everything is faceswaping
everything is faceswaping
CDid anyone run into this issue where if you have solid background (mainly gray / white / blue) the part of the image the runs through the k-sampler has a weird color shift. It only happens with solid background. I assume it's somewhere in a rgb/srgb color shift?
�all sdxl character loras i've trained on civitai end up with extra limbs and many deformities, will a onetrainer sdxl fine-tune into a lora solve this issue? i guess it's best to finetune on sdxl base? or does other realistic checkpoints work as well. i'd like the lora to be flexible and to be used with multiple checkpoints, so then it's best to use sdxl base, yes? @Dr. Furkan Gözükara
Fnope not that i know
Fi dont think there is a way that will do all of you want
Fsdxl is just weak
Fare you inpainting?
�ok, thanks doc. I've done some flux finetunes for character. problem with flux fine tunes is to carry over skin textures properly. and it also feels like the face inpainting in swarmui makes skin more plastic. feels like the skin textures in my training images were not learned properly. would close-ups be a solution, and should face always be visible on all photos, or could you use a photo of only stomach i.e. with detailed skin etc?
Fyou can do tiled training
Ffinally added cancel feature this was hard
F
Fonly left thing is preset save and load hopefully
�tiled training! yes I read about it, any guide somewhere on how to actually do it? or maybe you wrote a guide already, Dr?
Four app can generate tiled images
Fdo that you will undersand logic
FPatreon
Get more from SECourses: FLUX, Tutorials, Guides, Resources, Training, Scripts on Patreon
A@Dr. Furkan Gözükara is a 5060TI 16gb good for ai?
Sholy moly, @Dr. Furkan Gözükara have you heard about LoRA-Edit yet?
S
Smaybe an interesting alternative to VACE
SYouTubechenjian gao
Project Page: https://cjeen.github.io/LoraEditPaper
Code: https://github.com/cjeen/LoRAEdit
Paper: https://arxiv.org/abs/2506.10082
Video editing using diffusion models has achieved remarkable results in generating high-quality edits for videos. However, current methods often rely on large-scale pretraining, limiting flexibility for specific ed...
Code: https://github.com/cjeen/LoRAEdit
Paper: https://arxiv.org/abs/2506.10082
Video editing using diffusion models has achieved remarkable results in generating high-quality edits for videos. However, current methods often rely on large-scale pretraining, limiting flexibility for specific ed...
Fyep
Fnot yet
Flooks nice
Sits pretty new, especially its performance versus VACE is interesting, I'm sure you can optimize it ^^
Ffinally stage is done too now only i need to make presets
J@Dr. Furkan Gözükara quick question, i'm training a model in flux and i wanna use 24 images instead of 12 images for training, what settings do i need to change in epochs / iterations? i remember u mentioned it in one of the tutorials but having trouble finding that part of your video again
Fyou can do more epochs
F12 is a little bit more
Jgot it thank you!
�@Dr. Furkan Gözükara does any of your apps extract first the first frame from a video to save as png? best would be to run through batch of videos in a folder, then save first frame of each video in a separate folder as png
Fframepack saves frames as png
Fwell for that you can quickly write a python code
Ftakes folder
Fand save png in a folder
Fi recommend like any llm like chatgpt gemini etc
�Tried to make a short ffmpeg python app. Made a venv, cant install ffmpeg successfully. Working with chatgpt to solve it, no luck so far
Flet me give you a code
Fa.py3.81KB
Fpython a.py "C:\temp" "C:\temp2"
�Wow thanks doc! Excited to try
Fi tested works. if you had followed our requirements video you must have ffmpeg and python already
�Yes i followed it on my other computer. Ill try your script there
Ffirst preset is ready
Fbetter naming
Shi, what is this video frame interpolation type for?
AI am running the forgev9 installer and love how smooth and fast it is. I get this error message whenever I use a face model from Reactor: RuntimeError: D:\a_work\1\s\onnxruntime\python\onnxruntime_pybind_state.cc:857 onnxruntime::python::CreateExecutionProviderInstance CUDA_PATH is set but CUDA wasnt able to be loaded. Please install the correct version of CUDA andcuDNN as mentioned in the GPU requirements page (https://onnxruntime.ai/docs/execution-providers/CUDA-ExecutionProvider.html#requirements), make sure they're in the PATH, and that your GPU is supported.
I have been able to use reactor in the past with older versions of forgeUI and it currently works inside ComfyUI. I have a RTX 5070 ti 16gb as my primary GPU. I have manually updated cuda, cudnn, onnyxruntime and onnyxruntime-gpu inside the venv folder and have also updated the PATH variables. What am I missing?
I have been able to use reactor in the past with older versions of forgeUI and it currently works inside ComfyUI. I have a RTX 5070 ti 16gb as my primary GPU. I have manually updated cuda, cudnn, onnyxruntime and onnyxruntime-gpu inside the venv folder and have also updated the PATH variables. What am I missing?
Fno idea but RIFE is best as far as i know
Fyes it is extremely cudnn depended
Fi recommend download cudnn
Fand add to both system and user path