Download for Windows | Features | What you need | Build it yourself
Better Comfy makes pictures and videos with the ComfyUI on your PC, without a single node to wire. Pick a model, write what you want, slide from Ultra Fast to Best and press Generate. It finds ComfyUI by itself (Pinokio, the portable build or ComfyUI Desktop), starts it when needed and reads your models, LoRAs and upscalers from it.
Features
Pictures. SDXL, Pony, Illustrious, SD 1.5, Flux, Qwen-Image, Anima, Z-Image and Krea 2. Better Comfy reads what a model is from inside the file and sets the size, steps, sampler, guidance, clip skip and the usual quality words for it. Five presets from Ultra Fast to Best, and a switch to enhance every picture afterwards. Speed models (Lightning, Hyper, DMD2, Turbo) are spotted and run with few steps by themselves. Every fine control is still there under Advanced.
Videos. WAN 2.2 and 2.1, from a picture or from words only. Seamless loops that find the smoothest loop point and blend the seam, ping-pong, crossfade, start to end picture. Smoothing up to 4x the frame rate, steady colours, MP4, WebM or GIF. Extend a video from its last frame and get both parts as one longer video. Drop a whole folder of pictures to animate them all. Six presets from Ultra Fast to Best: Fast HD makes a video small and quick and then sharpens it to Balanced size with an upscale model, in about half the time, and Quality and Best are made larger and sharpened further.
Edit and upscale. Paint over a part of a picture, say what should be there, and only that part is drawn new. Make any picture bigger and sharper with its own model, prompt and seed, or just enlarge it with an upscale model.
Compare. The same picture with one setting changed (seeds, guidance, steps, sampler, scheduler, preset, LoRA strength or model), side by side.
Characters (work in progress). Make a character from a few pictures of it and keep it in its own tab. On the Image page you pick it from a row of previews, say how close it should stay (free to 1-to-1) and it appears in whatever scene you write. It uses the LoRA you trained for it, its pictures through the optional Character reference add-on, or both.
Train (work in progress). Train a LoRA of your own in one tab: drop in pictures, give it a name and a trigger word, pick a model and press Start. It uses the training built into ComfyUI, so nothing extra is needed, and the result lands in your LoRA library and as a character. It works with SDXL, Pony, Illustrious and SD 1.5 models.
Storage. A tab that shows what takes up space, biggest first. Delete, move or compress big pictures and videos.
A LoRA library. Every LoRA in your ComfyUI with what it was made for (Illustrious, Pony or plain SDXL told apart from the training details, and settable by hand), its trigger words (suggested from the file), notes, the page you got it from, a favourite strength and a preview picture. Sort them with tags (character, art style, pose, detailer and your own), sort them by size or date, and blur the names and previews of NSFW ones if you like. With a Civitai link saved, Fetch fills in the trigger words, a small cover picture and what it was made for. WAN LoRAs that come as a high and a low file are paired up and share their trigger words. Click a trigger word to put it into the prompt, or delete a LoRA from the drive right there.
A model library. Your checkpoints and diffusion models side by side, each marked as such, with what they are made for, tags, notes, a cover, the page you got them from and their size. Fetch, favourites, delete and Use on Image work like in the LoRA library. When a model needs a text encoder or VAE you don't have yet (Anima, Flux, Qwen-Image, Z-Image, Krea 2), it says so and downloads the official one with one click.
A market. Browse Civitai (civitai.red or civitai.com) for LoRAs and checkpoints inside the app: search, filter by
what a model is made for, sort, keep favourites. Downloads go straight into the right ComfyUI folder, are checked
against Civitai's checksum and arrive with their link, trigger words, cover and what they are made for. Every model
shows its downloads and likes, and every file whether it is safe: .safetensors (only numbers, can't run code) and
Civitai's virus and hidden code scans. SFW only until you switch NSFW on. Some creators only allow downloads when logged
in: for those, add your Civitai API key in Settings.
A queue. Pictures and videos one after the other, with time estimates learned on your PC. Take single pictures out of a job, stop one while the rest goes on, and choose what happens when the queue is done: nothing, free the graphics memory, stop ComfyUI, close the app, sleep or shut down.
A gallery. Everything you made, with its settings. Reuse them, view pictures big with zoom, join videos, animate or upscale from there. Pictures keep their settings inside the file, and dropping one onto ComfyUI opens its workflow. Find duplicates compares the pictures of the collection you have open (or the vault) locally, without AI, and stacks the look-alikes: nothing is deleted, you go through a stack and keep what you want. Select 2 to 4 pictures to see them side by side with zoom that follows, or export many at once, with the prompt and settings taken out if you like. Favourite with F, mark with X and delete everything marked at once (favourites are never deleted), sort by date, type, model or LoRA (or look-alikes first), filter and search, and keep pictures in collections with sub-collections a few levels deep (also smart ones that fill themselves by model or LoRA). Videos play with their sound, a time bar and volume in the big viewer, GIFs loop at a speed you set and can be gone through frame by frame, and a slideshow runs through it all. Trim, crop, turn and mirror pictures, GIFs and videos, or save a frame as a picture: always as a new one next to the original. Drag pictures in from a folder or another app, and out again, vault ones included.
A vault. Pictures and videos that only open with your password, encrypted with Argon2id and AES-256-GCM and only ever decrypted in memory. Turn on Private next to Generate and the result goes straight into the vault, without a copy in your folders or in ComfyUI's. Move pictures in later, take them out again, or animate, upscale and edit them without them leaving the vault. Collections and sub-collections inside the vault, whose names are encrypted too and only show while it is open, favourites and marks for deletion. A recovery key, auto lock, and a re-encrypt option when you change the password.
Drives. Keep pictures, videos and vault files on a USB stick or another drive. What is on it shows in the gallery while it is plugged in (vault files once the vault is open too), sorted as before, and nothing is forgotten while it is out. Vault files stay encrypted on the drive, which also keeps an encrypted copy of what is on it.
Deleting for good. Deleted files can go to the Recycle Bin, be shredded (overwritten, renamed and then deleted) or be handed to Eraser if it is installed. Deleting a picture also takes its prompt out of the prompt history and the queue list. Clean up in Settings shows what making pictures leaves behind (uploads and temporary files in ComfyUI, its job history, unused thumbnails, masks, logs) and removes it, and ComfyUI is tidied after every job.
A prompt helper. Small language models that improve, extend, shorten or write prompts, as tags or as sentences
depending on the model, and write motion for videos. Uncensored writing models (Qwen3.5 2B, 4B or 9B) and TIPO, a model
made only for Danbooru tags. The writing model keeps every person, position and action you name and describes the
picture plainly, as a camera would see it. They run on the processor, so the graphics card stays free. All optional: pick them in
Settings and each downloads once (1 to 5 GB) to a folder you pick, or use a .gguf model you already have. Tag models
also get tag suggestions while you type, with a tag blacklist you can edit.
Screen capture blocking. Screenshots, recordings and screen sharing can't see Better Comfy, either always or while the vault is open.
Performance. A card in Settings looks at your PC (page file, memory, power mode, where the models are, the graphics driver) and Better Comfy's own options, and opens the right setting for each recommendation.
Little things. Your own presets, live previews while a picture is drawn, prompt history, Ctrl+Enter to generate,
a Stop button that slides out of Generate while something is made, count arrows that double and halve,
Ctrl+Up and Down to weight a word, {a|b|c} to pick one per picture, accent colours and a tray icon.
Handy. Ctrl+K jumps to any tab, setting, model, LoRA or character, and F1 lists every shortcut. Deletes and a few other
actions can be undone for a few seconds. A panic key hides the app and locks the vault from anywhere. Settings, Privacy has a log of
everything the app connected to since it opened. Light mode, moving backgrounds, an interface size setting, one-zip backups. Pictures you put in
Pictures\Better Comfy\Dot art show up as Dot art backgrounds: the picture made of slowly drifting particles.
What you need
- Windows 10 or 11, 64 bit, and a graphics card that runs ComfyUI.
- ComfyUI. The easiest way is Pinokio: open Discover and install ComfyUI. The portable ComfyUI and ComfyUI Desktop work too.
- Models in ComfyUI's model folders. For pictures, an SDXL checkpoint in
models/checkpointsis enough to start. For videos, a WAN 2.2 image to video pair (or a WAN 2.1 model) inmodels/diffusion_models, its text encoder inmodels/text_encodersand the WAN VAE inmodels/vae. Settings, Setup check lists what is there and what is missing.
Install
Download BetterComfySetup from the latest release and run
it. It installs for your Windows account only, so no admin rights are needed, and it keeps itself up to date.
The setup isn't code signed yet, so Windows SmartScreen may say it "protected your PC". Click More info and then Run anyway. Every release lists the SHA-256 checksum of its files.
On the first start Better Comfy looks for ComfyUI. If it can't find it, pick the ComfyUI folder once and you are done.
Where your files are
| What | Where |
|---|---|
| Pictures | Pictures\Better Comfy (changeable) |
| Videos | Videos\Better Comfy (changeable) |
| Settings, gallery, LoRA and model notes, covers, the vault | %LocalAppData%\BetterComfy |
| Prompt helper (if set up) | the folder you pick |
| The program | %LocalAppData%\Programs\BetterComfy |
Updates only replace program files. Uninstalling removes the program and leaves your pictures, videos, settings and gallery where they are.
Privacy
Everything you make stays on your PC. Better Comfy only talks to the ComfyUI on this computer, and a ComfyUI it starts itself only listens on this computer. There is no account, no telemetry and no upload. Better Comfy goes online to ask GitHub whether a new version exists (you can switch that off under Settings, Updates) and, only when you ask for it, to download prompt helper models or missing model parts once from Hugging Face and GitHub, to read a model's Civitai page when you press Fetch, to talk to Civitai while the Market is open, and to fetch the Character reference add-on from GitHub and its models from Hugging Face if you choose to install it. Covers you see in the Market stay in memory and are never saved. A Civitai API key, if you add one, is kept encrypted for your Windows account and only sent to Civitai with downloads. The prompt helper itself runs on your PC.
Building
You need Python 3.12 and the .NET 8 SDK.
.\build-release.ps1The script sets up the Python packages from requirements.txt, builds Better Comfy and writes the setup, the update zip
with its checksum and a source zip to dist. For development, run python main.py after
pip install -r requirements.txt.
License
Copyright (C) 2026 slashneck. Better Comfy is free software under the GNU General Public License version 3: you can use, change and share it, and any version you share has to stay under the same license with its source code available. The license doesn't cover the Better Comfy name or logo (GPLv3 section 7e), so please give modified versions their own name.
Better Comfy is not made by or connected to the ComfyUI team. It includes Qt, PySide6, FFmpeg, cryptography, argon2-cffi, a Danbooru tag list and the Python runtime. See THIRD-PARTY-NOTICES.md.
