LoRA Cast

Put the people you trained identity LoRAs for into a picture you already have — one face at a time, so nobody's face bleeds onto someone else.

Install it Try the interface Source on GitHub

The problem it solves

A face LoRA applies to the whole image. Load two people's LoRAs into one render and their faces blend, or everyone ends up wearing the same face. Masking and inpainting help, but the LoRA is still loaded for the whole picture.

LoRA Cast never renders two identities together:

You open a picture. Made with any tool, from anywhere.
LoRA Cast proposes a box for each face. You say who each one is.
Each face is redrawn on its own. Cropped with room around it, rendered image-to-image with exactly one LoRA, then blended back with colour matching, a feathered edge and matched grain.

Because every face is its own render with one LoRA, identities cannot bleed. Two LoRAs that share a trigger word can even appear in the same picture.

Install it

One line on macOS or Linux. It fetches LoRA Cast, sets it up and opens it in your browser:

curl -fsSL https://raw.githubusercontent.com/Stigmavlc/lora-cast/main/install.sh | bash

Prefer to read before you run? The script is here, and the manual steps are in the README.

What you need

Python 3.10+

Already on most machines. The installer checks and tells you if not.

ComfyUI or Draw Things

ComfyUI on Windows, Linux and macOS. Draw Things on macOS, with brew install draw-things-cli.

Your own models

Tuned on Krea 2. Other model families work through your own exported workflow file.

Identity LoRAs

One per person you want to place. LoRA Cast never trains anything — it uses what you have.

After setup it opens at http://127.0.0.1:8420. From then on it's a double-click: Launch LoRA Cast.command on macOS, launch-lora-cast.bat on Windows.

Using it

Open a picture and LoRA Cast proposes a box for each face. The boxes are yours to correct: drag on the picture to add one, move or resize it, rotate it for a tilted head, or delete it. Click a box to say who it is, then press Apply.

Setting	What it does
Face strength	How much of the original face is thrown away before redrawing, 0.30–0.95. Around 0.40 the expression stays natural; by 0.70 the likeness is strongest but faces start to pull into a grimace.
LoRA weight	Per person, 0.1–2.5. Some LoRAs need well above 1.0 to show up properly.
Expression	Per face, optional. The whole head is redrawn, so without it a face comes back with the expression its LoRA was trained on.
Seed	The same picture, settings and seed give the same result. Change it for a different take.

Applying again always starts from the original picture, never from a previous result.

The Draw Things trap

Draw Things only applies a LoRA that was imported in the Draw Things app, where it becomes a .ckpt. A raw .safetensors file renders happily and applies nothing at all, with no warning at any point.

LoRA Cast refuses those out loud. When you add a person it runs a check: the same small picture is rendered twice with the same seed, with and without the LoRA, and the two are compared. Identical renders mean the LoRA did nothing, and that person can't be assigned to a face until it passes.

Your pictures stay yours

There is no account, no upload and no telemetry. LoRA Cast is a small local server on 127.0.0.1 that talks to the rendering engine already on your machine. Pictures, LoRAs and results never leave your computer, which is also why there is no "render it in the browser" button here — the GPU doing the work is yours.

Responsible use

Only use LoRAs of people who have agreed to it, and never of minors. Do not use LoRA Cast to impersonate anyone, to deceive, or to make intimate images of someone without their consent. Follow the licences of the models and LoRAs you use.