for image to video does it use the same resolution for the 1.3b model?
for image to video does it use the same resolution for the 1.3b model?
Fno it is 720p
F14b model
Fthus it need more vram
Fall right all ready time to test on 80 GB A100 and make kaggle notebook
Ftesting on massed compute A100
Ffull model uses 52.16 gb - 14b image to video
1Is that the FP32?
Fbf16
F
Fbut this is almost as if fp32 vram
Fwe need 1.3b image to video model :d
Fi will test on h200 lets see the speed
Zall very exciting
Fye really cool
Fi will publish on Patreon hopefully today before sleep
Ffinal test
F14b text to video broken atm though i expect fix tomorrow
Fdownload speed with special tech 
Fh200 is fast
Fh200 is 1.84 second id on my rtx 3090 ti around 8.5 second / it - 1.3b model
Ftext to video 14b fits into 24gb vram - min 17 gb
Fh200 is 36 second / it for 720x1280 image to video at 14b for 81 frames
Fwaiting first image to video
Flol i worked full day today at least 14 hours now
Fjust wow
F
YOH
Yyou have to tell me how to do this specific vid from prompt to approach
Yso gooood
Yokay imma try it lol
Ypls tell me you already dropped a tutorial in your patreon
Clooks like he did
Ajust amazing
Aalso refreshed my Patreon subscription
ADoctor I think it is very near kling tbh
AI'm going to try today and experiment with it
Tso around 30 minutes on h200 per generation? That's nuts
Thow many iterations per frame does it do?
A@Dr. Furkan Gözükara what gpu should I rent for the large 14b model on massed ?
Dgood grief, how big are these models - 60+gb to download
Dits a bit confusing how such big models can be run with consumer gpus, but I'm hopeful
Twith extreme quantization and quality loss
TConsumer GPU need to get better, not the other way around
TI was hopeful 5090 release would drop GPU compute prices but they continue to rise. L40S time is ever more expensive.
Tmaybe because libraries does not fully support 5090 as Dr. Furkan reporting.
Seems vary strange how long it takes
Seems vary strange how long it takes
T@Dr. Furkan Gözükara Hello, I watched this video https://www.youtube.com/watch?v=jHlGzaDLkto
which is about how to use 5090 for Swarm UI
However, I got this error .
I got several questions:
@Dr. Furkan Gözükara
which is about how to use 5090 for Swarm UI
However, I got this error .
I got several questions:
- I got this error (can you help me solve it?) Is it 5090-specific?
- I forgot (I asked you before but I since my computer was fried so I don't have the record) where to put the trained Flux Dev models with Dreambooth ; I remember somewhere I need to put something like 'diffusion_models' (?). Can you give me a a specific answer where to create this folder (or if my folder name is corrrct to put my Dreambooth-trained models)
- I forgot how you type the yoloface command in the input textbox (again sorry, I lost the record when my PC is fried)
@Dr. Furkan Gözükara
YouTubeSECourses
In this video I have intensively compared RTX 5090 speed on FLUX DEV, FLUX Fill, SD 3.5 Large, SD 3.5 Medium, Stable Diffusion XL (SDXL) and Stable Diffusion 1.5 (SD 1.5) models. For each benchmark, I have compared RTX 5090 against RTX 3090 TI so we see the speed improvement. Moreover, I have tested FP8 vs 16-bit precision for FLUX and SD 3.5 La...
F80gb best
Fbut 48gb will work too
