Short answer
For the everyday loop of prompt, look and keep, HEISS UI is quicker. It runs many models and video on the ComfyUI you already have, sets each one up for you, and adds a phone studio and an encrypted Hidden space.
If you work on one image at a time, selecting regions, inpainting and painting control layers, InvokeAI’s canvas is the better tool. Nothing here matches it. For building pipelines, there’s ComfyUI’s graph.
Pick HEISS UI if
- You mostly write prompts, compare results and keep the good ones.
- You already run ComfyUI and want to use the models and node packs you have.
- You want each model ready to prompt, and its missing pieces in one click.
- You want more video in the same studio: Wan 2.2, HunyuanVideo 1.5 and MiniMax H3, in beta.
- Your own ComfyUI workflows should work as simple controls.
- You want a phone studio and a locked, encrypted place for private images.
Pick InvokeAI if
- You inpaint, outpaint and paint over results on a canvas with layers.
- You use regional prompts, control layers and reference images while you edit.
- You want one app with its own engine, without running ComfyUI.
- Several people share one install. Invoke has a multi-user mode.
Only in HEISS UI
- Runs on your ComfyUI. The models and node packs you already have, nothing reinstalled.
- Your ComfyUI workflows. Drop in the JSON and they become simple controls.
- More video. HunyuanVideo 1.5 and MiniMax H3, next to Wan 2.1 and 2.2.
- 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 | InvokeAI | ComfyUI |
|---|---|---|---|
| Made for | Prompt, look, keep | Editing one image | Building workflows |
| Engine | Your ComfyUI | Its own | Itself |
| ComfyUI workflows | Drop in the JSONthey become simple controls | Its own format | Native |
| Video | BetaWan 2.1 and 2.2, HunyuanVideo 1.5, MiniMax H3 | Wan 2.2through the workflow editor | Yes |
| Missing pieces | Listed firstsizes shown, one click | Starter bundlesin the Model Manager | Templateslist missing files |
| On a phone | Phone studiomade for one thumb | Not covered in its docs | App Mode |
| Private images | Hiddenencrypted | No | No |
| Image models | 21 families | ManyFlux.2, Krea 2, Z-Image, Qwen-Image, Ideogram 4 | Allusually on release day |
| Inpaint, outpaint | No | Yeson the canvas | With nodes |
| Regions, control layers | No | Yes | With nodes |
| Multiple users | No | Yes | No |
| Licence | MIT | Apache-2.0 | GPL-3.0 |
Checked 29 September 2026 against the InvokeAI README, its 6.14 release notes and invoke.ai.
Canvas or prompt box.
InvokeAI’s Unified Canvas is the heart of it. You generate into layers, mask a region and inpaint it, extend the image outwards, draw a sketch for a control layer, and give one part of the picture its own prompt. It has SAM segmentation for picking objects and a gallery organised in boards. If your work is making one image right, this is a very good way to do it.
HEISS UI is for the other way of working: many images, quickly, and keeping the good ones. You write a prompt, previews stream in, you queue the next one, and the gallery keeps prompts, models and LoRAs searchable. For changing an image it has a start image with a Change slider, reference images for Flux.2 and Qwen-Image 2.1, and SeedVR2 upscaling with a compare slider. No brushes, no masks.
Invoke after Adobe.
Invoke’s founding team joined Adobe, and the hosted Invoke platform shut down. The open-source app carries on: stewardship passed to Lincoln Stein and Vic (blessedcoolant), core maintainers since the start, and it’s funded through GitHub Sponsors. It’s actively developed. Version 6.14 in August 2026 added Wan 2.2 video, Krea 2, Flux.2 Dev, ERNIE Turbo, Ideogram 4, multi-GPU and Intel XPU support, and 6.14.2 followed on 27 September.
The hosted platform shut down as the founding team joined Adobe; two long-time core maintainers now lead the project with the community.
6.14 notes list per-model gaps openly, for example no reference images for Krea 2 and text-to-image only for Ideogram 4.
Models and video.
Invoke supports a long list: SD 1.5 to SDXL and SD 3.5, the Flux.1 family, Flux.2 Dev and Klein, Z-Image, Krea 2, Anima, Qwen-Image and its edit model, Ideogram 4 and ERNIE-Image. Its release notes say which features each new model has on the canvas. Video arrived in 6.14 with Wan 2.2, through the workflow editor and its templates.
HEISS UI runs 21 image families and 7 video families (beta) through ComfyUI, from SD 1.5 to Flux.2, Qwen-Image 2.1, Krea 2 and Ideogram 4, and Wan 2.1, Wan 2.2, HunyuanVideo 1.5 and MiniMax H3 for video. Guides: Qwen-Image, Flux.2 Dev, Wan 2.2.
Where InvokeAI is the better pick.
Inpainting, outpainting, regional prompts, control layers, and several people on one install. If your work is making one image exactly right, Invoke’s canvas is the tool for it.
Running both on one computer.
They don’t share an engine, so each loads its own copy of a model into memory. Run one at a time on a single graphics card. Model files can be shared on disk: Invoke’s Model Manager installs from a local path, and ComfyUI reads extra folders from extra_model_paths.yaml.
Questions.
Is InvokeAI still maintained after Adobe?
Yes. The founding team joined Adobe and the hosted platform closed, but the open-source app continues under long-time maintainers. Version 6.14.2 came out on 27 September 2026.
Is InvokeAI better than ComfyUI?
For editing on a canvas, with inpainting, layers and regional prompts, many find it easier. ComfyUI is more flexible, gets new models first and runs any pipeline you can build.
Does InvokeAI use ComfyUI?
No. It has its own engine and its own node workflows. ComfyUI workflow files don’t run in it.
Does HEISS UI have inpainting like Invoke?
No. HEISS UI has start images, reference images for Flux.2 and Qwen-Image 2.1, and SeedVR2 upscaling, but no masks or canvas. For inpainting, InvokeAI or Krita AI Diffusion are the better tools.
Can InvokeAI make video?
Yes, since 6.14 in August 2026, with Wan 2.2 through the workflow editor. HEISS UI runs Wan 2.1, Wan 2.2, HunyuanVideo 1.5 and MiniMax H3 in beta.
Sources: InvokeAI on GitHub, 6.14.0 release notes, invoke.ai, ComfyUI, HEISS UI README.
