amd/FLUX.2-klein-4B-amdnpu

🚀 FLUX.2 [klein] 4B on AMD AI PC NPU

FLUX.2 [klein] 4B is a compact rectified flow transformer for fast text-to-image generation and image editing, with state-of-the-art quality at sub-second speeds on consumer hardware. For more information about how FLUX.2 [klein] works, please see the Black Forest Labs blog post.

More details about this model can be found on the original Hugging Face model card: black-forest-labs/FLUX.2-klein-4B.

This model repo contains the optimized ONNX models required to run the image generation pipeline for FLUX.2 [klein] 4B on AMD NPUs.

Model Details

The folder structure is organized to mirror the main components of the diffusion pipeline (scheduler, text encoder, tokenizer, transformer, and VAE decoder).

├─ scheduler/
├─ text_encoder/
├─ tokenizer/
├─ transformer/
└─ vae_decoder/

The scheduler folder contains the scheduler configuration (timesteps, betas, alphas, etc.) used during the diffusion sampling process.

The text_encoder folder contains the text encoder model used to convert the input prompt into conditioning embeddings for the diffusion model.

The tokenizer contains the tokenizer configuration and vocabulary files required to preprocess the text prompt before it is fed to the text encoder.

The transformer folder contains the rectified flow transformer model used in the diffusion process. The transformer is exported and structured specifically to leverage the AMD NPU accelerator for the denoising steps.

The vae_decoder folder contains the VAE decoder model used to map latent representations back to the image space. The VAE decoder is also structured to make use of the NPU accelerator for efficient image reconstruction.

Note: Transformer and VAE decoder models are optimized and structured to run on AMD NPUs. The other components (text encoder, tokenizer and scheduler) are shared between GPU and NPU pipelines, but are provided here for completeness.

Model Details	Description
Person or organization developing model	Benjamin Consolvo (AMD), makn87amd (AMD)
Original model authors	Black Forest Labs
Model date	June 2026
Model version	1.8.0
Model type	Rectified flow text-to-image and image-editing model
Information about training algorithms, parameters, fairness constraints or other applied approaches, and features	Distilled 4B-parameter FLUX.2 [klein] model for real-time text-to-image and multi-reference image editing. See the original model card for architecture and safety details.
License	Apache 2.0 — see black-forest-labs/FLUX.2-klein-4B
Where to send questions or comments about the model	Community Tab and AMD Developer Community Discord

⚡ Intended Use

Getting Started

To get started with this model, visit github.com/amd/sd-sandbox.

⚓ Ethical Considerations

AMD is committed to conducting our business in a fair, ethical and honest manner and in compliance with all applicable laws, rules and regulations. You can find out more at the AMD Ethics and Compliance page.

⚠️ Caveats and Recommendations

Please visit the original model card for more details on limitations, out-of-scope use, and responsible AI development: black-forest-labs/FLUX.2-klein-4B.

📌 Citation Details

Please cite the original FLUX.2 [klein] model and refer to Black Forest Labs for official citation guidance.