BENCHMARKS
USAGE
HARDWARE GUIDE
MODEL CARD
VARIANT COMPARISON
Accuracy vs. Size Tradeoff
Every variant is benchmarked against the FP16 baseline. Choose the right balance of accuracy and footprint for your workflow.
Benchmarks are performed on representative hardware. Accuracy is relative to model FP16 baseline.
| VARIANT | FORMAT | SIZE | PERPLEXITY | ACCURACY | LATENCY | BEST FOR | DOWNLOAD |
|---|
| FP16 (baseline) | — | 6450.00 MB | -- | 100% | — | Reference only | — |
ACCURACY RETENTION BY VARIANT (%)
FILE SIZE BY VARIANT (MB)
INTEGRATION GUIDES
Usage Code Snippets
llama.cpp
Python / ONNX
8b4b CLI
import torch
model = load_quantized("FLUX-1-Kontext-dev-GGUF")
output = model.generate("Hello world")
🔒
PROTECTED_CONTENT
Registry integration snippets, deployment guides, and API access keys are restricted to members.
SIGN IN TO UNLOCK
COMPATIBILITY GUIDE
Which variant for your hardware?
Choose the right quantization path for your device and performance target.
🧠
Desktop CPU
BESTHigh quality CPU inference for GGUF and ONNX.
📱
Mobile / TFLite
RECOMMENDEDFast on-device inference with TFLite and CoreML.
🖥️
GPU / AWQ
OPTIONALHigher throughput with GPU-backed AWQ and GPTQ.
☁️
Cloud API
VERSATILEServe quantized variants with low latency.
MODEL CARD
About This Model
OVERVIEW
FLUX.1-Kontext-dev is an open-weight 12-billion-parameter Rectified Flow Transformer developed by Black Forest Labs. Unlike traditional text-to-image models, Kontext supports both image generation and image editing within a unified architecture. Users can provide an image together with natural language instructions to perform object replacement, background modification, style transfer, text editing, character preservation, and iterative multi-turn image refinement. The GGUF version is a quantized deployment format optimized for local inference through ComfyUI-GGUF, significantly reducing VRAM requirements while maintaining high visual quality. Kontext achieves strong consistency across multiple edits and is designed for creative workflows requiring context-aware image transformations. It is one of the leading open-weight image editing models available for local deployment.
INTENDED USE
llmNone