Does anyone have any information on when Amina-Turbo will be released?
https://redd.it/1th0u76
@rStableDiffusion

balancing batch automation vs manual cherry-picking for large character sets?

hey guys, quick workflow question for the power users here!

so i’ve been trying to scale up my generation workflow lately, basically running huge batches using wildcards and dynamic prompts to cycle through a massive library of different character styles and fandom concepts.

the issue i’m running into is consistency vs time. if i let a massive script run overnight, i get tons of variety, but the quality is all over the place and i end up spending hours just manually cherry-picking the good gens and weeding out the bad hands/weird anatomy. but if i micro-manage every single prompt and seed, it takes forever.

how do you guys optimize your pipelines when you're generating a ton of content? do you rely heavily on automated scoring/filtering tools, or do you just accept the manual curation grind as part of the process?

would love to know how you keep your sanity while managing a huge output volume lol. thanks!!

https://redd.it/1th5dbk
@rStableDiffusion

Full Head swap model that make sure Facial features are so strong as well as head size matching of the target

Hey guys, I hope everyone is having great day.
I'm currently working on a project where I need to swap entire head between two images.
I have tried all sort of models, both open source and commercial and always got stuck between two priorities when one gets fulfilled the other doesn't.
First priority is that facial features should look so strong so that the person is so well recognizable as the source.

Second ( which is what most commercial models fail with), is that head should be resized to match target.

Third (not really strong priority semi priority) : adaption of body color or style, for example changing body color slightly to match head color of the source.

There other things like, Copying Facial emotions from target and head position, but these are not priorities. For commercial models I think i have tried every possible model out there.
And for open source models, I have tried bfs with Qwen basically have tried everything in this repo https://huggingface.co/Alissonerdx/BFS-Best-Face-Swap and it worked well for head size matching target, but facial expressions got so weak.
I was wondering can I find a workflow that fulfills my priorities very well, even if it requires large models size.

https://redd.it/1th6ohk
@rStableDiffusion

HY World + Sharp, 360 Panorama Gaussian Splat

https://redd.it/1th9yg0
@rStableDiffusion

LumiPic: Oumoumad's (LTX lora fame) SDR->HDR conversion LoRAs for Qwen, soon Kline Base 4 & 9
https://huggingface.co/oumoumad/LumiPic

https://redd.it/1thc7m8
@rStableDiffusion

Installing ComfyUI + PyTorch for AMD ROCm 7.2, using official drivers.

https://preview.redd.it/0z8hpiefc02h1.png?width=1280&format=png&auto=webp&s=a188e2160d709030b7ed661d4acd4a42f5bde886

Just upgraded my desktop (16G Radeon RX 9070 XT, 32G system RAM) from ROCm 6.4 to ROCm 7.2 on Windows 11 and this is the process I went through. Sources of information:

* [https://www.reddit.com/r/ROCm/comments/1qj9eom/rocm\_72\_official\_installation\_instructions/](https://www.reddit.com/r/ROCm/comments/1qj9eom/rocm_72_official_installation_instructions/)
* [https://www.reddit.com/r/ROCm/comments/1qj9eom/rocm\_72\_official\_installation\_instructions/](https://www.reddit.com/r/ROCm/comments/1qj9eom/rocm_72_official_installation_instructions/)
* [https://rocm.docs.amd.com/projects/radeon-ryzen/en/latest/docs/install/installrad/windows/install-pytorch.html#](https://rocm.docs.amd.com/projects/radeon-ryzen/en/latest/docs/install/installrad/windows/install-pytorch.html#)

Edit: this is a fairly manual process where I only install what is need to run. There are probably easier ways to get ComfyUI working, such as using the ComfyUI desktop installer, or using the AMD "A.I. bundle" that is an optional part of the Adrenalin installer.



# Install Python 3.12.0 (later version should work too)

[https://www.python.org/ftp/python/3.12.0/python-3.12.0-amd64.exe](https://www.python.org/ftp/python/3.12.0/python-3.12.0-amd64.exe)

MD5 checksum: 32ab6a1058dfbde76951b7aa7c2335a6

I choose "custom install" and set "Install for all users" to "c:\\Program Files\\Python312" (Note:

Make sure you let it add the python paths to the environment)

# Install AMD Display driver Adrenalin Edition 26.2.2

(Uninstall any old driver first. I used the AMD Cleanup Utility: [https://drivers.amd.com/drivers/amdcleanuputility.exe](https://drivers.amd.com/drivers/amdcleanuputility.exe) from [https://www.amd.com/en/resources/support-articles/faqs/GPU-601.html](https://www.amd.com/en/resources/support-articles/faqs/GPU-601.html))

[https://drivers.amd.com/drivers/whql-amd-software-adrenalin-edition-26.2.2-win11-c.exe](https://drivers.amd.com/drivers/whql-amd-software-adrenalin-edition-26.2.2-win11-c.exe) [https://www.amd.com/en/resources/support-articles/release-notes/RN-RAD-WIN-26-2-2.html](https://www.amd.com/en/resources/support-articles/release-notes/RN-RAD-WIN-26-2-2.html)

Note: my GPU is used for A.I. only, I don't know if this driver is any good for gaming.

# Install ComfyUI

cd d:\\ (assuming it will be installed at d:\\ComfyUI

git clone [https://github.com/comfyanonymous/ComfyUI.git](https://github.com/comfyanonymous/ComfyUI.git)

cd d:\\ComfyUI

(Activate virtual env)

py -V:3.12 -m venv 3.12.venv

.\\3.12.venv\\Scripts\\activate

At this point, if you try to run ComfyUI you'll get an error about "CUDA not found", that is because the PyTorch installed by default is the NVIDIA/CUDA version.

# Install ROCm 7.2 specific version of PyTorch via PIP

a. Setup ROCm environment.

pip install --no-cache-dir [https://repo.radeon.com/rocm/windows/rocm-rel-7.2.1/rocm\_sdk\_core-7.2.1-py3-none-win\_amd64.whl](https://repo.radeon.com/rocm/windows/rocm-rel-7.2.1/rocm_sdk_core-7.2.1-py3-none-win_amd64.whl) [https://repo.radeon.com/rocm/windows/rocm-rel-7.2.1/rocm\_sdk\_devel-7.2.1-py3-none-win\_amd64.whl](https://repo.radeon.com/rocm/windows/rocm-rel-7.2.1/rocm_sdk_devel-7.2.1-py3-none-win_amd64.whl) [https://repo.radeon.com/rocm/windows/rocm-rel-7.2.1/rocm\_sdk\_libraries\_custom-7.2.1-py3-none-win\_amd64.whl](https://repo.radeon.com/rocm/windows/rocm-rel-7.2.1/rocm_sdk_libraries_custom-7.2.1-py3-none-win_amd64.whl) [https://repo.radeon.com/rocm/windows/rocm-rel-7.2.1/rocm-7.2.1.tar.gz](https://repo.radeon.com/rocm/windows/rocm-rel-7.2.1/rocm-7.2.1.tar.gz)

b. Install torch, torchvision and torchaudio for ROCm AMD GPU support.

pip install --no-cache-dir [https://repo.radeon.com/rocm/windows/rocm-rel-7.2.1/torch-2.9.1%2Brocm7.2.1-cp312-cp312-win\_amd64.whl](https://repo.radeon.com/rocm/windows/rocm-rel-7.2.1/torch-2.9.1%2Brocm7.2.1-cp312-cp312-win_amd64.whl)

[https://repo.radeon.com/rocm/windows/rocm-rel-7.2.1/torchaudio-2.9.1%2Brocm7.2.1-cp312-cp312-win\_amd64.whl](https://repo.radeon.com/rocm/windows/rocm-rel-7.2.1/torchaudio-2.9.1%2Brocm7.2.1-cp312-cp312-win_amd64.whl) [https://repo.radeon.com/rocm/windows/rocm-rel-7.2.1/torchvision-0.24.1%2Brocm7.2.1-cp312-cp312-win\_amd64.whl](https://repo.radeon.com/rocm/windows/rocm-rel-7.2.1/torchvision-0.24.1%2Brocm7.2.1-cp312-cp312-win_amd64.whl)

At this point you can type "python main.py" to launch comfy. I can generate 1024x1024 anima images with er\_sde + simple 30 steps in about 28 seconds.

If you have problem generating image you can try the following.

# Confirm that PyTorch is correctly installed

1. Verify if Pytorch is installed and detecting the GPU compute device. python -c "import torch" 2>nul && echo Success || echo Failure Expected result: Success
2. Enter command to test if the GPU is available. python -c "import torch; print(torch.cuda.is\_available())" Expected result: True
3. Enter command to display installed GPU device name. python -c "import torch; print(f'device name \[0\]:', torch.cuda.get\_device\_name(0))" Example result: device name \[0\]: Radeon RX 7900 XTX device name \[0\]: <Supported AMD GPU>
4. Enter command to display component information within current environment. python -m torch.utils.collect\_env Example result:PyTorch version: 2.9.1+rocm7.2.1 Is debug build: False CUDA used to build PyTorch: N/A ROCM used to build PyTorch: 7.2.53211-158bd99533OS: Microsoft Windows 11 Pro (10.0.26200 64-bit) GCC version: Could not collect Clang version: Could not collect CMake version: Could not collect Libc version: N/A

When I tried it I got some strange warning, but the test seems to run ok: Here is the log:

C:\\Users\\AMD> python -c "import torch" 2>nul && echo Success || echo Failure Success

C:\\Users\\AMD> python -c "import torch; print(torch.cuda.is\_available())" Program: Unknown command line argument 'Files\\Python312\\Lib\\site-packages\_rocm\_sdk\_core\\lib\\llvm\\bin\\offload-arch.exe'. Try: 'C:\\Program --help' True

C:\\Users\\AMD> python -c "import torch; print(f'device name \[0\]:', torch.cuda.get\_device\_name(0))" Program: Unknown command line argument 'Files\\Python312\\Lib\\site-packages\_rocm\_sdk\_core\\lib\\llvm\\bin\\offload-arch.exe'. Try: 'C:\\Program --help' device name \[0\]: AMD Radeon RX 9070 XT

C:\\Users\\AMD> python -m torch.utils.collect\_env Program: Unknown command line argument 'Files\\Python312\\Lib\\site-packages\_rocm\_sdk\_core\\lib\\llvm\\bin\\offload-arch.exe'. Try: 'C:\\Program --help' <frozen runpy>:128: RuntimeWarning: 'torch.utils.collect\_env' found in sys.modules after import of package 'torch.utils', but prior to execution of 'torch.utils.collect\_env'; this may result in unpredictable behaviour Collecting environment information... PyTorch version: 2.9.1+rocm7.2.1 Is debug build: False CUDA used to build PyTorch: N/A ROCM used to build PyTorch: 7.2.53211-158bd99533

OS: Microsoft Windows 11 Pro (10.0.26100 64-bit) GCC version: Could not collect Clang version: Could not collect CMake version: Could not collect Libc version: N/A

Python version: 3.12.0 (tags/v3.12.0:0fb18b0, Oct 2 2023, 13:03:39) \[MSC v.1935 64 bit (AMD64)\] (64-bit runtime) Python platform: Windows-11-10.0.26100-SP0 Is CUDA available: True CUDA runtime version: Could not collect CUDA\_MODULE\_LOADING set to: GPU models and configuration: AMD Radeon RX 9070 XT (gfx1201) Nvidia driver version: Could not collect cuDNN version: Could not collect Is XPU available: False HIP runtime version: 7.2.53211 MIOpen runtime version: 3.5.1 Is XNNPACK available: True

CPU: Name: AMD Ryzen 7 7800X3D 8-Core Processor Manufacturer: AuthenticAMD Family: 107 Architecture: 9 ProcessorType: 3 DeviceID: CPU0 CurrentClockSpeed: 4201 MaxClockSpeed: 4201 L2CacheSize: 8192 L2CacheSpeed: None Revision: 24834

Versions of relevant libraries: \[pip3\] numpy==2.4.5 \[pip3\] torch==2.9.1+rocm7.2.1 \[pip3\] torchaudio==2.9.1+rocm7.2.1 \[pip3\] torchvision==0.24.1+rocm7.2.1 \[conda\] Could not

collect

https://redd.it/1th9x3x
@rStableDiffusion

Anyone using LTX Desktop?

Hey Guys I have tried the LTX Desktop and it is really fast. It generated 10 sec video 720p 9:16 in just 2-3 minutes maximum.

I want to know if anyone else is using it, as I want to do some more stuffs with it.

https://redd.it/1thhj2v
@rStableDiffusion

ComfyUI Tutorial: Realistic AI Lip Sync Dubbing with LTX 2.3 LORA Low Vram workflow (6 Gb Vram,16 Gb of Ram)
https://youtu.be/5hmismj1LQc

https://redd.it/1thm9n8
@rStableDiffusion

Review of comfyUI cloud



https://redd.it/1thow1g
@rStableDiffusion

This took me like a Whole Week to Do. Steve got to Catchup Somehow.
https://www.youtube.com/watch?v=9QWClP2i-BU

https://redd.it/1thp2jn
@rStableDiffusion

LTX 2.3 Cinematic Look

https://redd.it/1thsfio
@rStableDiffusion