English fallback: This guide has not been translated into Chinese.

FLUX.2 Klein Examples¶

These examples generate images with the FLUX.2 Klein 9B model through either TeleFuser or the official Diffusers pipeline.

Model Source¶

Model	HuggingFace	ModelScope	Purpose
FLUX.2 Klein 9B	black-forest-labs/FLUX.2-klein-9B	black-forest-labs/FLUX.2-klein-9B	Four-step distilled text-to-image model

Feature Support¶

Feature	Support	Notes
Text-to-image	Supported	TeleFuser and Diffusers reference paths
Multi-GPU inference	Partial	The TeleFuser script accepts `--gpu_num`; use only validated degrees
LoRA	Unsupported	No LoRA option is exposed
Quantization	Unsupported	The examples use BF16 weights
CPU offload	Unsupported	No CPU-offload option is exposed
Feature cache	Unsupported	No cache configuration is exposed
Server API	Supported	The TeleFuser example exposes the standard pipeline functions

Requirements¶

GPU: one H100 or A100 80 GB is recommended for the 9B BF16 model
Software: the standard TeleFuser installation; Diffusers is required for the official comparison script
Input assets: none

Install TeleFuser by following the development setup.

Model Directory¶

/path/to/FLUX.2-klein-9B/
|-- transformer/
|-- vae/
|-- text_encoder/
\-- tokenizer/

The TeleFuser script expects this local Diffusers-style directory. The official script also accepts a Hugging Face model ID through --model_id.

Quick Start¶

TELEAI_EXAMPLE_OUTPUT_DIR=work_dirs \
python examples/flux2_klein/flux2_klein_text_to_image_h100.py \
  --model_root /path/to/FLUX.2-klein-9B \
  --prompt "A sunlit mountain lake"

The command writes work_dirs/flux2_klein_text_to_image_h100.png.

Examples¶

TeleFuser Inference¶

`flux2_klein_text_to_image_h100.py`¶

python examples/flux2_klein/flux2_klein_text_to_image_h100.py \
  --model_root /path/to/FLUX.2-klein-9B \
  --prompt "A detailed architectural photograph"

Key options:

Option	Default	Description
`--model_root`	`${TF_MODEL_ZOO_PATH}/FLUX.2-klein-9B`	Local model directory
`--gpu_num`	`1`	GPU count
`--height`, `--width`	`1024`	Output dimensions, divisible by 16
`--seed`	`42`	Random seed

Diffusers Reference¶

`flux2_klein_text_to_image_official.py`¶

python examples/flux2_klein/flux2_klein_text_to_image_official.py \
  --model_id /path/to/FLUX.2-klein-9B \
  --prompt "A detailed architectural photograph"