---
title: FLUX.2-klein-base-4B-bf16
canonical_url: "https://www.modelscope.cn/models/mlx-community/FLUX.2-klein-base-4B-bf16"
md_url: "https://www.modelscope.cn/models/mlx-community/FLUX.2-klein-base-4B-bf16.md"
repository: mlx-community/FLUX.2-klein-base-4B-bf16
last_updated: 2026-07-07
license: apache-2.0
pipeline_tag: text-to-image-synthesis
tasks:
  - text-to-image-synthesis
base_model:
  - black-forest-labs/FLUX.2-klein-base-4B
base_model_relation: finetune
parameters: 8.0B
tensor_type:
  - BF16
  - I64
library_name:
  - mlx
  - lora
  - safetensors
  - diffusers
  - pytorch
frameworks:
  - pytorch
downloads: 38
stars: 0
tags:
  - mlx
  - safetensors
  - apple-silicon
  - text-to-image
  - diffusion
  - mmdit
  - flux2
  - klein
---

# FLUX.2-klein-base-4B-bf16

> FLUX.2-klein-base-4B-bf16 - mlx-community 在 ModelScope 开源的模型。mlx-community/FLUX.2-klein-base-4B-bf16

mlx-community/FLUX.2-klein-base-4B-bf16 是 ModelScope 魔搭社区上的 8.0B 参数text-to-image-synthesis模型，采用 apache-2.0 许可，基于 black-forest-labs/FLUX.2-klein-base-4B 构建。

- **Repository**: mlx-community/FLUX.2-klein-base-4B-bf16
- **License**: apache-2.0
- **Tasks**: text-to-image-synthesis
- **Parameters**: 8.0B
- **Base model**: black-forest-labs/FLUX.2-klein-base-4B
- **Tags**: mlx, safetensors, apple-silicon, text-to-image, diffusion, mmdit, flux2, klein
- **Downloads**: 38
- **Stars**: 0
- **Last updated**: 2026-07-07

Source: https://www.modelscope.cn/models/mlx-community/FLUX.2-klein-base-4B-bf16

---

# mlx-community/FLUX.2-klein-base-4B-bf16

MLX (bf16) snapshot of **[black-forest-labs/FLUX.2-klein-base-4B](https://huggingface.co/black-forest-labs/FLUX.2-klein-base-4B)**
(Apache-2.0) for Apple Silicon — the **quality tier** companion to the distilled
[`FLUX.2-klein-4B-bf16`](https://huggingface.co/mlx-community/FLUX.2-klein-4B-bf16). Same compact
rectified-flow **MMDiT** (5 double + 20 single blocks) + Qwen3-4B 3-layer-tap conditioner + FLUX.2
VAE, but the base checkpoint is **not guidance-distilled**: it runs classic two-pass
classifier-free guidance (guidance ~4.0) + **negative prompts** over ~28 steps for stronger prompt
adherence (dense scenes, text, fine attributes) at ~2× the forward cost. Standard diffusers tree;
int8/int4 produced at load (int4 DiT ~2.35 GB → q4 pipeline fits a 16 GB Mac).

**4B only** — the 9B klein variants are FLUX Non-Commercial; this Apache-2.0 4B is the wrapped tier.

## Use (Swift / MLXEngine)
```swift
import MLXKlein
import MLXToolKit
let pkg = Klein4BBaseT2IPackage(configuration: .base(quant: .int4, snapshotPath: "<this repo dir>"))
try await pkg.load()
let r = try await pkg.run(T2IRequest(
    prompt: "a rustic picnic table with a chalkboard sign reading 'PICNIC', three green apples, "
          + "sunflowers, sunny meadow, photorealistic",
    negativePrompt: "blurry, deformed, extra objects",
    guidanceScale: 4.0, width: 1024, height: 1024, seed: 42)) as! T2IResponse
```
Code: https://github.com/xocialize/flux2-klein-swift