Short answer
If you make mostly images, want to choose each file yourself, or already run ComfyUI, HEISS UI fits better. Every missing piece is shown with its size before anything downloads, 21 image families run built in, and you get a searchable gallery, a phone studio and Hidden on top.
If video is the point and your card has 6 to 12 GB, start with WanGP. It covers more video models, including LTX-2, runs select ones from 6 GB, and sets up the speed-ups for you.
Pick HEISS UI if
- You make mostly images, with 21 families built in, and some video on the side.
- You want to see each file, its size and other versions before anything downloads.
- You already run ComfyUI, with your own models, LoRAs and node packs.
- Your own ComfyUI workflows should become simple controls.
- You want a gallery you can search, a phone studio and a locked Hidden space.
Pick WanGP if
- Video is what you make, and you want LTX-2, MiniMax H3, Wan 2.2 and more in one place.
- Your card is small and you’d like builds and settings chosen for it.
- You want audio and speech models next to video.
- You don’t want to run ComfyUI.
Only in HEISS UI
- You choose the files. Sizes, sources and other versions shown before anything downloads.
- Runs on your ComfyUI. With the models, LoRAs and node packs you already have.
- Your ComfyUI workflows. Drop in the JSON and they become simple controls.
- A phone studio. Made for one thumb, with its own sign-in.
- Hidden. An encrypted place for private images.
At a glance.
| Feature | HEISS UI | WanGP | ComfyUI |
|---|---|---|---|
| Who picks files | Yousizes, sources and other versions shown first | WanGPpicked for your hardware on first run | You |
| Engine | Your ComfyUI | Its own | Itself |
| Your own checkpoints | Any file in ComfyUI’s folders | Through finetune definitions | Yes |
| Image models | 21 familieseach set up for you | SeveralKrea 2, Qwen-Image, Z-Image, Flux.2, HiDream | All |
| ComfyUI workflows | Drop in the JSONthey become simple controls | Noits own engine | Native |
| On a phone | Phone studiomade for one thumb | Not covered in its docs | App Mode |
| Private images | Hiddenencrypted | No | No |
| Made for | Images first, video in beta | Video first, images and audio too | Anything you can wire up |
| Video models | 7 familiesWan 2.1, Wan 2.2 5B and 14B, HunyuanVideo 1.5, MiniMax H3 | Manyalso LTX-2, LongCat, Kandinsky | Allthrough templates |
| Small cards | Through ComfyUIwhich moves parts to system RAM; GGUF via ComfyUI-GGUF | From 6 GBfor select models; int8, fp8, GGUF, NVFP4 | With the right files |
| LTX-2 | No | Yes | Yes |
| Audio, speech | No | Yes | Yes |
| Licence | MIT | WanGP Community License 2.0 | GPL-3.0 |
Checked 29 September 2026 against the WanGP README and licence.
Video first or image first.
WanGP started as a way to run Wan on “GPU poor” machines and grew into an app for video, images and audio. Its README lists Wan 2.1 and 2.2 with their offshoots, MiniMax H3, LTX-2 up to 2.5, Hunyuan Video, LongCat and Kandinsky, plus a mask editor, pose and depth extractors, a prompt enhancer and video upscalers. For video it does far more than HEISS UI.
HEISS UI started from images. Its video is marked beta: Wan 2.1, Wan 2.2 5B and 14B, HunyuanVideo 1.5 and MiniMax H3 run built in, Wan 2.2 14B and HunyuanVideo 1.5 also from a start image, and videos get their own gallery.
Who picks the files.
WanGP’s downloads are “architecture-aware”: when you first generate with a model, it fetches the files it considers best for your hardware. That’s convenient, and part of why it works on small cards. Using a checkpoint of your own goes through its finetune definitions, which its docs explain.
HEISS UI does it the other way round. Pick a model and its setup lists each missing part, the text encoder, the VAE, a companion file or a node pack, with its size and source. Where several builds exist, Other versions shows them all. Get all adds up the sizes and checks free space before anything starts, and nothing downloads until you press the button.
WanGP’s README describes downloads that pick the model files suited to your hardware automatically.
Custom models and fine-tunes are added through a finetune definition file.
Small cards.
WanGP runs select models from 6 GB and supports GTX 10-series cards and newer, as well as AMD RDNA 2 to 4. Its install scripts add the speed-ups that fit your setup. On ComfyUI, and so in HEISS UI, a larger model on a smaller card still runs, only slower, because ComfyUI moves parts of it in and out of memory. GGUF builds load through the ComfyUI-GGUF node pack, which HEISS UI installs in one click, though it doesn’t load MiniMax H3 GGUFs yet.
What runs where: 8 GB, 12 GB, AMD Radeon. Model guides: Wan 2.2, MiniMax H3, HunyuanVideo 1.5.
Your ComfyUI, your files.
HEISS UI never installs an engine. It uses the ComfyUI you have, with the models, LoRAs and node packs already in it, and writes images and videos to ComfyUI’s output folder. If you built a video workflow in ComfyUI, export it and import it into HEISS UI: the prompt, size, seed, steps, frames and start image become controls. WanGP keeps its own folders and its own engine, and its README notes it can reuse LoRAs stored for another app.
Where WanGP is the better pick.
Video, especially on a small card: more models including LTX-2, low-memory builds and speed-ups set up for you, plus audio and speech. If that’s your week, WanGP is the easier home.
Questions.
Is WanGP better than ComfyUI for video?
It’s easier on a small card, with builds and speed-ups chosen for you and more video models ready to go. ComfyUI is more flexible and gets models first, but you set up the files and workflow yourself.
Does WanGP use ComfyUI?
No. WanGP has its own engine and web interface. HEISS UI, SwarmUI and Krita AI Diffusion are front ends that run on ComfyUI.
Can WanGP choose which model files I download?
It picks the files it considers best for your hardware when you first generate. Your own checkpoints go in through finetune definitions. HEISS UI shows every part with its size and lets you pick another version.
Which video models does HEISS UI run?
Wan 2.1, Wan 2.2 5B, Wan 2.2 14B, HunyuanVideo 1.5 and MiniMax H3, in beta. Wan 2.2 14B and HunyuanVideo 1.5 also start from an image. LTX-2 isn’t built in yet.
Is WanGP free?
Yes, to use locally, including at a company, under the WanGP Community License 2.0. It isn’t an OSI open-source licence: you may not sell or white-label WanGP itself.
Sources: WanGP on GitHub, WanGP licence, finetunes docs, ComfyUI-GGUF, HEISS UI README.
