Run Stable Diffusion on a cloud GPU from India: ComfyUI or Forge in minutes, paid in rupees
Tutorial · Published 2 November 2026 · prices on this page are live
Stable Diffusion and FLUX want a GPU with real memory, and a laptop in Pune or Hyderabad rarely has one. Renting is the usual answer: a 24 GB card for the hours you draw, paid in rupees, with ComfyUI or Forge already installed. This is the whole process, start to finish, with today’s prices.
Which card you need
The model decides the memory. Stable Diffusion 1.5 is happy on 8 GB. SDXL wants 12 to 16 GB for comfortable 1024-pixel work. FLUX.1 dev in its full form fills a 24 GB card; the quantised versions fit 12 to 16 GB with a little less quality. Training a LoRA for SDXL needs 24 GB. The cheapest card for each size today:
| For | USD | INR | NPR |
|---|---|---|---|
| Stable Diffusion 1.5, an hour of drawing | USD 0.08 | INR 8 | NPR 12 |
| SDXL at 1024 pixels, an hour | USD 0.08 | INR 8 | NPR 12 |
| FLUX.1 dev, an hour | USD 0.18 | INR 17 | NPR 26 |
| A full afternoon on a 24 GB card, four hours | USD 0.72 | INR 68 | NPR 104 |
Step by step
- Create an account and add credits. Google or email; then Credits, where the smallest top-up is USD 5.00, INR 200 or NPR 200. Indian cards need international transactions switched on in the bank’s app.
- Pick the software. ComfyUI (image & video generation) if you like node graphs and want FLUX and video models; Stable Diffusion WebUI Forge if you want the classic prompt-and-sliders interface with the best memory handling; Stable Diffusion WebUI (A1111) if you have extensions that only run there.
- Pick the GPU. From the live board, a card with the memory from the table above. Every offer shows the host’s country, machine and measured reliability; the price is locked for that machine when you rent.
- Wait for Running, then open the app. About a minute on a warm host, a few minutes on a cold one. The interface opens in your browser; nothing is installed on your computer.
- Load a model. The official images download popular checkpoints on first start; for others, paste a Hugging Face or Civitai link into the model manager, or upload a file. A 6 GB checkpoint arrives in a minute or two on a datacentre connection.
- Draw. Generate, upscale, inpaint, build a workflow. Download the images you keep; everything else stays on the machine’s disk.
- Stop when you are done. GPU billing ends at once. Keep the disk, with your models and workflows, at the storage rate shown before renting, so the next session starts without downloads; destroy the instance to stop paying for the disk too.
What a batch of images costs
A thousand SDXL images at 1024×1024 with 30 steps, on today’s cheapest hosts of each card:
| GPU | Typical time | USD | INR | NPR |
|---|---|---|---|---|
| RTX 3090 | 2.4 h | USD 0.72 | INR 68 | NPR 106 |
| RTX 4090 | 1.2 h | USD 0.72 | INR 68 | NPR 106 |
| A100 SXM4 | 1.2 h | USD 0.78 | INR 74 | NPR 116 |
| RTX 5090 | 1.1 h | USD 0.83 | INR 77 | NPR 123 |
| RTX A5000 | 2.8 h | USD 1.04 | INR 98 | NPR 154 |
Faster cards cost more per hour and finish sooner, so the price per image is closer than the hourly rate suggests. If you draw in long sessions, the RTX 4090 usually gives the best price per image; for a quick experiment the cheapest 12 GB card is fine.
Tips from the first week
- Set a timer on the machine when you rent it, or the idle stop, so a tab left open overnight does not cost a night.
- Keep models on the instance disk between sessions rather than re-downloading: 100 GB of disk costs far less per day than an hour of GPU time.
- Want the model to draw your product or your face? Train a LoRA first: the Kohya tutorial takes about an hour on an RTX 4090.
- The ComfyUI guide covers workflows, custom nodes and the video models in more depth.
Questions
Can I run FLUX on a rented GPU?
Yes. FLUX.1 dev in full precision needs a 24 GB card; the quantised versions run on 12 to 16 GB. ComfyUI ships with FLUX workflows, and the model manager pulls the weights from Hugging Face.
Do I need to install anything?
No. ComfyUI, Forge or A1111 opens in your browser from the console. Your own computer only needs a browser; the GPU work happens on the rented machine.
Where are my images stored?
On the rented machine’s disk, which stays with the instance until you destroy it. Download the images you keep; nothing is stored by PowerPod itself beyond the disk you rent.
Can I use Civitai models and LoRAs?
Yes. Paste the download link into the model manager, or upload the file through the app. Check the licence of each model for commercial use.