Doc when the Hunyuan img2vid is released will this work with the same app or will it be a new one, w
Doc when the Hunyuan img2vid is released will this work with the same app or will it be a new one, whats your plan?
Fi plan to implement into this app
Fonce diffsynth supports
Fi plan to add hunyuan text to video as well to this app
Fupdated to v19
Fnow testing massed compute and runpod with new torch
Fexcellent news now bf16 model weights exists
Fi will update
Fhalf space
Fimportant fix coming
FGitHub
The current version has updated content: Wan-Video supports multiple Attention implementations. But the effect is far inferior to the branch: wan-train. I tried to use the same picture + the same p...
Jso we are currently affected by this effect?
FFfp32 vs fp16 weights
Flooks same great
Fi am not sure how wide it is
�Your app now supports the fp32 version as well?
J@Dr. Furkan Gözükara Is it possible to control which face gets the inpaint if there's multiple faces? Seems like it doesn't even work with masking/edit. <segment:yolo-face_yolov9c.pt-1,0.7,0.5>
Fnope
FPatreon
Get more from SECourses: FLUX, Tutorials, Guides, Resources, Training, Scripts on Patreon
�so how did you use the fp32? through diffusers?
SDr Furkan this graphic show thta the 5070 ti have AI TOPS Equivalent to 4090 that means that could be same as fast in procces AI task like your app WAN video?
Fwe were using fp32 weights
Fnow we will use fp16 half space 
Fbut you can set precision to fp32 in app which would double VRAM
Fit has lesser VRAM
Fpay attention to cuda cores that is what matters most + vram
Simean 4070 ti super, vs 5070 ti super in AI task thediference of AITOPS ahve any impatc between them?
�So your app was already using those 6 parts of the model and it's fp32 rught? And we are able to use those 6 part foles through your app on 48gb vram?
Fit was using fp32 weights in bf16
Fi expect 10% difference between 2
�Can we now use those weights in fp32 through your app?
Fyes both works
Fjust dont download new model weights :d
�That's great so what changes we need to select in the ui in order to use those weights in fp32
Fchange torch_dtype
Jso instead fp16 using fp32 this will be much slower right? i have to test to see if there is a difference in the video
Fi tested fp16 weight vs fp32 weight
Fno difference
Fbut inference in fp32 i didnt test as it would require
Fastronomical vram 
SDr Furkan this model WAN could work in AMD gpus?
Fnope
Fi mean at least my code
Famd has to bring like 64 gb consumer gpus
Fthen community will make it work
Fbut for mere 24 gb no one gonna waste time
Jor anyone else knows how to fix this?
Sthe incremental in amount of vram remember me when intel only give us 4 cores in mainstrain,,, AMD must do somthing to change this
Jfound this in the docs. You can control the index apparently
