<h1 id="flux-3-image-klein-or-forge-neo-extension">FLUX 3 Image Klein | Forge Neo Extension</h1>Generate with <a target="_blank" rel="ugc" href="https://huggingface.co/black-forest-labs/flux-3-image-klein">FLUX 3 Image Klein</a> inside <a target="_blank" rel="ugc" href="https://github.com/Haoming02/sd-webui-forge-classic/tree/neo">Forge Neo</a>, using the checkpoint, preset and Generate button you already know: text-to-image, editing with up to 10 reference images, and layouts drawn with boxes in the prompt, even on 8 GB GPUs.Requires an up-to-date Forge Neo (the <code>neo</code> branch). On an older version it stays disabled and says so in the console, and Forge Neo keeps working as usual.<hr /><h1 id="features">Features 🚀</h1><h3 id="native-generation">🖼️ Native Generation</h3><ul><li>Runs inside txt2img and img2img: no separate tab, no extra environment</li><li>The distilled model makes an image in 16 steps, about 25 seconds at 1024x1024 on an RTX 2070 with the INT8 file</li><li>The base model runs 50 steps with CFG, made for LoRA training and research</li><li>INT8 files for 8 GB GPUs, the full BF16 files, and NVFP4 for RTX 50 cards</li><li>Uses the same text encoder as Krea 2 and Iris 3B, and the same autoencoder as FLUX.2</li></ul><h3 id="reference-images">🧩 Reference Images</h3><ul><li>Edit a picture or combine up to 10 of them, with ImageStitch Integrated and img2img</li><li>An edit with one reference takes about a minute on an RTX 2070</li></ul><h3 id="layout-by-boxes">📐 Layout by Boxes</h3><ul><li>Add a list of boxes at the end of the prompt, in the format of Black Forest Labs' guide: <code>[top, left, bottom, right]</code> on a 0-1000 grid, with a short description and a name such as <code>&lt;vase_1&gt;</code></li><li>Works with sd-dynamic-prompts installed</li></ul><h3 id="draw-the-boxes">🖱️ Draw the Boxes</h3><ul><li>The companion Regional Boxes extension lets you drag on a frame to draw each box, and writes the list into the prompt</li><li>It also places, adds, moves and removes things for box edits</li></ul><h3 id="edit-by-boxes">✂️ Edit by Boxes</h3><ul><li>With a reference image, add, remove, move or replace things by giving their boxes. The rest of the picture stays as it was</li></ul><h3 id="lora">🔌 LoRA</h3><ul><li>FLUX 3 LoRAs with the usual <code>&lt;lora:name:weight&gt;</code> syntax, with the BF16 and INT8 files alike</li></ul><h3 id="safe-by-design">🛡️ Safe by Design</h3><ul><li>Changes no Forge Neo file: remove the extension and everything is as it was</li><li>Checks your Forge Neo at startup and tells you which file is missing instead of failing with a generic error</li><li>Other models are not affected</li></ul><hr /><h1 id="recommended-settings">Recommended Settings ⚙️</h1><ul><li>Distilled model (<code>_sd_</code> files): Euler, schedule FLUX.3, 16 steps, CFG 1. The flux3 UI preset sets these</li><li>Base model: Euler, schedule FLUX.3, 50 steps, CFG 3.5</li><li>Leave the negative prompt empty and use sides that are a multiple of 16. For edits with a reference, use Denoising strength 1</li></ul><h1 id="installation">Installation 📦</h1><ol><li>Open Forge Neo WebUI and go to Extensions → Install from URL</li><li>Paste: <code>https://github.com/eduardoabreu81/flux3-image-forge-neo</code></li><li>Click Install and restart the WebUI</li><li>Accept the license on <a target="_blank" rel="ugc" href="https://huggingface.co/black-forest-labs/flux-3-image-klein-comfy">black-forest-labs/flux-3-image-klein-comfy</a> and place the checkpoint, the Qwen3-VL 4B text encoder and <code>flux2-vae</code> in the usual folders (the full file list is in the README)</li><li>Pick the flux3 UI preset, the checkpoint, and the text encoder and autoencoder under VAE / Text Encoder</li></ol><hr />Documentation: <a target="_blank" rel="ugc" href="https://github.com/eduardoabreu81/flux3-image-forge-neo/wiki">Wiki</a> · <a target="_blank" rel="ugc" href="https://github.com/eduardoabreu81/flux3-image-forge-neo#-installation">File list</a> · <a target="_blank" rel="ugc" href="https://github.com/eduardoabreu81/flux3-image-forge-neo#-whats-new">What's new</a> · <a target="_blank" rel="ugc" href="https://github.com/eduardoabreu81/flux3-image-forge-neo">GitHub</a>License: the extension is AGPL-3.0. The FLUX 3 Image Klein weights are released under the <a target="_blank" rel="ugc" href="https://huggingface.co/black-forest-labs/flux-3-image-klein-comfy/blob/main/LICENSE.md">FLUX Kommunity License</a> and are not included.

FLUX 3 Image Klein | Forge Neo Extension

v0.1.1: First release on Civitai<ul><li>FLUX 3 Image Klein in Forge Neo: the 16-step distilled model and the base model, in BF16, INT8 and NVFP4 files, with the flux3 UI preset and the FLUX.3 schedule.</li><li>Reference images and layout by boxes: edit a picture or combine up to 10 of them, and place, add, move or remove things by giving their boxes.</li><li>Older GPUs: works on cards without bf16 (RTX 20xx), tested on an 8 GB RTX 2070.</li><li>Plays well with other extensions: the Forge check no longer breaks when another extension hooks the loader first.</li></ul>Full changelog: <a target="_blank" rel="ugc" href="https://github.com/eduardoabreu81/flux3-image-forge-neo#-whats-new">README</a> · Source: <a target="_blank" rel="ugc" href="https://github.com/eduardoabreu81/flux3-image-forge-neo">GitHub</a>

Files with SHA256: 149e2d490dae1f42fa0e8da19cd62e221dd50dcc0cd8f7fc8de52fcdb4cc41b8