Video models that run on 8 GB
18 models with published weights that fit in 8 GB — a phone, a base iPad, an Air. Counted at four-bit quantisation, weights only, leaving the machine about a third of its memory and a gigabyte for context: room for roughly 7 billion parameters. A mixture of experts is counted in full, because it is held in full even though only a few experts compute.
The memory of a phone, a base iPad or an entry-level laptop. What fits is small: models of a few billion parameters, quick and cheap to run, good at summarising, classifying and simple extraction, and out of their depth on long reasoning. This is also where on-device makes the most sense, because the alternative is a network round trip for something that takes a moment.
Models that generate moving footage, most of them now with sound. Priced by the second of output, sometimes doubled for higher resolution, which makes this the most expensive thing in the catalogue to experiment with — a minute of 1080p can cost more than a million tokens of text. Check the maximum clip length and whether it will continue from an image you supply.
- Cosmos-1.0-Diffusion-7B-Text2WorldNVIDIA7.0B≈4.5 GB at 4-bit
- CogVideoX-5bZ.ai5.6B≈3.6 GB at 4-bit
- CogVideoX-5b-I2Vzai-org5.0B≈3.2 GB at 4-bit
- CogVideoX1.5-5B-SATzai-org5.0B≈3.2 GB at 4-bit
- Wan2.2-TI2V-5BWan-AI5.0B≈3.2 GB at 4-bit
- Wan2.2-TI2V-5B-DiffusersWan-AI5.0B≈3.2 GB at 4-bit
- LTX-VideoLTX.io1.9B≈1.3 GB at 4-bit
- CogVideoX-2bZ.ai1.7B≈1.1 GB at 4-bit
- stable-video-diffusion-img2vidStability AI1.5B≈1.0 GB at 4-bit
- stable-video-diffusion-img2vid-xtStability AI1.5B≈1.0 GB at 4-bit
- stable-video-diffusion-img2vid-xt-1-1Stability AI1.5B≈1.0 GB at 4-bit
- i2vgen-xlali-vilab1.4B≈0.9 GB at 4-bit
- text-to-video-ms-1.7bali-vilab1.4B≈0.9 GB at 4-bit
- FastWan-QAD-FP8-1.3BFastVideo1.3B≈0.8 GB at 4-bit1 also selling it hosted
- Wan2.1-T2V-1.3BWan-AI1.3B≈0.8 GB at 4-bit1 also selling it hosted
- Wan2.1-T2V-1.3B-DiffusersWan-AI1.3B≈0.8 GB at 4-bit
- Wan2.1-VACE-1.3BWan-AI1.3B≈0.8 GB at 4-bit
- stable-virtual-cameraStability AI1.3B≈0.8 GB at 4-bit