workload requires extremely detailed, high-resolution images for print-grade commercial media.
  • You need fine-grained control using complex textual instructions, compositional prompts, or scene logic.
  • Your workload requires integrated text, code, or tool-calling alongside image modeling in one system.

COMMUNITY

What developers say about FLUX.2 Klein 4B

LMArena—Speed—Sources: LMArena, Artificial Analysis

Summarised from public developer discussions and write-ups. Opinions are the sources’, not LLM.API’s.

  • Developers on Hugging Face note that the 4B model is a rectified flow transformer designed for sub-second image generation and multi-reference editing on consumer hardware.
  • Technical reviews report that while the 4B model is ultra-lightweight and fast, it sacrifices semantic understanding and fine detail compared to the larger 9B variant.
  • Developers on GitHub highlight that the 4B version is the only variant in the family offering a permissive Apache 2.0 license, making it ideal for commercial production.

SOURCES

Frequently Asked Questions

  • What is FLUX.2 Klein 4B?

    FLUX.2 Klein 4B is a 4B-parameter image generation model from Black Forest Labs optimized for fast, efficient, high-quality image synthesis.

  • What modalities does FLUX.2 Klein 4B support via LLM.API?

    FLUX.2 Klein 4B supports text-to-image generation and image-to-image transformation through the LLM.API image generation endpoints.

  • What is FLUX.2 Klein 4B best suited for?

    FLUX.2 Klein 4B is best for rapid, low-cost image generation where lightweight deployment, iteration speed, and decent visual quality are priorities.

  • How is FLUX.2 Klein 4B priced on LLM.API?

    On LLM.API, FLUX.2 Klein 4B is billed per generated image or image step, with exact pricing defined in the LLM.API model catalog.

  • How do I access FLUX.2 Klein 4B through the LLM.API?

    Call the LLM.API image generation endpoint with the FLUX.2 Klein 4B model identifier and your API key in the Authorization header.

  • What is the typical latency of FLUX.2 Klein 4B on LLM.API?

    Typical text-to-image requests return in a few seconds, depending on resolution, step count, and current LLM.API load.

  • Does FLUX.2 Klein 4B have a context window like text models?

    FLUX.2 Klein 4B does not use a token-based context window; it consumes prompts as text strings and conditioning inputs for image generation.

  • How does FLUX.2 Klein 4B compare to larger FLUX.2 models?

    FLUX.2 Klein 4B generally trades some visual fidelity and detail for significantly lower compute cost, faster responses, and easier deployment.

  • Are there safety or content limitations when using FLUX.2 Klein 4B on LLM.API?

    Yes, FLUX.2 Klein 4B usage is subject to LLM.API safety filters and content policies, which may block disallowed or unsafe generations.

  • What are key limitations of FLUX.2 Klein 4B?

    FLUX.2 Klein 4B may struggle with very fine text rendering, complex multi-object scenes, and ultra-photorealism compared to larger image models.

COMPARE

Competitive Models

Get one key to every model

Swap your API key. Keep your code.

Browse available models Talk to sales