i got to sleep hopefully see you tomorrow
i got to sleep hopefully see you tomorrow
VHave a good night, tty soon
XGG's captain
Fi am trying to add teacache myself
Ffirst try with teacache
F1.3b teacache works but rest not 
Fok this time gonna work i assume
J
Fyes almost there
JI need it
Fye doing final tests
Fas it does more steps it will become faster
FJsame quality
Fyes i am testing 0.15 now
F0.1 didnt speed that much
FGitHub
Teacache v2
I am open to contributions suggestions
left original right is teacache
output.mp4
Prompt: super fast car going on a high way
Negative Prompt: still ...
I am open to contributions suggestions
left original right is teacache
output.mp4
Prompt: super fast car going on a high way
Negative Prompt: still ...
F0.15
Fthis really did speed up
Jnot bad at all
F704 to 490 second
Ftesting 0.2
Fthere is certainly some quality loss though
Fwow 0.2
F0.2 : Generation Duration: 315.25
F0.2 increased duration to 315/705
Jnot good
Fi agree
Fi think we can set as 0.15
Ftesting 720p model
Fi also need to update models to single files
Fthey will be used for training
JDon't know if any of you guys also run into this? https://discord.com/channels/1243166023859961988/1349721178700386364
XI pressed start after editing this part of "Hugging_Face_Upload_Download_V12.ipynb" inside of runpod
XAND GETING THIS ERROR
FPatreon
Get more from SECourses: FLUX, Tutorials, Guides, Resources, Training, Scripts on Patreon
FPatreon
Get more from SECourses: FLUX, Tutorials, Guides, Resources, Training, Scripts on Patreon
X@Dr. Furkan Gözükara do i need to upload "workspace/trained_models/config_dreambooth-20250311-230222.toml" into my model file in swarm_ui also?
JI'm not at home. I'll test teacache on wan 2.1 when I get home
Xokay finally uploaded my .safetensors to huging face and genrated frist wan 480p video :))))
XXTrying 720pn now image to video
Xgetting an error because of 99% mem utilization.
Flooks amazing
TRequired a clean install for me, but its definitely working. Very nice speed increase.
TThe speedup for later iterations is phenomenal. Makes increasing step on the generations more attractive.
TKLING 1.6 Pro mode now generates 5s@30fps img2vid in 60 seconds.
How fast is this model, insane compared to WAN 2.1
How fast is this model, insane compared to WAN 2.1
Teven assuming they use h200's on backend
Xgetting this error trying to run 720p image to video