Get Nook
  1. Nook
  2. Labs
  3. nl.VI
  4. NL-2610.007

NL-2610.007Ongoing · Set up[nl.VI]

Which Wan video models run on an 8 GB card?

The Nook team

Opened 5 October 2026 · 2 min read

Abstract

Nook's Video page runs Wan models through stable-diffusion.cpp, and now offers four text-to-video models: Wan 2.1 1.3B in 8 bits, and Wan 2.2 TI2V 5B, Wan 2.1 14B and Wan 2.2 A14B in 4 bits. Only Wan 2.1 1.3B has run on our RTX 4060 so far, at about five minutes for a two-second clip. The 5B model is listed as needing 8 GB, the 14B models 16 GB. We want to know what actually runs on 8 GB, how long a clip takes, how much video memory (VRAM) each model really uses, and where each one starts to spill out of it.

Stage
Set up
Comments
Only the 1.3B model timed so far. RTX 4060 8 GB; stable-diffusion.cpp master-841-6b3edaa.
Subjects
Images and video (nl.VI); Memory (nl.MM)
Cite as
NL-2610.007 · usenook.ai/labs/wan-video-on-8-gb.html

Full text

The question

Which open video models can someone with an 8 GB graphics card actually use, and what does each one cost them in time? Model pages list a minimum, but a minimum says nothing about how long a clip takes, or what happens a little below it.

The models

ModelFileListed needOn our card
Wan 2.1 T2V 1.3BQ8_0; 832 × 480 at 16 fps; the text encoder runs on the processorFits 8 GBRuns: about 5 min for a 2 s clip
Wan 2.2 TI2V 5BQ4_K_M, with the Wan 2.2 VAE; 832 × 480, 49 frames at 24 fps8 GBNot yet run
Wan 2.1 T2V 14BQ4_K_M16 GBNot yet run
Wan 2.2 T2V A14BQ4_K_M, a low-noise and a high-noise model; 10 steps plus 8 high-noise steps, CFG 3.516 GBNot yet run

Each model also needs the umt5 text encoder, 4.1 GB, which Nook currently downloads once per model.

The plan

  1. Close everything else that uses the card, and record the free memory before each run (NL-2610.004 shows why).
  2. Run each model on the same three prompts at its default size, length and steps.
  3. Record the time per clip, peak dedicated memory and peak shared memory, so a spill shows (NL-2610.002).
  4. Run the 14B models on the 8 GB card anyway, to see whether they finish at all once they spill, and how slowly.
  5. Publish the clips beside the numbers, so the quality can be judged too.

So far

  • All four models are in Nook's catalogue, with sizes and checksums, and marked against the card with the half-gigabyte allowance from NL-2610.004.
  • Wan 2.1 1.3B runs on the RTX 4060, with its text encoder on the processor: a two-second clip at 832 × 480 takes about five minutes. The other three have not been run on a graphics card yet.

This entry will be revised as the runs come in, and becomes a result when they are done.

Run it on your own GPU.

Create AI videos and images on your own computer with Nook. Open models, local generation, no generation credits.