---
title: FLUX.3 klein Bbox Canvas
emoji: 🎛️
colorFrom: indigo
colorTo: gray
sdk: gradio
sdk_version: 5.49.1
app_file: app.py
pinned: false
python_version: 3.12.12
short_description: generare and edit images with FLUX.3 klein bbox canvas
---

# FLUX.3 image-small — layout-control canvas (private)

A custom-HTML canvas on top of the **diffusers-native FLUX.3 modular pipeline**
([`diffusers-new-model-addition-f3` PR #29](https://github.com/huggingface/diffusers-new-model-addition-f3/pull/29)), running the **step-distilled** model AOT-compiled with aokit,
on ZeroGPU.

Drag a bounding box for every element that matters, give each box an id and a
description, write a one-line scene prompt, and FLUX.3 renders the scene with
every element placed inside its box. Toggle the **Bounding boxes** overlay on the
result, or **Replay** to regenerate.

> **Private.** Weights are read at runtime from the private repo
> `diffusers-internal-dev/f3-small-image-distilled` (never copied here). Keep this Space private.

Layout control uses FLUX.3's region format: the final prompt is the scene caption
followed by a JSON list of regions, each `{"id", "bbox": [y0, x0, y1, x1] (0-1000), "desc"}`.

## Setup

Needs an `HF_TOKEN` secret with read access to `diffusers-internal-dev/f3-small-image-distilled` and the private AOTI artifact `diffusers-modular/f3-small-image-distilled-aoti-sm120`.