PhD. Computer Engineer. Produces Content For Stable Diffusion, SDXL, LoRA Training, DreamBooth Training, Deep Fake, Voice Cloning, Text To Speech, Text To Image, Text To Video, Generative AI, LLMs
Full precision uses lesser VRAM than mixed BF16 or FP16 precision Very weird SDXL DreamBooth full fine tuning 1024x1024 Adafactor Although speed difference is significant Mixed precision : 1.22 second / it Full precision : 1.82 second / it
PhD. Computer Engineer. Produces Content For Stable Diffusion, SDXL, LoRA Training, DreamBooth Training, Deep Fake, Voice Cloning, Text To Speech, Text To Image, Text To Video, Generative AI, LLMs
I hate my motherboard so much. Constantly ram instability error even though I am already down clocking RAM Shame on you GIGABYTE. If a motherboard can't handle 4 sticks of RAM why do you sell it as 4 sticks! I have been trying to make it work for 3 hours now! Z690 UD DDR4 increased voltages a lot too
PhD. Computer Engineer. Produces Content For Stable Diffusion, SDXL, LoRA Training, DreamBooth Training, Deep Fake, Voice Cloning, Text To Speech, Text To Image, Text To Video, Generative AI, LLMs
Halk Bankası (Halkbank), sosyal medya içerik üreticisi olarak (otomatik %15 vergi kesintisi) IBAN hesabıma Stripe Express ödemesi gelmedi. Bu durumu bankanıza ilettim ve cevabınız olumsuz oldu. Hiçbir açıklama da yazmamışsınız. Yani illa gidip başka bir bankada mı hesap açalım? Bu vurdumduymazlığı anlayamıyorum.
PhD. Computer Engineer. Produces Content For Stable Diffusion, SDXL, LoRA Training, DreamBooth Training, Deep Fake, Voice Cloning, Text To Speech, Text To Image, Text To Video, Generative AI, LLMs
𝗠𝗶𝗻𝗶𝗺𝗮𝘅 looks like new SOTA. And sadly there is nothing, not even close in Open Source :(
Creative AI Coach & Director • Become an indispensable AI expert in just one week • Clients include NBCUniversal, Disney, BBC, ITV, Sky and more
Now that all my favourite AI video generators can do image-to-video, I thought I'd try a quick comparison test using the same prompt and image input. Prompt: 𝗔 𝗰𝘂𝘁𝗲 𝗳𝗹𝘂𝗳𝗳𝘆 𝗺𝗼𝗻𝘀𝘁𝗲𝗿 𝘄𝗮𝘃𝗲𝘀 𝘁𝗼 𝗰𝗮𝗺𝗲𝗿𝗮, 𝘀𝗺𝗶𝗹𝗲𝘀 𝗮𝗻𝗱 𝘁𝗵𝗲𝗻 𝘄𝗮𝗹𝗸𝘀 𝗼𝗳𝗳 𝘁𝗼 𝗿𝗶𝗴𝗵𝘁 𝗮𝗻𝗱 𝗲𝘅𝗶𝘁𝘀 𝗳𝗿𝗮𝗺𝗲. Tests like this only tell you so much. Results will be different with every re-roll, but I tried to give each model at least a few chances to show what it could do. Here's what I learnt... 🔹 𝗠𝗶𝗻𝗶𝗺𝗮𝘅 - Got closest to the prompt and did a great job with the animation and character consistency. 🔹 𝗞𝗹𝗶𝗻𝗴 - Very expressive. Lots of dynamic movement but the character changed somewhat from the initial frame. 🔹 𝗗𝗿𝗲𝗮𝗺 𝗠𝗮𝗰𝗵𝗶𝗻𝗲 - Performed the worst in this particular test with some very odd moves and warping/deflation! 🔹 𝗥𝘂𝗻𝘄𝗮𝘆 (Gen-3) - Great character consistency and image quality with a nice wave and walk cycle. Not great prompt adherence though. It’s missing the smile and walk off at the end. I've linked each of these in the comments so you can try them out for yourself. For now at least I'll be keeping all 4 in my AI tool-box. Although Luma's Dream Machine performed the worst, it still has its uses and is particularly good for transitions where the first and last frames are totally different images. Its very quick at generating now too - almost as fast as Runway's Turbo mode but without the pause you often get when using Gen-3. Kling can create some real money-shots on a single roll, but you have to wait for them. I'm fairly new to Minimax, but will definitely be using it more now it accepts image inputs. If I had to choose one right now it would be Hailuoai's Minimax. Not only is it extremely capable but it's also currently free! 👉 Want to take a deeper dive into generative AI? Check out the live, online course I've been running for almost a year now. See link in the comments. 👀 #minimax #kling #lumadreammachine #runwaygen3
PhD. Computer Engineer. Produces Content For Stable Diffusion, SDXL, LoRA Training, DreamBooth Training, Deep Fake, Voice Cloning, Text To Speech, Text To Image, Text To Video, Generative AI, LLMs
Discord app has been completely banned in Türkiye yesterday. Dear Discord team are you working on the issue to solve it ASAP? I am a yearly paid Discord Nitro subscriber and I have a professional Discord server with over 8000 members. I need Discord for my business and professional life. Moreover I use it communicate with my University Students. Now I have to use a VPN to connect my Discord server to help my subscribers. I am bringing USD to Türkiye from through Discord as well. How is this exactly helping the economy? Moreover, now I have to connect internet through a VPN. Now my question to everyone is that do you know any reliable VPN that allows me to run only Discord APP through VPN? I tried Warp VPN but it is only exclusive split tunneling based not inclusive. My discord server : https://lnkd.in/dmcNTPrg It is for fully education, programming and AI Also please share this for awareness because Discord is an extremely useful tool that can be used for good purposes.
PhD. Computer Engineer. Produces Content For Stable Diffusion, SDXL, LoRA Training, DreamBooth Training, Deep Fake, Voice Cloning, Text To Speech, Text To Image, Text To Video, Generative AI, LLMs
AI is going to bring massive, massive unemployment in every profession until all workforce is done by AI and humans don't need to work. So best to learn AI, fully utilize it to the max to leverage yourself. Of course Allah knows best but this is my future prediction.
PhD. Computer Engineer. Produces Content For Stable Diffusion, SDXL, LoRA Training, DreamBooth Training, Deep Fake, Voice Cloning, Text To Speech, Text To Image, Text To Video, Generative AI, LLMs
Doing 4 new trainings for FLUX Fine Tuning 2 datasets , 15 images and 256 images 2 separate trainings for each one , Batch Size 1 vs Batch size 7 - both has individually researched Learning Rate Time to finalize current best configs hopefully :)
PhD. Computer Engineer. Produces Content For Stable Diffusion, SDXL, LoRA Training, DreamBooth Training, Deep Fake, Voice Cloning, Text To Speech, Text To Image, Text To Video, Generative AI, LLMs
I still don't understand why LinkedIn still parses all links like this. This is literally a security vulnerability that we can't see links! LinkedIn
PhD. Computer Engineer. Produces Content For Stable Diffusion, SDXL, LoRA Training, DreamBooth Training, Deep Fake, Voice Cloning, Text To Speech, Text To Image, Text To Video, Generative AI, LLMs
Huge news for Kohya GUI - Now you can fully Fine Tune / DreamBooth FLUX Dev with as low as 6 GB GPUs without any quality loss compared to 48 GB GPUs - Moreover, Fine Tuning yields better results than any LoRA training could
Director of Operations and Tech @ TNV Method | ex - AI Infrastructure and Software Engineer in Amazon.
1yThis is dreambooth training on SDXL and just not lora?