FreeLLMNVIDIA Blackwell GPUs cut FLUX 3 Image Klein latency up to 3.46× on single B200
TensorRT LLM VisualGen optimizations deliver up to 3.46× faster text‑to‑image and 4.45× faster image‑to‑image generation on a Blackwell B200 GPU.
NVIDIA’s TensorRT LLM VisualGen runtime was used to accelerate FLUX 3 Image Klein, a diffusion model from Black Forest Labs, on Blackwell GPUs.
On a single B200 GPU, NVFP4 precision already gave 1.62× faster text‑to‑image (T2I) and 1.58× faster image‑to‑image (I2I) compared with BF16.
Further optimizations—fused GEMM kernels, FP8 attention, cache‑DiT step caching and sparse Sol‑Attention—boosted performance to 3.46× faster T2I and 4.45× faster I2I at 2048×2048 resolution.
Multi‑GPU scaling with eight B200 cards reduced median T2I latency from 14.90 s to 3.00 s and I2I from 2.73 s to 1.08 s, achieving 6.01× and 4.44× denoising speedups respectively.
The optimized pipeline keeps most of the generation time in denoising, which dropped from 96.8% to around half when caching was applied.
What it means for you
Developers can now integrate FLUX 3 Image Klein into design tools with markedly lower latency, enabling real‑time image editing workflows. The gains also show that high‑resolution diffusion models can run efficiently on a single Blackwell GPU, reducing reliance on large GPU clusters for many users.
Sources
- Speed Up Image Generation with FLUX 3 Image Klein on NVIDIA Blackwell GPUs, NVIDIA Developer (official), 6 Nov 2026
Written automatically from the sources above, then checked by our code: every number and name must appear in them, and no sentence may copy a source. Spotted a mistake? Report it.
Which AI can you use for free today?
Every free AI model, ranked by blind human votes, with the memory and speed it needs on your own GPU.