ti2v-models

[YIFanH]

HI, thank you for your excellent work in open source,I want to try using the model ability of the video below, but I couldn't find the corresponding model download link, do you have any open source plans this week?

[hyang0511]

The model will be online today.

[maazel-rhymes]

HI, thank you for your excellent work in open source,I want to try using the model ability of the video below, but I couldn't find the corresponding model download link, do you have any open source plans this week?

The model is now public.

[YIFanH]

thanks

[ai-anchorite]

thanks for releasing!

be interesting to see other people's experiences.

I'm finding it spiking to 30GB vram during the pre-inference phase (with off-load), which is taking ages,
then inference acts very similar to t2v at about 8GB. The vae and masking phase doubles the total generation time however.

Around 90 minutes on a 3090.

Looks like input images need to be landscape ratio (16:9). the portrait i tried on the test-run (doh) was zoomed in, so assume a hard crop.

Very curious about prompting. Are long, descriptive prompts required?

[xiao2mo]

Hi, is there any chance for training code release?