Novita AI for ads

What Novita AI does for ad work

Novita AI sells two things from one account. The first is GPU rental. A graphics card (GPU) is the chip behind AI images and video, and Novita rents them in its cloud by the second. The second is model APIs, where you send a prompt and pay per output with nothing to install.

For ad work, the GPU side is where you run open models yourself. You start an instance from Novita’s official ComfyUI template, load FLUX or Wan, and generate stills or clips until you stop it. The Runpod hub covers running ComfyUI on a rented card in full.

Novita’s price table lists four single cards, and their memory decides which models fit. The RTX 4090 has 24 GB and the RTX 5090 32 GB. The RTX 6000 Ada has 48 GB, the H100 SXM 80 GB.

The API side is where a marketer buys clips by the second. MiniMax H3, a video model, costs $0.08 per second of video at 768P on Novita’s pricing page (read 12 November 2026). On that date its Image filter lists one model, Ming Image 0.1 Design, and no FLUX or Seedream, so FLUX stills mean renting a GPU.

What Novita costs per hour and per clip

A 6-second MiniMax H3 clip at 768P costs $0.48, from the per-second rate on the pricing page. On the GPU side, an RTX 5090 costs $0.56 an hour on demand and an H100 SXM $3.39.

Spot instances cost about half: the H100 SXM drops to $1.70 an hour. The RTX 5090 shows no spot price. Storage bills on top, past a free 60 GB container disk.

Every GPU and storage rate, the H3 Max rates and the billing rules are on Novita AI pricing. Cloud GPU prices sets Novita beside the other GPU clouds, and the prices table carries its row. A new account’s free credit and the missing promo code are covered in Novita AI promo code and free credits.

Can you use Novita outputs in paid ads?

Yes for most ad categories, once the model you run allows it. Novita’s terms, updated 5 August 2026, leave output rights open. You keep copyright in what you upload: “You retain copyright and any other proprietary rights that you may hold in the Input.” The terms say nothing about who owns the output.

Section 10.1 gives Novita a licence to your output only “for the purpose of providing the Service to you”. Section 10.2 says that “By Default, Novita AI will not use your Content to train our own models or to improve the Services.”

Terms of Service, updated 5 August 2026

10. User content You retain copyright and any other proprietary rights that you may hold in the Input.

10.1. Grant of License By making Input available to the Service, you grant Novita AI a worldwide, royalty-free, and non-exclusive license to reproduce, view, and use the Input for the purpose of providing the Service to you (including, without limitation, to generate Output).

7. Purchases and payment All fees are non-refundable and exclusive of any applicable taxes, which the Customer is solely responsible for paying.

Some ad categories are out whatever the model. Section 3 says the services “may not be used in connection with: (i) intoxicants, illegal drugs, or other controlled substances; (ii) alcoholic beverages; or (iii) games of chance.” Section 9(2) bars content “that falsely depicts a real person without their consent, in a manner that could deceive a reasonable person.”

You carry the legal risk. Section 24 has you “defend, indemnify, and hold us harmless” against third-party claims, intellectual property among them. Section 23 caps Novita’s liability at the fees you paid in the prior six months or $500, whichever is less. An Enterprise customer whose cap is negotiated in a signed contract gets that cap instead.

The Acceptable Use Policy, section 5(6), also bars use “as part of any effort to compete with Novita AI”.

The model’s licence decides the rest:

• Wan 2.2, run on a rented GPU, is released under Apache 2.0 (model card), which allows commercial use.
• MiniMax H3 is sold on Novita’s API as the “overseas edition”, and its API reference names no licence. MiniMax’s community licence, dated 2 August 2026, covers hosted APIs. Its section V.4 bars using the model “or any of their Outputs or results outside the Applicable Territory”, which excludes the US, the EU, the UK and South Korea.

Novita does not say which document governs H3 clips bought through its API. So whether they can run in ads in those four markets is unclear. The rights table holds Novita’s row by use.

Billing, spot reclaim and refunds

The meter starts before you generate anything. Novita’s GPU pricing docs say billing begins “when the instance is successfully created and enters the pulling state”, while it downloads its software image, and stops “when the instance is stopped”. Time is “accurate to the second and settled hourly”.

Spot instances trade price for interruption. The spot page promises “Default 1-hour protection, 1-hour advance termination notice”. After the first hour, Novita can take the card back with an hour’s warning. That suits a batch of stills you can rerun, less so a long video render.

Money you load stays loaded. Section 8 of the terms says “All fees are non-refundable” and adds that Novita “does not offer refunds for unused credits”. The exceptions are a verified outage that leaves the service “materially unavailable for an extended period”, a duplicate or incorrect charge, and refunds the law requires, claimed within 30 days.

One Trustpilot reviewer says a balance of $100 is needed to run more than three GPU instances, a rule the site does not show (Trustpilot). Load a small amount first.

Users praise Novita’s model serving and fault its filters

29 public sources from December 2025 to November 2026 mention Novita. Fifteen are Reddit threads, 8 Hacker News comments and 6 Trustpilot reviews. Almost all of them concern Novita’s language-model API, often reached through OpenRouter. Only one touches its GPU instances.

Quality of serving leads the praise. Users say open models run at full quality on Novita where cheaper hosts degrade them (6 of 29 sources: Hacker News, Reddit). Low token prices with frequent discounts come next (4 of 29 sources: Reddit, Hacker News).

Content filters lead the complaints. Users report prompts refused or cut off that other hosts answer (5 of 29 sources: Reddit, Reddit). For ad work that matters less than it sounds: the reports concern chat models, and none involve image or video.

Some models served by Novita hallucinate, stall or return empty replies (4 of 29 sources: Reddit, Hacker News). Credits drain faster than expected (3 of 29 sources: Trustpilot, Trustpilot).

Novita AI pros and cons

Novita AI alternatives

Runpod is the closest match: the same per-second GPU rental, with ComfyUI as an official template and network volumes that keep your models between sessions. Its RTX 4090 costs $0.34 an hour on Community Cloud and $0.89 on Secure Cloud, read 13 November 2026.

So Novita’s on-demand RTX 4090 sits a cent under Runpod’s Community Cloud rate at list price. Novita’s spot rate, which can be reclaimed after an hour, is half that. Novita AI vs Runpod sets every GPU and rule side by side.

• All-in-one suites

  Novita AI

  GPU cloud billed by the second, RTX 4090 from $0.33 an hour, plus MiniMax H3 video by API at $0.08 a second.

  $0.08 per second of video at list price · checked 12 Nov 2026

  Best for Marketers running open models on a rented RTX 4090 or 5090 who want the lowest list price per hour, or MiniMax H3 clips by API.

• All-in-one suites

  Runpod

  GPU cloud billed by the second: rent an RTX 4090 from $0.34 an hour and run ComfyUI, Wan 2.2, LTX-2 or FLUX yourself.

  $0.01 per image at list price · checked 13 Nov 2026

  Best for Marketers running open models such as Wan 2.2, LTX-2 or FLUX in ComfyUI, paying by the second for GPU time: about $0.01 an image.

For other GPU clouds and hosted apps, see Runpod alternatives. To set a rented card up, how to run ComfyUI on Runpod goes step by step, and Wan 2.2 on Runpod covers video. The steps carry over to any cloud that runs a ComfyUI image.

Questions buyers ask about Novita AI

Is Novita AI good for ad work?

For renting a GPU at a low list price, yes. Users praise its full-quality model serving (6 of 29 sources) and fault its content filters (5 of 29). Those reports concern its language-model API; almost none cover its GPUs.

Does Novita AI have an API for video?

Yes. MiniMax H3 costs $0.08 per second of video at 768P, read 12 November 2026. The same account sells APIs for language and image models too.

Which GPUs does Novita AI rent?

Four single cards on its price table: the RTX 4090, RTX 5090, RTX 6000 Ada and H100 SXM. All but the RTX 5090 also come as cheaper spot instances.

Is Novita AI legit?

Yes. Novita AI publishes its GPU prices, billing rules and dated terms (5 August 2026). Public reports praise its serving (6 of 29 sources) and fault its filters (5 of 29). Its Trustpilot profile is unclaimed, with 9 reviews at a TrustScore of 2.5. The 6 from the last 12 months are the ones counted here.