WhichClip

Wan local AI video: free and unlimited, if you have the GPU

Running Wan 2.2 on your own GPU is the only AI video path with no credits, no daily cap and no vendor watermark. The weights are Apache 2.0. The catch is hardware: the 5B model needs about 24GB of VRAM (e.g. RTX 4090) for 720p and takes under 9 minutes per 5-second clip. Without that GPU, use hosted free tiers like Google Flow instead.

Last checked: 2026-10-26 · By WhichClip · Independent, not affiliated with any tool vendor

Weekly WhichClip

One email a week: which free credits changed, which tool to use this week, one prompt that worked. No spam.

By signing up you agree we email you WhichClip tips. Unsubscribe anytime. Questions: borderlessdesk.hello@gmail.com

Wan models you can download (checked 2026-10-26)

ModelWhat it makesGPU memory (official README)Speed stated by Wan
Wan2.2-TI2V-5BText-to-video and image-to-video, 720p (1280×704) at 24fpsAt least 24GB VRAM (e.g. RTX 4090) with the README's offload flagsA 5-second 720p video in under 9 minutes on one consumer GPU, without optimization
Wan2.1-T2V-1.3BText-to-video, 480p recommended (720p less stable)8.19GB VRAMA 5-second 480p video in about 4 minutes on an RTX 4090, without quantization
Wan2.2-T2V-A14B / I2V-A14BText- or image-to-video, 480p and 720p (MoE)At least 80GB VRAM on a single GPUNot published for consumer GPUs
Wan2.2-S2V-14BAudio-driven (speech-to-video), 480p and 720pNot published in the lines we readNot published
Wan 3.0Hosted only (e.g. Pika, PixVerse)No weights: Wan-AI's Hugging Face page lists only Wan 2.1 and 2.2-family models—

Figures are quoted from the official Wan-Video GitHub READMEs and Hugging Face model cards. Your speed depends on your card, drivers and settings. Community tools (ComfyUI, Diffusers, LightX2V) can lower memory needs, but those figures come from third parties.

Local Wan vs hosted free tiers

Wan 2.2 on your GPUGoogle Flow freePika / PixVerse free
Cost per clip$0 (your electricity)$0, 50 credits a dayPika: packs only. PixVerse: 30 daily credits
LimitYour GPU timeAbout 5 Lite clips a day. Video may be blocked about 2–5 PM UTCCredits and parallel-job caps
Visible watermarkNone from the vendorToggle (forced in India, South Korea, Vietnam)Pika: none listed. PixVerse: yes on free
Invisible markNone mentioned in the READMESynthID on everythingNot published
Quality ceiling720p, Wan 2.2 (2025)Veo 3.1 with native audioNewer hosted models (Wan 3.0, Seedance)
SetupPython, PyTorch 2.4+, command line (or ComfyUI)Browser sign-inBrowser sign-in
LicenseApache 2.0. Wan claims no rights over your outputsGoogle termsVendor terms (Pika Free: no commercial license)

Run Wan 2.2 locally (5B model, 720p)

  1. Check your GPU. The official README says the 5B command runs on a GPU with at least 24GB of VRAM, e.g. an RTX 4090. With 8–12GB, start with Wan2.1-T2V-1.3B at 480p instead (8.19GB).
  2. Get the code. git clone https://github.com/Wan-Video/Wan2.2.git, then pip install -r requirements.txt. The README asks for torch 2.4.0 or newer.
  3. Download the weights. huggingface-cli download Wan-AI/Wan2.2-TI2V-5B --local-dir ./Wan2.2-TI2V-5B (or via ModelScope).
  4. Generate. python generate.py --task ti2v-5B --size 1280*704 --ckpt_dir ./Wan2.2-TI2V-5B --offload_model True --convert_model_dtype --t5_cpu --prompt "your shot". Add --image for image-to-video.
  5. Prefer a UI? Wan 2.2 has been integrated into ComfyUI and Diffusers (per the Wan README, July 2025).
  6. Write prompts with the shot formula. One subject, one action, one camera move. See the prompt cheat sheet.
Honest hardware note: no NVIDIA GPU with enough VRAM means local Wan isn't really free. Renting a cloud GPU costs money, and laptops without a strong discrete GPU are likely to fail or be very slow. In that case use Google Flow's 50 daily credits or the other free tiers.

FAQ

Is Wan AI free and open source?

Wan 2.1 and Wan 2.2 are. Their weights are on Hugging Face and GitHub under the Apache 2.0 license, and the Wan team says it claims no rights over your generated content. Newer Wan versions such as Wan 3.0 are offered as hosted models; Wan-AI's Hugging Face page lists only Wan 2.1 and 2.2-family models (checked 2026-10-26).

What GPU do I need to run Wan locally?

For Wan2.2-TI2V-5B at 720p, the official README says at least 24GB of VRAM, e.g. an RTX 4090. Wan2.1-T2V-1.3B needs 8.19GB for 480p. The larger A14B models need at least 80GB on a single GPU.

How long does Wan take to make a video?

Wan says TI2V-5B makes a 5-second 720p video in under 9 minutes on one consumer GPU without optimization, and the 1.3B model makes a 5-second 480p video in about 4 minutes on an RTX 4090.

Does local Wan add a watermark?

The official Wan 2.2 README doesn't mention adding a visible or invisible watermark. Generated files stay on your machine. Still label realistic AI video where a platform asks you to.

Can I use Wan videos commercially?

Wan is released under Apache 2.0, which allows commercial use, and the team claims no rights over generated content. The license text also says you're responsible for how you use it. Read the full license before you rely on it.

Want these changes in your inbox? Join Weekly WhichClip (free) ↑

Sources

Vendor pages fetched 2026-10-26 (Dubai). Prices and credits change. Check the vendor page before you pay.

More from WhichClip