Dive into the seamless world of GPU resource sharing with Compute Grid! Whether you're looking to rent GPUs for your projects or host your own for others to utilize, our platform is designed to simplify your journey. Subscribe for tutorials, tech insights, and success stories that showcase how Compute Grid empowers both individual developers and companies to scale their computational power as needed. Join us to revolutionize the way GPU resources are shared globally!
Thanks for the video! Are the uploaded files persistent? As well as the generated images? Or are they deleted whenever the instance is restarted? Thank you!
Compute Grid is a full virtualization GPU Marketplace, meaning we ensure that no one has access to anything done in the instance, or the instance itself, but you the renter. Unfortunately to keep ensuring this privacy we do destroy the instance & all the instance data immediately when renters terminate them. I will say though, we are currently in development to a solution to this that would still guarantee 100% privacy for our users, but allow them to still retain & access files & models.
Hello, i was trying to follow your tutorialbut i guess there was an update or somerhin. Impossible to install instantID. Improt fails, any idea what to do? Not able to in my pcnor the online instance
That terminal is my computer's terminal. I use Linux so its readily available to me from my home menu. If you use windows you should be able to access it by going to the menu and searching for "cmd" "Command Prompt" or "PowerShell". For Mac it should be under "terminal" instead, but should still be searchable from your computers main menu.
🎯 Key points for quick navigation: 00:00 *📹 The video compares the 8B and 70B parameter models of Llama3, focusing on their differences and potential use cases.* 00:41 *💡 Parameters in a neural network refer to the number of learnable weights or connections within the architecture, controlling the model's behavior.* 01:10 *⚖️ A model with more parameters will always perform better compared to one with less parameters, given the same training data.* 02:07 *💻 To run the 70B model, a stronger computer with a good GPU and CPU is required, but comput grid provides access to such resources.* 06:38 *📊 The 70B model has a larger manifest size (39 GB) compared to the 8B model (4.7 GB).* 09:06 *📝 The 70B model can generate more creative and engaging responses due to its increased parameter count.* 11:21 *👥 The 70B model can improve a bad resume by rewriting it in a more professional tone and highlighting relevant skills.* 13:00 *💼 The 70B model can also provide better reasoning through family relationships, understanding complex concepts and providing accurate answers.* Made with HARPA AI
I tried to replicate your error but wasn't able to, but it seems to me that the image is causing the error. When it gets taken in be the img2vid node it can't read any of the embeddings or data from the image, so I would verify that the image you are using is a compatible type, like png or jpg. If that doesn't work I would re-install the model & nodes.
Great tutorial, use InstantID but always on the look out for any tips to improve my usage. I used to use Zoe for depth maps but came across a Reddit post analysing the differences between Zoe and MiDaS (tried to link it but got an error, easy to find with a quick Google though), and they found MiDaS performed much better and uses less VRAM. P.S. I've been playing D&D for decades and ranger is my favourite class and loved the reference :)
It's funny but if I didn't know any better I'd swear you were Sarah Sunstone, she also has a very soft melodic voice by the way thank you for this tutorial
Hey there, thank you for your tutorials, they’re very clear and beginner friendly. A small question if that’s ok, instant ID seems and Face ID seem to work kinda poorly with anime characters as far as consistency goes. Any advice on how to tackle that without training a Lora?
We are so glad you are enjoying the videos! For your question, I need a little more clarity. May I ask are you trying to input a human reference photo and make it in anime style with poses & prompt applied? Or is it an anime reference photo and that you are then trying to generate poses & prompts?
@@compute-grid the latter. I've created an original anime character (I have 30 images now) and Im trying to create multiple poses, expressions, and clothing setups. any ideas for how to approach it? also thank you so much for answering!!!
WOW! 30 images are impressive. Unfortunately though, I don''t believe InstantID will be the best model for you to use as it was trained on human face proportions which is very different compared to Anime faces (Eye size, nose position, etc). It could still work, if the face keypoints line up where they need to be, but you would probably need to find a different model trained on anime faces. If I come across one I will be sure to come back here & let you know.
Love your simple way of explaining things. I need to use SD 1.5 at times. How can I do that with ComputegridAI? Is there an easy install method? Oh and how do we add custom models, loras, and controlnets from Civitai or Huggingface?
Hello. We are really glad you enjoyed our video! We are currently in the works to create a tutorial for installing custom models that can't be found in the pre-installed Manager, but for the time being our video covering Logos, "Create Stunning AI Logos with Compute Grid, ComfyUI, and Stable Diffusion | Step-by-Step Guide" breaks down how to install a model from CivitAI at the 3:00 minute timestamp. As for an easy way to downgrade SDXL to SD1.5, we weren't able to find one since SDXL is pre-installed on all of the instances, you would have to SSH in and install in through the command line. Some SD1.5 models do still work with SDXL, so it may still be worth giving it a try! Thank you again for the comment!
@@compute-grid Will check out the video you mentioned, thanks. I want to create interpolation videos. Do you know if ipiv's Morph - img2vid AnimateDiff LCM / Hyper-SD will work with SDXL models?
There is a video up now documenting the file transfer process from instance to your personal machine and vice versa. (How To Transfer Files Within a Compute Grid Instance: ru-vid.com/video/%D0%B2%D0%B8%D0%B4%D0%B5%D0%BE-93eOGpXWDus.html) Also thank you very much for bringing this to our attention. We greatly appreciate the feedback as we try to make our platform better & better!
thank you for this video i would like to see more on comic book art the extra control you mentioned ,merging the character with other backgrounds maybe some openpose stuff would be great
We are glad you enjoyed the video! Also, we are currently in the works for a more in depth InstantID video. If you click the bell icon for our account, it will notify you when new videos come out!
Do you mean how to use other models in the workflow? We are using the SDXL base model, but if you want to install and use others, there is a Manager node that is unfortunately not pictured in the video, because we didn't make it standard for all instances of ComfyUI to have it until after this video was released. However, our other videos do showcase using the Manager, but it is still fairly easy to use. You can use it to install custom models & nodes and upload them directly into your workflow. It should be on your right hand side, on the main menu if you are using a Compute Grid Instance. If you want to use a custom model that isn't available on the manager though, you would then have to use the command line to tell SDXL & ComfyUI where it is located on your machine and to use it.
It's kinda unfair how great your videos are turning out, while I have the same settings and just get those strange messy videos with stuff blurring out all the time, people diffusing like I was using animatediff with a shatter lora 😂
I am sorry to hear that. Different images and subjects will usually need some tweaking with the settings. I recommend just make sure your CFG is relatively low, maybe 2.5. I used the default sampler "Euler", but perhaps try "dpmpp_3m_sde" or any other samplers to help. There are also some controlnet models that can be beneficial such as the "SDXL-controlnet: OpenPose (v2)" to detect poses for any complicated images you are using. If you want us to do a second video with more complex images and fine tuning let us know!
@@compute-grid oh, so you could use sdxl controllers with svd? 🤔 that sounds like a great idea! Well anything that could help me get better results with my rubbish 8gb vram is welcome 😂🙏🏼
