Run Wan 2.2 and LTX on your own GPU
The models are open weights, so the model was never the hard part. The hard part is the ComfyUI graph, the custom node versions that have to agree with each other, the VRAM tuning and the twenty settings nobody documents. SimpliGen ships those as presets you pick from a list.
The models, already wired
Wan 2.2 and a lightweight low VRAM build of it, LTX 2.5, LTX Director and MiniMax H3 for video; FLUX.2 Dev and Qwen-Image for stills. Each one is a preset carrying its workflow, its node pins and its settings.
No graph to build
Pick a preset, write a prompt, generate. The node graph exists, runs underneath and is the reason the output looks right, but you never open it.
Nothing to maintain
When a model moves or a custom node breaks against a new engine build, we ship a new pack version and you click update. That is the part that eats evenings when you self-host it.
Will Wan 2.2 fit on your card? Four questions, no sign-up
Video models are decided by VRAM more than by anything else, and the honest answer differs between an 8 GB laptop and a 12 GB desktop. Answer four questions for a verdict built from what those same cards really do in our telemetry.
What it is not, so you can stop reading if this is you
- It is not a Stable Diffusion front end. We ship current open weight models, and SD 1.5 and SDXL are not among them. If you want an interface for the Stable Diffusion family specifically, this is the wrong tool and we would rather you knew now.
- You cannot import your own ComfyUI workflows. Presets are authored against our render pipeline so that updates can be shipped safely. If your value is in graphs you built yourself, keep ComfyUI; it is free and good.
- It is Windows and NVIDIA only. No macOS or Linux build, and the local engine is CUDA based so AMD and Intel cards cannot run it. The full requirements list the cards.
- The models are large and you download them once. A single video model can be 10 to 30 GB, ideally on an SSD. You can also point SimpliGen at model files you already have instead of fetching them again.
If the check above said yes, you are a download away from Wan 2.2 running on your own card. If it said no, Cloud generation runs the same presets on our GPUs. Either way the licence is refundable within 14 days if it will not run properly on your computer.
Questions
- Which Wan versions are included?
- Wan 2.2 Video, and a lightweight low VRAM build of Wan 2.2 for smaller cards. They ship as preset packs that update in place, so a newer Wan release reaches you as a pack version rather than a reinstall.
- Can I run Wan 2.2 on 8 GB of VRAM?
- The lightweight build is the one aimed at smaller cards, at lower resolutions and shorter durations. Expect 30 to 60 minutes for a 5 second clip on an 8 GB laptop. Start at 480p and 5 seconds, then scale up. The hardware check on this page gives the verdict for your exact card.
- Is it ComfyUI underneath?
- Yes. SimpliGen embeds ComfyUI with a portable Python runtime and renders the preset into a real workflow before submitting it. When we hit a bug in ComfyUI itself we fix it upstream, so everyone gets the fix whether they bought anything or not.
- Do I have to download the models myself?
- No. Each preset knows its models and fetches them with progress tracking, and presets share model files so installing a second preset from the same family is usually small. If you already have the files, you can point SimpliGen at them.
- Does it run LTX 2.5 as well?
- Yes, LTX 2.5 and LTX Director both ship as packs, alongside MiniMax H3. LTX and Wan want different settings and each preset carries its own, which is most of what SimpliGen is for.