Home Creators Posts Import Register Favorites Logout

Downloads

Content

Henge uses MiniMax H3 to swap characters, transfer motion, swap heads and sharpen videos; a 12 second Character Swap took about 4 minutes on an RTX 5090.

All-in-1 video studio by Alisson Anjos with four tools in one browser page, a planner AI that writes the plan for you and SAM 3.1 tracking for several people. Runs on your own PC or on RunPod, Massed Compute and SimplePod; the installer sets up its own ComfyUI, Python 3.12, PyTorch 2.13 with CUDA 13 and 14 model bundles.

Patreon exclusive posts index to find our scripts easily, Patreon scripts updates history to see which updates arrived to which scripts and amazing Patreon special generative scripts list that you can use in any of your task.

Join discord to get help, chat, discuss and also tell me your discord username to get your special rank : SECourses Discord

Please also Star, Watch and Fork our Stable Diffusion & Generative AI  GitHub repository and join our Reddit subreddit and follow me on LinkedIn (my real profile)

Latest Zip File : Henge_Installer_v4.zip

Windows Requirements

For RunPod, SimplePod and Massed Compute no requirements needed so just follow their Instructions_READ txt files and run commands

11 November 2026 V4

  • The startup files will start with following ComfyUI startup parameters and you can easily edit them now

    • --disable-pinned-memory --use-sage-attention --enable-triton-backend --disable-comfy-compiler

  • Edit Windows_Start_Henge.bat to change as you wish

Henge is an all-in-1 video studio: swap characters, transfer motion, swap heads and sharpen videos on your own PC. Henge, made by Alisson Anjos, runs the MiniMax H3 video model with its own LoRAs and a planner AI that reads your video and pictures and writes the plan for you. Our standalone installer sets up the whole studio with one double-click: its own ComfyUI, Python, Node.js, FFmpeg and every library, plus a fast model downloader. Below you will see every step and every feature with real screenshots and real results from our PC.

Why Henge with our installer

  • Four studios in one app: Character Swap, Motion Transfer, Head Swap and Sharpness, all in one browser page.

  • The AI plans for you: a planner model watches your guide video and your pictures, writes the prompt and shows its plan before anything renders.

  • Real results in minutes: on an RTX 5090, a 12 second Character Swap took 4 minutes 9 seconds and a 12 second Head Swap 3 minutes 3 seconds, planning included.

  • Several people at once: click each person in the video, and SAM 3.1 tracks them so each one becomes their own character.

  • Ready-made examples: open a Community example, add your own picture and generate.

  • 1-click install: Henge, its own ComfyUI, Python 3.12, PyTorch 2.13 with CUDA 13 and precompiled libraries, in one folder that touches nothing else.

  • Fast, verified model downloads: 14 bundles, from the essentials to all 159 GB of public models, fetched in parallel pieces and checked against the source hashes. All 43 files took about 44 minutes on our connection.

  • Ready in seconds: Henge, its bridge and ComfyUI were ready 10 to 16 seconds after the double-click on our PC.

Here is the studio right after a real Character Swap with a new background. Our gladiator picture took over the High Kick Spin example in a neon city in 4 minutes 38 seconds on an RTX 5090.

1: four studios in one page. 2: your guide video. 3: the Henge result. 4: sharpen it or swap the head next. 5: every generation stays in Recent and History.

1-Click Installation

What You Download

You get one small zip file with the installer, the launchers and the model downloader.

Extract it into a permanent folder with a short path, for example C:\Henge, then double-click Windows_Install_Or_Update_Henge.bat. The same file updates Henge later.

Latest PyTorch, CUDA 13 and Precompiled Libraries

The installer makes its own Python 3.12 environment, its own ComfyUI with the Henge nodes, portable Node.js and FFmpeg, all inside the install folder. It installs every package with uv, a very fast package installer.

PyTorch 2.13 with CUDA 13 comes with SageAttention, Flash Attention, TorchAO and xFormers precompiled for Windows, so nothing has to be compiled on your PC.

Every Library Checked

At the end the installer builds the studio and checks every library.

All 125 packages passed with no conflicts. Our clean install took 95 seconds on our PC, with the packages already downloaded once before.

Cloud GPUs: RunPod and Massed Compute

No strong GPU at home? The zip also has 1-click installers for RunPod and Massed Compute. They set up the same studio and download the models for you.

The step-by-step commands are in the README inside the zip. On RunPod, open Connect, then HTTP 8288 to use the studio in your browser.

Requirements

On Windows you need Python 3.12, Git and an NVIDIA driver with CUDA 13. This tutorial shows every step, and the same setup runs all our AI apps.

Model Downloader

Double-click Windows_Start_Model_Downloader.bat and the model library opens in your browser. Pick bundles by task, or every public model at once.

Recommended studio, the main Henge tools, is 103.47 GB. All public models is 159.12 GB in 43 files. The downloader shows the free space on the drive before you start.

Fast Parallel Downloads

Every file is split into pieces that download in parallel, 8 connections by default.

On our connection the files came in at about 70 to 100 MB per second. Pause keeps every finished piece, so the next start continues from there.

Every File Verified

When the download ends, every file is checked against its source hash.

All 43 files of All public models arrived and passed the check in about 44 minutes. Individual models lets you add or repair a single model at any time.

First Start: Set up Henge

Double-click Windows_Start_Henge.bat. Henge, its bridge and ComfyUI were ready 10.8 seconds after the double-click on our PC. The first time, Henge opens a short setup wizard.

The installer already set up Henge's own ComfyUI, so the connection test finds all 7 base models right away. The next pages choose the assistant and an optional login.

Community Examples

The fastest start is a ready-made example: a guide video, its pictures and its prompt, loaded together.

Where to find it: click the Community tab at the top of the studio.

Pick any example card:

Open an example in Character Swap or Motion Transfer, then put your own picture where it says "You".

Choose the Part of the Video

When you add a guide video, Henge opens an editor to pick the part you want.

Drag the handles or type the start and end times. Only the selected part is planned and rendered, and the scissors button opens the editor again.

The AI Writes the Plan

Click Generate and the planner model studies your video and pictures first. You see what it understood before anything renders.

Where to find it: the Generate button under the scene box; Auto-approve plan is in the Settings column.

Leave Auto-approve off and Henge stops at the plan for you to check:

Check the role of each picture, read its warnings, look at every shot, then approve. Tick Auto-approve plan to render right after planning.

Live Preview While It Renders

While the video renders, Henge shows a live preview of the sampling.

You see the step, the speed and the time left, so you know early how the result will look.

Motion Transfer

Motion Transfer moves anyone or anything with the performance of the guide video: people, animals, objects.

Where to find it: pick Motion transfer in the Mode list, then add your guide video, or open a Community example.

Here is a real run with the Dancing Husky example:

Here the husky from our picture dances the German shepherd's dance of the guide. The 12 second clip rendered in 4 minutes 55 seconds, the first model load included.

Your Result and Its Downloads

Every result stays in the studio with everything that made it.

Download the video, a side-by-side comparison, a comparison with the references, everything as a zip, or the ComfyUI workflow. Continue with Sharpen or Head swap in one click, and every generation is kept in Recent and History.

Character Swap

Character Swap replaces a person in the video with the character in your picture and keeps the motion, the camera and the scene.

Where to find it: pick Character Swap in the Mode list, add the guide video and your character picture.

Here is a real run with the High Kick Spin example and our knight picture:

Our knight took over the High Kick Spin example. The 12 second clip took 4 minutes 9 seconds on the RTX 5090, planning included.

Head Swap

Head Swap changes only the head and keeps the body, the clothes and the scene. Henge writes the prompt for you.

Where to find it: pick Head Swap in the Mode list, then add the guide video and a face picture.

Here is a real run with the Fridge Dessert example and our sea captain picture:

The old sea captain from our face picture took over the Fridge Dessert example in 3 minutes 3 seconds.

Several People at Once

When the video has more than one person, click each one and Henge tracks them with SAM 3.1.

Where to find it: the Subjects and references panel under the video.

Pause on a frame and click a person:

Each subject gets its own color and its own picture. When the video has a camera cut, mark each person once in every shot.

Two People Swapped

Here both people of the guide were swapped at once, each into their own character.

The astronaut and the gladiator follow both people through both shots. The 12 second clip took 4 minutes 2 seconds.

Sharpness

Sharpness makes any clip crisper, an upload or one of your generations, and keeps its motion and look.

Where to find it: pick Sharpness in the Mode list, then choose one of your results or upload a clip, and pick 1080p or 2K.

Then set the method below the scene box:

Choose 1080p or 2K, and the Open-H3 Refiner + LMS method upscales the result after sharpening.

Here is our knight Character Swap from 480p after one Sharpness run to 1080p, frames from the same moments:

Hair, beard, the bottle labels and the armour edges come out much crisper, and the motion stays the same. The 12 second clip took 17 minutes 23 seconds on the RTX 5090, the first model load included.

First Frame: Edit One Frame, the Whole Video Follows

In Motion Transfer you can edit one frame of the guide, and Henge carries that change through the whole clip.

Where to find it: in Motion transfer, open the FFP card under the Reference pictures.

The card opens the frame editor:

Pick a frame, say what to replace and with what, and FIBO 2.0 edits it while the pose, camera and scene stay. Our red shirt became knight armour in about 8 seconds. Qwen-Image 2.1 and FLUX Klein are there too when you install them.

The Result

Click Use as first frame, then Generate. Henge picks the Motion Transfer FFP model by itself.

The armour stays on through every axe swing and the fall into the river. The 10 second clip took 2 minutes 26 seconds.

New Background

In Character Swap and Motion Transfer you can also put the people into a new place. Add one picture, and Henge places it behind them as a moving plate.

BiRefNet separates the people in every frame, and the picture moves with the original camera, so the shot still feels real.

The Result

Here the gladiator from our picture took over High Kick Spin, and our neon city picture replaced the room:

The gladiator holds through every kick, and the city follows the camera. The 12 second clip took 4 minutes 38 seconds.

The Studio Assistant

Not sure which mode to use? Ask the assistant in plain words. It runs on the same planner model, so it needs no extra account or key.

We asked how to turn the man into a knight. It described the video, picked Character Swap, marked the man as Subject 1 with our picture and asked before generating. Claude or an OpenAI-compatible model such as LM Studio can be picked in Settings too.

Start, Status and Stop

Windows_Start_Henge.bat starts the studio, Windows_Henge_Status.bat checks that its three parts run, and Windows_Stop_Henge.bat stops them. Everything stays inside the install folder: models, jobs and logs.

Related posts

All posts in this topic: AI Video Generation and Animation Apps · New to SECourses? Start Here: Best and Most Popular SECourses Posts

Files

Comments

Hipno

OMG THANK YOU SO MUCH!!

Hipno

Does this include background swap and with person blending in properly with lighting and the motion of the camera?

SECourses

Background swapping and camera following are included. Lighting integration is less complete. - Background swap: Upload a picture through New background; Henge separates the person and replaces the scene behind them. - Camera motion: Tracks the original background and applies pan, rotation and zoom to the replacement. It moves a flat image, so complex 3D perspective changes remain limited. Implementation (https://github.com/alisson-anjos/Henge/blob/main/backend/moving_plate.py) - Person blending: The initial composite preserves the person’s original lighting. H3 generation can be instructed to match lighting, shadows and perspective, but that result depends on the model; the swap step has no dedicated relighting pass. Prompt behavior (Henge/Henge/docs/prompt-format.md:21)

SECourses

yes i can make that workflow but it is not Henge

Hipno

ohh haha true that

SECourses

by the way i am about to add replace background preset so soon :D 1 click

Hipno

OMG i cant wait!!

Hipno

thank you Dr!

Vik An

Очень рад что подписался к вам. Много интересной инфы по ИИ. Огромное спасибо!!!

SECourses

welcome and thanks. i also just saw your youtube comment thanks

Christoph Behrmann

I can/have to use MiniMax H3 video LoRA to keep consistency for my one-eye calico cat?

SECourses

you can do image to video or give as reference. if you train lora it can even improve more