Hello everyone. I am Dr. Furkan Gözükara. PhD Computer Engineer. SECourses is a dedicated YouTube channel for the following topics : Tech, AI, News, Science, Robotics, Singularity, ComfyUI, SwarmUI, ML, Artificial Intelligence, Humanoid Robots, Wan 2.2, FLUX, Krea, Qwen Image, VLMs, Stable Diffusion
Get more from SECourses: Tutorials, Guides, Resources, Training, FLUX, MidJourney, Voice Clone, TTS, ChatGPT, GPT, LLM, Scripts by Furkan Gözü on Patreon
I tested this and have some feedback. a) cant see/modify system prompt, b) Top K and Temperature dont appear to have any effect. c) nothing is getting output to the terminal (logging, etc) - would like option to see verbose.
When I have a running instance, ThinLinc runs in a loop after entering the id and password. It runs without success. I've done several tests with and without https... but it's the same. Does this happen to you?
I just tested the new JoyCaption v7 and it works wonderfully on my Windows 10 machine now! Awesome work @Dr. Furkan Gözükara ! I guess the earlier versions weren't working because of my limited VRAM. I currently have 3080 10GB, waiting on the 5000 series to ugprade. So thanks for including the startup options to choose which version of it to run
Something I noticed when designating an Output Folder for Batch Processing it works good but then it still puts files in the default output folder as well. It's not really that big of a deal though, just thought I'd mention it
I'm about to install Caption Editor and test it. Oh and since Flux works way better with natural language prompts, maybe in the near future you can make Gradio app of something like this so we can run it locally? With the option to choose whichever LLM we've downloaded https://huggingface.co/spaces/gokaygokay/FLUX-Prompt-Generator
It happened to me yesterday, Did you deploy a RTX A6000 [Spot] by any chance? I ended up having to terminate it since it wouldn't let me login and then deployed a regular A6000 and then it worked like normal
I'm on ThinLinc, but I can't copy and paste, is this normal? I want to paste this: chmod +x Massed_Compute_Kohya_FLUX.sh ./Massed_Compute_Kohya_FLUX.sh at the location of the photo, is this what I'm supposed to do?
i've done a flux lora with "ohwx man" and another with captions generated by JoyCaption. just an anecdote on a small sample size but wanted to share: the subject had glasses (wearing them in every one of the 20 training pics), and with the lora that had the JoyCaptions, fewer output images have him wearing glasses compared to the lora captioned with just "ohwx man". probably because of JoyCaption explicitly captioning that the man is wearing glasses? if i add "wearing glasses" to my prompt the glasses are there as expected, and they match the style from the training images. interested to hear thoughts and if anyone else experienced the same.
also, i did separate loras of two different people, then prompted something like "photo of ohwx man and phwx woman smiling" - applied both loras and turned the strength of both down to 0.8. there was some bleeding of facial features but still produces images really close to the likenesses of the subjects.