v3.0.2 - major overhaul

[C0untFloyd]

Major changes/improvements only:

• Streamlined UI
• Rewrote threading/processing
• Added processing of video parts only
• Most models using ONNX Architecture now (memory hungry but can be run in parallel and supported on more platforms)
• Selective overriding of fps possible
• Trimming of top of face mask, usefull for hairlines
• Added streaming to OBS virtual cam
• Post-processing major speed increase

The number of threads for a system with 8 Gb GPU shouldn't be higher than 2! Running codeformer enhancement you might even need to use just 1.

---

This discussion was created from the release v3.0.2 - major overhaul.

[Siraj-HM]

Applied providers: ['CUDAExecutionProvider', 'CPUExecutionProvider'], with options: {'CUDAExecutionProvider': {'do_copy_in_default_stream': '1', 'cudnn_conv_algo_search': 'EXHAUSTIVE', 'device_id': '0', 'gpu_external_alloc': '0', 'enable_cuda_graph': '0', 'gpu_mem_limit': '18446744073709551615', 'gpu_external_free': '0', 'gpu_external_empty_cache': '0', 'arena_extend_strategy': 'kNextPowerOfTwo', 'cudnn_conv_use_max_workspace': '1', 'cudnn_conv1d_pad_to_nc1d': '0', 'tunable_op_enable': '0', 'tunable_op_tuning_enable': '0', 'enable_skip_layer_norm_strict_mode': '0'}, 'CPUExecutionProvider': {}}
inswapper-shape: [1, 3, 128, 128]
Module does not exist
Traceback (most recent call last):
  File "E:\roop-unleashed\installer\installer_files\env\lib\site-packages\gradio\queueing.py", line 388, in call_prediction
    output = await route_utils.call_process_api(
  File "E:\roop-unleashed\installer\installer_files\env\lib\site-packages\gradio\route_utils.py", line 219, in call_process_api
    output = await app.get_blocks().process_api(
  File "E:\roop-unleashed\installer\installer_files\env\lib\site-packages\gradio\blocks.py", line 1437, in process_api
    result = await self.call_function(
  File "E:\roop-unleashed\installer\installer_files\env\lib\site-packages\gradio\blocks.py", line 1123, in call_function
    prediction = await utils.async_iteration(iterator)
  File "E:\roop-unleashed\installer\installer_files\env\lib\site-packages\gradio\utils.py", line 503, in async_iteration
    return await iterator.__anext__()
  File "E:\roop-unleashed\installer\installer_files\env\lib\site-packages\gradio\utils.py", line 496, in __anext__
    return await anyio.to_thread.run_sync(
  File "E:\roop-unleashed\installer\installer_files\env\lib\site-packages\anyio\to_thread.py", line 33, in run_sync
    return await get_asynclib().run_sync_in_worker_thread(
  File "E:\roop-unleashed\installer\installer_files\env\lib\site-packages\anyio\_backends\_asyncio.py", line 877, in run_sync_in_worker_thread
    return await future
  File "E:\roop-unleashed\installer\installer_files\env\lib\site-packages\anyio\_backends\_asyncio.py", line 807, in run
    result = context.run(func, *args)
  File "E:\roop-unleashed\installer\installer_files\env\lib\site-packages\gradio\utils.py", line 479, in run_sync_iterator_async
    return next(iterator)
  File "E:\roop-unleashed\installer\installer_files\env\lib\site-packages\gradio\utils.py", line 629, in gen_wrapper
    yield from f(*args, **kwargs)
  File "E:\roop-unleashed\installer\roop-unleashed\roop\ui.py", line 691, in start_swap
    batch_process(list_files_process, use_clip, clip_text, processing_method == "In-Memory", progress)
  File "E:\roop-unleashed\installer\roop-unleashed\roop\core.py", line 284, in batch_process
    process_mgr.initialize(roop.globals.INPUT_FACES, roop.globals.TARGET_FACES, options)
  File "E:\roop-unleashed\installer\roop-unleashed\roop\ProcessMgr.py", line 90, in initialize
    p = str_to_class(module, classname)
  File "E:\roop-unleashed\installer\roop-unleashed\roop\utilities.py", line 272, in str_to_class
    return class_ or None
UnboundLocalError: local variable 'class_' referenced before assignment

Do I need to download any checkpoints ?

Edit: I think I found the issue when I previewed and then started Faceswap; it crashed.

[Lordmau5]

Getting the same error when I select Codeformer for the post-processing, the other 2 work fine though

Seems like the class and filenames need to be in correct casing.

Renaming the class in roop/processors/Enhance_Codeformer.py to Codeformer instead of CodeFormer and changing it to also say Codeformer here fixed it for me:
https://github.com/C0untFloyd/roop-unleashed/blob/main/roop/ProcessMgr.py#L69

[C0untFloyd]

Yes, merging the dev branch didn't rename the codeformer file with correct upper-/lowercase. Fixed it a second ago, please update!

[Niiickel]

Launching roop unleashed - please edit windows_run.bat to customize commandline arguments
Check for Updates? [y/n]y
HEAD is now at 628b7e8 Hotfix: renaming of codeformer class
Already up to date.

but new commit is not downloaded or does the windows_run.bat only update main releases? So how do I get the new commit?

[C0untFloyd]

Renamed it directly in github, please try updating again. Running windows_run.bat is correct.

[Siraj-HM]

Codeformer is really slow in my 8GB VRAM 3070 than GFPGAN

[C0untFloyd]

Is this your first roop install? It used to be about 10-20 times slower than this. Compared to back then, this is lightning speed! Or are you using more than 2 threads?

[kyl688]

Can you add the output path?

[kyl688]

And I've just got this error when add the files from a folder. It was ok with the older version. I've just updated 15 minutes ago.

[C0untFloyd]

You know you can just drag'n drop files onto the files box? That box is really meant for Colab use which you don't seem to use.

[kyl688]

When I process multiple files, the first one is ok but from the second it has something wrong like this.

[kyl688]

Yes, I know that I can drag n drop files. Just want to let you know it was ok with the old version but not this new update.

[kyl688]

Is there anyway to go back to the old version? The old one seems good enough for me. I can wait until you fix those bugs to update.

[DarthKX]

older version was working, I get this on all machines now:
ProcessMgr' object has no attribute 'swap_mode

Any idea why? Thx in advance

[vicento]

same here and also on Colab install

[C0untFloyd]

Can one of you just dump the complete error output here? I can't reproduce it and I'm bad at guessing...

[DarthKX]

Traceback (most recent call last):
File "C:\AI\roop-unleashed\installer_files\env\lib\site-packages\gradio\routes.py", line 523, in run_predict
output = await app.get_blocks().process_api(
File "C:\AI\roop-unleashed\installer_files\env\lib\site-packages\gradio\blocks.py", line 1437, in process_api
result = await self.call_function(
File "C:\AI\roop-unleashed\installer_files\env\lib\site-packages\gradio\blocks.py", line 1123, in call_function
prediction = await utils.async_iteration(iterator)
File "C:\AI\roop-unleashed\installer_files\env\lib\site-packages\gradio\utils.py", line 508, in async_iteration
return await iterator.anext()
File "C:\AI\roop-unleashed\installer_files\env\lib\site-packages\gradio\utils.py", line 501, in anext
return await anyio.to_thread.run_sync(
File "C:\AI\roop-unleashed\installer_files\env\lib\site-packages\anyio\to_thread.py", line 33, in run_sync
return await get_asynclib().run_sync_in_worker_thread(
File "C:\AI\roop-unleashed\installer_files\env\lib\site-packages\anyio_backends_asyncio.py", line 877, in run_sync_in_worker_thread
return await future
File "C:\AI\roop-unleashed\installer_files\env\lib\site-packages\anyio_backends_asyncio.py", line 807, in run
result = context.run(func, *args)
File "C:\AI\roop-unleashed\installer_files\env\lib\site-packages\gradio\utils.py", line 484, in run_sync_iterator_async
return next(iterator)
File "C:\AI\roop-unleashed\installer_files\env\lib\site-packages\gradio\utils.py", line 853, in gen_wrapper
yield from f(*args, **kwargs)
File "C:\AI\roop-unleashed\roop-unleashed\roop\ui.py", line 691, in start_swap
batch_process(list_files_process, use_clip, clip_text, processing_method == "In-Memory", progress)
File "C:\AI\roop-unleashed\roop-unleashed\roop\core.py", line 328, in batch_process
process_mgr.run_batch_inmem(v.filename, v.finalname, v.startframe, v.endframe, fps,roop.globals.execution_threads, skip_audio)
File "C:\AI\roop-unleashed\roop-unleashed\roop\ProcessMgr.py", line 239, in run_batch_inmem
future.result()
File "C:\AI\roop-unleashed\installer_files\env\lib\concurrent\futures_base.py", line 451, in result
return self.__get_result()
File "C:\AI\roop-unleashed\installer_files\env\lib\concurrent\futures_base.py", line 403, in __get_result
raise self._exception
File "C:\AI\roop-unleashed\installer_files\env\lib\concurrent\futures\thread.py", line 58, in run
result = self.fn(*self.args, **self.kwargs)
File "C:\AI\roop-unleashed\roop-unleashed\roop\ProcessMgr.py", line 172, in process_videoframes
resimg = self.process_frame(frame)
File "C:\AI\roop-unleashed\roop-unleashed\roop\ProcessMgr.py", line 299, in process_frame
gender = 'F' if self.swap_mode == "all_female" else 'M'
AttributeError: 'ProcessMgr' object has no attribute 'swap_mode'

I also used all male option here, not sure why the debug info calls for all female.

I just tried now with CUDA and first found option for the face, this time it's running. But it's the first video running with this new version (which looks great btw!), I tried many and it would give me that same error, on 2 different PCs (RTX 3080 & RTX 3060)

I also didn't know if tensorrt or CUDA is best, which is faster in your opinion? Before I only used CUDA and I will up what someone else said, would love the ability to run the older version too, just to have a backup to turn back to while some bugs are fixed. And like a idiot I updated my main working instance. I did duplicate the roop_unleashed folder, but I can't seem to make this copy run either.

If not too hard to do, having access to previous version would be great.

Thx in advance

[DarthKX]

It does seem to work with CUDA however now... maybe it was the TensorRT that was breaking things... I'm not 100% sure, I need to make more test. The bug where it continues to process after the file is done, seems to also be hit or miss. The processing bar now goes into thousands of % :), before it just updated the XXX/Xs one in the top right corner.

[DarthKX]

in the previous version this would stop when clicking stop, here it doesn't :(, so I suppose a refresh of the page or restarting the app to fix this? As such, access to the previous version would really be appreciated if it's not too much work on your part, of course.

I should mention that this issue is random (and was random with previous version too), sometimes it happens, sometimes not.

Thanks in advance.

Oh and I randomly get this error too:
[mov,mp4,m4a,3gp,3g2,mj2 @ 00000297e3384740] moov atom not found

But that was happening before as well

[vicento]

hi, have this error on Colab and also local install :
Error
ProcessMgr' object has no attribute 'swap_mode

[vicento]

work if i choose CPU only, Coreml bugs on mac now

[DarthKX]

This can run on Mac too? Is there a tutorial on how to install it?

[C0untFloyd]

This is coded in Python and uses a Web-UI so it certainly runs on Mac,Linux, Windows whatever. Default installation doesn't use the special hardware though. There's no tutorial yet, so I'm linking the old roop wiki pages here:
Basic Setup

Enable HW-Acceleration (select CoreML in Settings Tab after that)
pip uninstall onnxruntime onnxruntime-coreml pip install onnxruntime-coreml==1.13.1

[C0untFloyd]

Note to myself:

Never ever hastily release something you merged together without a full day of testing again.
Closing this discussion now as all bugs should be resolved with v3.0.5

[ShrekLovesAll]

I only just updated and have to say that this overhaul is a banger. Great stuff countfloyd. Actually I understated how insane this is. I was in shock when I saw a codeformer enhanced swap with CPU face detection fly through 2000 frames in mere minutes. It really works the CPU and CUDA together well. And the memory management is so much better now too.

[DarthKX]

Don't be too hard on yourself, C0untFloyd. The new changes are great, the interface is better that way, and overall you did an amazing job! Bugs will be bugs :D, that will happen on any dev project ;) . I also look forward to test the Text Masking option. Keep up the amazing work, and thanks again for your work.