i will first do wan 2.1 lora
i will first do wan 2.1 lora
TI think this number changes if you select 14B co-efficients. this is also a bit confusing. I also notice torchcompile kills loras
F0.15 only changes with 1.3b model
Tso you like 0.15 for I2V and T2V 14B right? I didn't buy 5090 to run 1.3B models 
Twill you use musubi-tuner? I just can't get good lora's for wan with the same dataset as used for good loras on hunyuan. I actually think captions might be required this time for more flexibilty during prompting. But not sure yet. I didn't caption hunyuan and they came out good, sometimes better than flux with kohya.
Fytes 0.15 for all 14b model
Fyes i plan to use musubi
Flots of people using lots of bug fixes
TMaybe I need to try again with later code. Also this guy just created a gui for it. https://github.com/Kvento/musubi-tuner-wan-gui there is also a gui from ttplanet https://github.com/TTPlanetPig/Gui_for_musubi-tuner - this is gradio
Fi plan to make my own gradio
TI just modified TTplanet's one. its a good starting point. Let me know how your Wan Lora success is. I'm going to keep experimenting, and will let you know what i discover too. I think there is a reason character lora's aren't being released as fast as "motion/animation" loras
Fsure. i am testing 0.2 teacache right now
Fit takes less than 10 min on h100 highest quality - 20 steps
TAlso TorchCompileModelWan (for wan21) breaks Lora's on 50xx but not 30xx. Must be the 3.3 triton, or torch nightlies. I just tested. Same workflow works fine on 30xx. but not 50xx with Torchcompile running on both.
Fvery likely
Fwow teacache 0.2 not bad
Fand super fast
TIts hard to find people with direct 5090 experience still. Shortages are tough.
TI was using 0.16 then 0.2 - but i was reading another thread where they reported 0.25 being the perfect spot
TNot my research - direct from Kijai: Benchmarks. Everything here is Wan2.1 720 @ 49 frames, sageattention, cuda 12.6, 3080 ti, listing thresh, starts, times, and quality observations
0.300 (rel_l1_thresh) a lot more subtle grainy noise than 0.250
0.00 (start) -> 13:08 (encoding time)
0.2875
0.00 -> 13:39 more grainy noise
0.280
0.00 -> 13:18 92% starts noisy
0.275
0.00 -> 14:55, near indistinguishable from 0.25
0.270
0.00 -> 14:41, 96%, nearly perfect
0.265
0.00 -> 16:09, 96%, near perfect
0.250 -> looks lossless
0.10 -> 16:30
2x run:
Best so far, but slower,
0.250
0.10 (start) -> 16:20
Near if not lossless at 0.250, at the cost of about 3 minutes when compared to 0.300.
in summary
0.300 ~ 13:00 I found a bit too grainy for my liking, though speed is nice
0.275 ~ 14:55 great balance between speed and quality
0.250 ~ 16:30 I found was basically lossless
0.300 (rel_l1_thresh) a lot more subtle grainy noise than 0.250
0.00 (start) -> 13:08 (encoding time)
0.2875
0.00 -> 13:39 more grainy noise
0.280
0.00 -> 13:18 92% starts noisy
0.275
0.00 -> 14:55, near indistinguishable from 0.25
0.270
0.00 -> 14:41, 96%, nearly perfect
0.265
0.00 -> 16:09, 96%, near perfect
0.250 -> looks lossless
0.10 -> 16:30
2x run:
Best so far, but slower,
0.250
0.10 (start) -> 16:20
Near if not lossless at 0.250, at the cost of about 3 minutes when compared to 0.300.
in summary
0.300 ~ 13:00 I found a bit too grainy for my liking, though speed is nice
0.275 ~ 14:55 great balance between speed and quality
0.250 ~ 16:30 I found was basically lossless
WThat is odd...
WHello everyone
WCan somone hel me to solve this?
WIm trying to merge two models in SD
Wbut im getting this error
W
WIts a VRam problem?
Astable-virtual-camera https://stability.ai/news/introducing-stable-virtual-camera-multi-view-video-generation-with-3d-camera-control?utm_source=x&utm_medium=social&utm_campaign=virtual+camera
Stability AI
Introducing Stable Virtual Camera, currently in research preview. This multi-view diffusion model transforms 2D images into immersive 3D videos with realistic depth and perspective—without complex reconstruction or scene-specific optimization.
AGitHub
Stable Virtual Camera: Generative View Synthesis with Diffusion Models - Stability-AI/stable-virtual-camera
Fyep. select cpu to load
Iwith the wan 2.1 zip file you provide is triton also included or do i need to install it manually? also is teacache using more vram?
Fincluded
Fteacache shouldnt impact vram
T@Furkan Gözükara SECourses have you seen a xformers fix for blackwell yet? I searched but haven't seen progress. I saw someone compiled wheels for python 3.12 but not 3.10, was thinking to upgrade venv for kohya to 312. Also for wan21. lora training, you must use captions, or the lora model can't do anything, no flexibility beyond the source images similarity. I tried at least 4 tests using musubi. Also still struggling to get good likeness as compared to hunyuan
Fi am working on this for weeks
FGitHub
Just compiled with below pip freeze (venv310) C:\f\x10>pip freeze annotated-types==0.7.0 colorama==0.4.6 deepspeed @ https://files.pythonhosted.org/packages/3d/65/1a6394f5d6dee851e9ea59e385f6d64...
FGitHub
I have compiled xFormers on xformers-0.0.30+c5841688.d20250306 torch==2.7.0.dev20250228+cu128 triton-3.2.0+git8f9b005b the compile worked I am able to install. Python 3.10.11 - Windows 11 However I...
Ti can't get it to compile, but someone did.
Fi compiled many times
Fit doesnt work
CHi guys, do you know of any AI for subtle 2D animations that are clean?
CJust very subtle movement for a cutout 2D image
Fwan 2.1 is best atm
H@Furkan Gözükara SECourses Have you seen this? https://riflex-video.github.io/
It looks like an amazing new way for video extension
It looks like an amazing new way for video extension
IHey @Furkan Gözükara SECourses I am running the wan2.1 on H100 14b model, while text to video works great, I'm having issues with Image to video, I'm following every step of your tutorials but when I try to generate i get an error (no lens). what am I doing wrong?
Fyes i told authors to add this but they didnt do yet
Fyou didnt download model?
Ahi. I get this error while installing Wan2.1 Windows ver 46:
How can I solve?
How can I solve?
Amessage.txt3.85KB
