Step1X-Edit

Overview

Step1X-Edit-FP8 is a new free open-source tool trying to compete with big names like GPT-4o and Gemini2 Flash.
It lets you edit pictures by just typing what you want like "make the sky sunset pink" or "erase the person in the image" and boom it does it.

It was developed by StepFun, a tech shop known for fast vision-language tools.

How It Works

It mixes a multi-skill language model like Qwen-VL with a DiT-style diffusion engine.
You drop a picture give a command and it edits smart without you drawing masks or fiddling with stuff.
They even made a huge special dataset over 1 million image-text pairs and built a new benchmark called GEdit-Bench to measure real-life edits.

Here’s a quick list of what it can do:

• Add or remove stuff. Like add a tree or remove a pole

• Change colors. Like make a gray car red

• Change materials. Like make a statue look like glass

• Swap backgrounds. Move your subject to a beach or a city street

• Boost portraits. Clean up wrinkles brighten smiles

• Edit text. Change a billboard from SALE to NEW

• Style stuff. Make a picture look like watercolor

What You Need To Run It

Here’s a quick table that lists what you need to run the Step1X-Edit model (batch size 1 no cfg distillation) for editing pictures

They tested it on a single H800 GPUand for best quality and faster generation speed they suggest using GPUs with 80GB of memory.