I tried adding Hunyuan's LoRA to this project. So far, the test results are fairly okay, but I'm not sure if this is correct. I hope this can inspire others. I think most people need this feature.

[chenpipi0807]

I tried adding Hunyuan's LoRA to this project. So far, the test results are fairly okay, but I'm not sure if this is correct. I hope this can inspire others. I think most people need this feature.

[chenpipi0807]

This is the test result of this LoRA for me:
https://civitai.com/models/1094156/titty-drop-hunyuan-video-lora?modelVersionId=1490124

通过网盘分享的文件：混元lora测试
链接: https://pan.baidu.com/s/1VtLpp0hIXYaBzMHocNCcLg?pwd=6666 提取码: 6666
--来自百度网盘超级会员v5的分享

[PumaAI487]

did you need start and end frame for it to work? tried just using start frame but no beuno

[Tablaski]

Thanks I've tried lots of things, the only times I could make a LoRa work it had to be through loading the full unquantized BF16 model and quite slow. I've been able to generate a good short 5s video with your work, but it seems the LoRas i've tried with it don't end up being strong. On a 30 s video it was barely moving, on 10 s it doesn't move that much.

That might be due to the LoRas being initially trained for 5 s though. For people that have already generated many longer videos using other means, would be interested to know how well the existing loras are faring

[chenpipi0807]

I only tried adding the initial frame and generating a 5 - second video, so my test may not be complete.

[Tablaski]

It seems to recognize the (video) Hunyuan LoRas i've tried (from civitai) as LyCoRis also, which is weird ?

For the duration i'm wondering if that could help to repeat/renew the prompt to later timeframes

Edit : i've generated several 10 s videos with several LoRas, it's definitely working. But sometimes barely... depending on the starting picture (I haven't been using and ending frame) and the prompt. It feels though that the LoRas are not at full power yet. I'm considering trying your loading node coupler with https://github.com/nirvash/ComfyUI-FramePackWrapper samplers that allow prompting at different time intervalls. Maybe it could reinforce the LoRa effect past the 5 - 10 s mark ?

[ethanfel]

Lora support has been available for a little while on the dev branch. They work very well and you can use multiple.

[Dhrhciebcy]

Why is there no issues tab in this project? I have 64GB of RAM and 6GB of VRAM and ComfyUI always goes into torch.OutOfMemoryError, how is that possible? Shouldn't it start also with 6GB of VRAM?

[ethanfel]

Why is there no issues tab in this project? I have 64GB of RAM and 6GB of VRAM and ComfyUI always goes into torch.OutOfMemoryError, how is that possible? Shouldn't it start also with 6GB of VRAM?

because Kijai isn't actively developing this (he can't dev everything) and doesn't want to have people posting issues all the time.
You only have 6gb of vram ?

[DavidChiapixel]

Has anyone encountered the access violation when running on ComfyUi?
This access violation happened consistently every time I run the workflow despite changing Pytorch several times
My Environment
OS: Windows 11 10.0.26100
GPU: NVIDIA RTX 4060 Laptop GPU (8GB VRAM)
RAM: 16GB
Python: 3.12.8150.1013
PyTorch Versions Tried: from 2.4 to 2.7 with cuda ranging from 121 to 128
Numpy: 1.26.4 (I have downgraded for compatibility concerns)
The workflow is basically the same as KJ with minor revision
Kijai_FLF2V_framepack1.json

Logs
`loaded completely 5149.0 470.1210079193115 True
STREAM b'IHDR' 16 13
STREAM b'caBX' 41 68962
b'caBX' 41 68962 (unknown)
STREAM b'IDAT' 69015 65536
Windows fatal exception: access violation

Current thread 0x000068b4 (most recent call first):
File "C:\Users\raven\Documents\ComfyUI\venv\Lib\site-packages\torch\storage.py", line 469 in getitem
File "C:\Users\raven\Documents\ComfyUI\comfy\utils.py", line 58 in load_torch_file
File "C:\Users\raven\Documents\ComfyUI\comfy\sd.py", line 722 in load_clip
File "C:\Users\raven\Documents\ComfyUI\nodes.py", line 970 in load_clip
File "C:\Users\raven\Documents\ComfyUI\execution.py", line 183 in process_inputs
File "C:\Users\raven\Documents\ComfyUI\execution.py", line 194 in _map_node_over_list
File "C:\Users\raven\Documents\ComfyUI\execution.py", line 222 in get_output_data
File "C:\Users\raven\Documents\ComfyUI\execution.py", line 347 in execute
File "C:\Users\raven\Documents\ComfyUI\execution.py", line 525 in execute
File "C:\Users\raven\Documents\ComfyUI\main.py", line 183 in prompt_worker
File "D:\Installation\Python\Lib\threading.py", line 1012 in run
File "D:\Installation\Python\Lib\threading.py", line 1075 in _bootstrap_inner
File "D:\Installation\Python\Lib\threading.py", line 1032 in _bootstrap

Thread 0x000046f8 (most recent call first):
File "D:\Installation\Python\Lib\asyncio\windows_events.py", line 774 in _poll
File "D:\Installation\Python\Lib\asyncio\windows_events.py", line 445 in select
File "D:\Installation\Python\Lib\asyncio\base_events.py", line 1954 in _run_once
File "D:\Installation\Python\Lib\asyncio\base_events.py", line 640 in run_forever
File "D:\Installation\Python\Lib\asyncio\windows_events.py", line 322 in run_forever
File "D:\Installation\Python\Lib\asyncio\base_events.py", line 673 in run_until_complete
File "C:\Users\raven\Documents\ComfyUI\main.py", line 315 in `