Currently, DraftCraft is optimized and bundled with a specific image diffusion model to ensure maximum stability and precise geometry matching. While manual swapping is technically possible since all files are stored locally in a single folder, official updates or alternative models are not supported or planned at this time.

The aspect ratio is locked to 1:1 and the resolution to 1024x1024 px. This constraint is tied to the underlying AI model architecture to guarantee the most stable, accurate results. While expanding these options is technically possible, they are strictly fixed for this version. For sharper details, we recommend using an upscaler rather than generating at a higher native resolution.

Yes. The cost of your Personal license is automatically deducted, so you only pay the difference.

Developed and tested on RTX 40 series with 8 GB VRAM (~10 s per image). RTX 30 series is expected to work — it uses the same officially supported CUDA architecture — but has not been tested by us. RTX 20 series is untested and not recommended: the PyTorch build we install may not support it. GTX cards and AMD/Intel GPUs will not work.

Yes, that is the benchmark configuration (~10 s per image). The "Keep Models on GPU" option is for 12+ GB cards.

By design: it shows live progress, per-step timing and VRAM usage. It is also the first place to look if anything behaves unexpectedly.

Yes — client presentations, competition boards, publications.

RealVisXL V4.0 Lightning + Depth ControlNet, downloaded from Hugging Face under their own licenses (OpenRAIL++ — commercial use of outputs permitted). Models are not bundled with the add-on.

No. One purchase, unlimited local generations.