InvokeAI is a studio for artists: a canvas with layers, inpainting and outpainting, generation with SDXL and FLUX, and node workflows for repeated jobs. On gpu.nz it starts from the InvokeAI template on a rented card, so you need no GPU of your own and the interface opens in the browser. Pick a card with 16 GB or more, wait for the start, and open the link from the server page.
In short
InvokeAI runs on cards with 16 GB or more. The most affordable suitable card right now is the RTX A4000 16 GB from 13 ₽ per hour, billed by the minute.
Updated 13 October 2026
SDXL works well in 12–16 GB: on an RTX 4080 with 16 GB a 1024×1024 image takes seconds. FLUX in fp8 needs 16 GB or more, and full-precision bf16 needs 24 GB, so take an RTX 4090 or another 24 GB card. Large batches and resolutions are easier on an L40S with 48 GB.
InvokeAI puts the canvas at the centre of the work: the picture, layers, masks and selections stay separate and change independently, and nodes appear only for repeated processes. ComfyUI builds everything as a graph, while Forge keeps the familiar A1111 interface. If you paint, fix and extend pictures on a canvas, choose InvokeAI. For full control of the graph, see ComfyUI.
| Task | Memory needed | What to rent |
|---|---|---|
| SDXL, images up to 1024×1024 | 12–16 GB | RTX 4080 (16 GB) |
| FLUX in fp8 | 16 GB or more | RTX 4080 (16 GB), RTX 4090 (24 GB) |
| FLUX in bf16 with headroom | 24 GB |
Prices and availability come from the live catalog and update automatically.
| Large batches and resolutions | 48 GB | L40S (48 GB) |
A Tesla T4 with 16 GB also handles SDXL, but generation on it is noticeably slower.
Models are not part of the template, so you download only what you need. Take a 100 GB disk: FLUX dev weighs about 24 GB and SDXL about 7 GB, and models add up quickly. The simplest route is the Model Manager: paste a file link or a Hugging Face repository ID.
For a large file, download it in the JupyterLab terminal with a direct link, then add it to the Model Manager by its local path.
wget -O sdxl_base.safetensors "<direct file link>"
# then add the file in the Model Manager by its local pathFor Civitai links add your token; closed Hugging Face models need an access token.