GPU for AI video generation from €0.30 an hour
Wan 2.2 5B makes 720p clips on a 24 GB card such as the RTX 4090. The 14B model gives better motion and needs 48 GB or more: an RTX A6000, L40S or an 80 GB card for speed. On VRAMfarm the video app opens in your browser with the model downloaded, from €0.30 an hour, billed by the minute.
Which video model runs on which card
| GPU | Memory | SDXL | Flux | Qwen-Image | Wan 2.2 5B video | Wan 2.2 14B video | Price |
|---|---|---|---|---|---|---|---|
| RTX 3090 | 24 GB | ✓ yes | ✓ yes | ✓ yes | ✓ yes | — | €0.30/h |
| RTX 4090 | 24 GB | ✓ yes | ✓ yes | ✓ yes | ✓ yes | — | €0.59/h |
| RTX 5090 | 32 GB | ✓ yes | ✓ yes | ✓ yes | ✓ yes | — | €0.76/h |
| RTX A6000 | 48 GB | ✓ yes | ✓ yes | ✓ yes | ✓ yes | ✓ yes | €0.72/h |
| L40S | 48 GB | ✓ yes | ✓ yes | ✓ yes | ✓ yes | ✓ yes | €1.06/h |
| RTX PRO 6000 | 96 GB | ✓ yes | ✓ yes | ✓ yes | ✓ yes | ✓ yes | €2.12/h |
| A100 | 80 GB | ✓ yes | ✓ yes | ✓ yes | ✓ yes | ✓ yes | €1.46/h |
| H100 | 80 GB | ✓ yes | ✓ yes | ✓ yes | ✓ yes | ✓ yes | €3.90/h |
Whether our ready-made app runs on the card: SDXL needs 12 GB, Flux 16 GB, Qwen-Image and Wan 2.2 5B 24 GB, Wan 2.2 14B 48 GB. With your own setup, offloading and smaller formats go lower, but slower. Prices are for one card per hour and change with the market; the price is fixed when you launch. Billing is per minute, minimum 15 minutes.
What you can make
Text to video
Describe the scene and get a 5-second clip. Good for concepts, social posts and storyboards.
Image to video
Animate a picture: a product shot, an illustration or a frame from Flux. The 14B image-to-video model keeps the look of the source well.
Batches overnight
Queue dozens of prompts in ComfyUI and let the card work. We email you if it sits idle after the queue ends.
Your own pipeline
HunyuanVideo, LTX-Video or your own models on a clean server with PyTorch, up to 192 GB on one card.
5B or 14B
The 5B model renders 1280 × 704 at 24 fps and fits 24 GB. The 14B model is two experts of 14B each, renders 480p or 720p at 16 fps, and moves more naturally. Start with 5B to find prompts, then switch to 14B for final clips.
How long a clip takes
Video is the heaviest AI task: a 5-second clip takes minutes, not seconds. On an RTX 4090 the 5B model needs 4–9 minutes per clip; the 14B model with a 4-step LoRA takes 2–4 minutes on an H100 and 7–14 on an RTX A6000. The Wan 2.2 guide has measured ranges.
Big downloads
The 14B app downloads about 70 GB of models after the start, which takes 10–15 minutes. You can save /workspace on stop, but the base models of ready-made apps are downloaded again on each launch.
What an hour of video costs
An hour on an RTX 4090 with Wan 2.2 5B: about €0.59, roughly 7–15 clips at 1280 × 704. An hour on an H100 with the 14B model: about €3.90.
Ready-made apps for this
The app installs itself with its models and opens in your browser. You pay only for the GPU time.
Guides on this
12 November 2026 Wan 2.2 video on a rented GPU: 5B or 14B, how much VRAM and how long per clip
Wan 2.2 5B renders 5 s of 720p on a 24 GB card in 4–9 minutes; 14B needs 48 GB and looks much better. Which card, clip limits and cost per clip.
11 November 2026 Cost per image, video clip and million tokens on a rented GPU, worked out
1,000 Flux images, one 5-second Wan 2.2 clip, a million tokens from 8B to 70B, a Flux LoRA: all-in costs with live euro prices and the maths shown.
11 November 2026 ComfyUI on a rented GPU: your own models, LoRAs, custom nodes and the 502 page
Run ComfyUI on a rented GPU from €0.59 an hour: when it is ready, where models and LoRAs go, how to add custom nodes and keep them for next time.
Frequently asked questions
Which GPU do I need for Wan 2.2?
The 5B model runs on 24 GB (RTX 3090, 4090). The 14B model needs about 30 GB at 720p, so our app starts it on 48 GB cards and up; with offloading it can run on 24–32 GB, but much slower.
Can I generate longer videos?
The models are trained on about 5 seconds. Longer videos are made by extending clips from their last frame or by joining several clips, which ComfyUI workflows can do.
Can I use the videos commercially?
Wan 2.2 is released under Apache 2.0, which allows commercial use. Check the licence of any other model you add.
Why is the first clip slow?
The first run loads the model into memory and compiles kernels. The following clips in the same session are faster.
Make your first clip
Top up from €10, open the video app on a 24 GB or 48 GB card and write a prompt.