---
title: Qwen-Image-Flash-8bit
canonical_url: "https://www.modelscope.cn/models/mlx-community/Qwen-Image-Flash-8bit"
md_url: "https://www.modelscope.cn/models/mlx-community/Qwen-Image-Flash-8bit.md"
repository: mlx-community/Qwen-Image-Flash-8bit
last_updated: 2026-07-25
license: other
pipeline_tag: text-to-image-synthesis
tasks:
  - text-to-image-synthesis
base_model:
  - nvidia/Qwen-Image-Flash
base_model_relation: finetune
parameters: 7.8B
tensor_type:
  - BF16
  - U32
library_name:
  - mlx
  - lora
  - safetensors
  - diffusers
  - pytorch
frameworks:
  - pytorch
downloads: 44
stars: 0
tags:
  - mlx
  - mlx-swift
  - text-to-image
  - qwen-image
  - dmd2
  - few-step
  - nvidia
  - quantized
---

# Qwen-Image-Flash-8bit

> Qwen-Image-Flash-8bit - mlx-community 在 ModelScope 开源的模型。Qwen-Image-Flash-8bit (MLX)

mlx-community/Qwen-Image-Flash-8bit 是 ModelScope 魔搭社区上的 7.8B 参数text-to-image-synthesis模型，采用 other 许可，基于 nvidia/Qwen-Image-Flash 构建。

- **Repository**: mlx-community/Qwen-Image-Flash-8bit
- **License**: other
- **Tasks**: text-to-image-synthesis
- **Parameters**: 7.8B
- **Base model**: nvidia/Qwen-Image-Flash
- **Tags**: mlx, mlx-swift, text-to-image, qwen-image, dmd2, few-step, nvidia, quantized
- **Downloads**: 44
- **Stars**: 0
- **Last updated**: 2026-07-25

Source: https://www.modelscope.cn/models/mlx-community/Qwen-Image-Flash-8bit

---

# Qwen-Image-Flash-8bit (MLX)

An **int8** MLX snapshot of [nvidia/Qwen-Image-Flash](https://huggingface.co/nvidia/Qwen-Image-Flash) —
NVIDIA's DMD2 four-step distillation of [Qwen/Qwen-Image](https://huggingface.co/Qwen/Qwen-Image).
28 GB total, versus 57.7 GB for
[the bf16 snapshot](https://huggingface.co/mlx-community/Qwen-Image-Flash-bf16).

This is the tier that makes a 20B-parameter image model reachable on ordinary Apple Silicon: a
**30.0 GB peak** at 1024²/4 steps instead of bf16's 57.4 GB, and it is **4× faster**.

## Contents

| file | precision | size |
|---|---|---|
| `transformer/model-int8.safetensors` | int8 attention + feed-forward + modulation, group 64 | 21.75 GB |
| `text_encoder/model-int8.safetensors` | int8 Qwen2.5-VL-7B **language model only** | 7.51 GB |
| `vae/` | unquantized | 0.25 GB |

The `img_in` / `txt_in` / `time_text_embed` / `norm_out` / `proj_out` projections are left at full
precision, and the VAE is never quantized — decode is where precision loss shows up as visible
colour and banding artifacts.

The text encoder carries **no vision tower**: text-to-image conditions on text alone, so the ViT is
dead weight in this pipeline and is omitted entirely.

## Measured quality

Against PyTorch fp32 goldens (diffusers 0.37.1), on identical injected inputs:

| | int8 | bf16 | fp32 oracle |
|---|---|---|---|
| DiT step-0 cosine | **0.9973** | 0.99836 | 1.0 |
| VL-7B prompt-embed cosine | **0.99992** | 0.9999926 | 1.0 |
| 1024²/4-step render | 19.8 s | 83.3 s | — |
| load | 2.3 s | ~60 s | — |
| peak memory | **30.0 GB** | 57.4 GB | — |

Renders at this tier are visually indistinguishable from bf16 at the same seed.

### Why there is no 4-bit tier

int4 was built and measured, not skipped: DiT step-0 cosine **0.9623** at group 64 and **0.9659** at
group 32, with the VL encoder at 0.9845. The 1024² render came out visibly soft and washed out, with
fine fur and snow detail gone. Finer scale groups did not rescue it — this DiT is intrinsically lossy
at 4 bits — so no 4-bit snapshot is published rather than shipping one that looks like that.

## Inference notes that are easy to get wrong

The distillation **internalized CFG 4.0**, and the packaged scheduler is **static shift-3**
(`use_dynamic_shifting: false`):

- `num_inference_steps = 4`, `true_cfg_scale = 1.0` — applying CFG again double-counts guidance the
  student already absorbed, and doubles the transformer evaluations per step for nothing.
- The four-step trajectory is sigmas `[1.0, 0.9, 0.75, 0.5, 0.0]`.
- Tested at 1024 × 1024. Use width/height divisible by 16.

## Use from Swift (MLXEngine)

```swift
import MLXQwenImageFlash
import MLXToolKit

let package = QwenImageFlashPackage(configuration: .init(quant: .int8))
try await package.load()
let response = try await package.run(T2IRequest(
    prompt: "A red fox in a snowy pine forest at golden hour, photorealistic",
    width: 1024, height: 1024, seed: 42)) as! T2IResponse
```

Port: [xocialize/qwen-image-edit-swift](https://github.com/xocialize/qwen-image-edit-swift) (MIT).
The package also selects this tier automatically when the engine's memory governor reports a budget
that cannot seat bf16 — and it resolves that *before* downloading, so a constrained machine fetches
this 28 GB snapshot rather than 41 GB of bf16 it could never load.

These are **pre-quantized** weights. Consumers never materialize bf16 at any point, which is the
difference between a tier that runs on a 32–48 GB machine and one that merely claims to: quantizing
at load would require holding the 41 GB bf16 transformer first.

## License

Governing terms: **[NVIDIA Open Model License Agreement](https://www.nvidia.com/en-us/agreements/enterprise-software/nvidia-open-model-license)**.
Additional information: Apache License 2.0 (`LICENSE`).

> Licensed by NVIDIA Corporation under the NVIDIA Open Model License

Commercial use, derivative models, and redistribution are permitted; if you redistribute these
weights you must pass on the Agreement and this notice (§3.1). See `NOTICE`.