Thanks for the tour. Good to know the tools and follow how technology transforms and manifests into different solutions. Not suspecting it was generated output, your photographic face had me fooled. The illustration is a nice test, and can see a bit of softening of the painted texture. Good observation on the dread of commiting to an output that might contain the unexpected, or need to be refined with iterations to meet expectations. But it’s free for now as you say. I’m curious about results on 3-quarter perspectives for more extreme parallax. We will see-Sub’d
Hi, Sorry to hear this didn't work for you. A couple of things worth trying if you haven't already: 1. In ComfyUI Manager do an 'Update ComfyUI' and an 'Update All'. 2. In ComfyUI Manager go to 'Installed Custom Nodes' and filter on installed Nodes. Look for any conflict triangles. If you click on a triangle it will show you what is conflicting, so look for any reference to Bilbox. 3. Rather than use my DropBox Workflow download, try adding the Bilbox Photo Prompt to any Workflow (but not my one) manually by double left clicking the mouse, type bilbo in the search box and select the Photo Prompt Node. If you don't see the Node in the search list, it hasn't installed, so try again to install the 'BilboX's ComfyUI Custom Nodes' via the Install Custom Nodes Option in ComfyUI Manager. I hope this helps and Good Luck.
She came in through the bathroom window, Protected by a silver spoon. But now she sucks her thumb and wonders By the banks of her own lagoon. Didn't anybody tell her? Didn't anybody see? Sunday's on the phone to Monday, Tuesday's on the phone to me. She said she'd always been a dancer, She worked at fifteen clubs a day, And though she thought I knew the answer, Well, I knew what I could not say. And so I quit the p'lice department, And got myself a steady job. And though she tried her best to help me, She could steal, but she could not rob. Didn't anybody tell her? Didn't anybody see? Sunday's on the phone to Monday, Tuesday's on the phone to me, oh yeah
I went to go check this out, then read their Privacy Policy while I was there. Holy crap! I couldn't get away from there fast enough. Not just all the information they collect about you, but that they flat out say they'll use it for third-party marketing and advertising purposes.
They may be reluctant to give you a free preview, because then you could just screen capture it and save it for free. Though the idea would work if it the preview was lower rez, B&W etc.
Hi, sorry to hear it didn't work for you. A couple of things worth trying/thinking about: ComfyUI works best with nVidia GPU/graphics. My GPU is AMD so I have all sorts of problems regular ComfyUI users don't have. The ComfyUI folder should be installed onto your C: drive, not a Naz box or USB drive. Git must be installed first. Have a look at more recent ComfyUI installation videos in case there have been any changes to the installation process since my video. Good luck - I hope you get there!
hi mate, great video. thank you for your explanation. if i have 50 prompts, is it possible for the generated image to be saved one by one, and not waiting for the full 50 batch? because it takes too long to see the image if i have more than 100 prompts to generate.
Hi, thanks for your comments! From my testing I only get the images in one lot at the end of the generation, which is the same if I generated a batch of say 10 images normally. So I think this is just the way ComfyUI works, but I guess there is a risk that if your ComfyUI bombs out due to GPU or memory issues on a large batch, you may lose all the images. As I have AMD GPU, which is not popular with AI programs, I don't generate large numbers of images in one go as it is more likely to bomb out for me. Hope this helps.
Thanks for the tip!! But it is not working. I refreshed the nodes after copying the embedding files, which were in safetensors format, "warning, embedding:negfix does not exist, ignoring" in the console. Do I need to restart?
Hi, sorry to hear this didn't work for you. I believe the file suffix for those embedding was .pt not .safetensors and both Embedding files should be saved into the Models\Embeddings folder. Also make sure there are no type-o errors when you put the Embedding details into your Positive prompt as per my example. Hope this helps.
Hi, sorry about that! Audio is a bit of an ongoing nightmare for me. I have now reverted from a more expensive XLR mic to a basic USB mic and hopefully my latest vids are less bassey. I did try all the usual filters in OBS Studio but I kept sounding like Barry White on my XLR mic! 🙂
Thanks I have photoshop so I do most of my postprocessing there but there were quite a few nuggets in you video, the custom nodes and workflow manager will be useful.
Hi, If you mean the ComfyUI Menu panel on the far right, it does not detach, its all one piece. Mine may look a little different as I have played with various different plug-ins, eg the 'Server' button which I no longer want but haven't worked out how to remove it yet. There is a setting in the ComfyUI Menu that will anchor your ComfyUI menu panel so that it is in the same location on every Workflow that you create. To do that, select the cog wheel icon at the top of the ComfyUI Menu, then scroll down to and select the 'Save Menu Position' option which is third from the bottom of the menu list. Hope this helps.
Hi, On my computer I do not get a progress bar in the Resource Monitor, only a 'time taken' at the end of the task. I only get a progress bar in the KSampler while its doing its thing. That said, what Resource Monitors each of us see may differ slightly, depending on your computer. I have an AMD CPU which ComfyUI doesn't really like, so I do not get a monitor for GPU, while those with nVidia GPU do get a GPU monitor. Hope this helps.
not sure if you figured out the size issue with the reference image. I ran a "upscale image" (can be a "upscale image by") node after the load image. and before the VAE encode "load image" -> "Upscale Image" -> "VAE encode" -> "KSampler" -> ETC that way I changed the reference from a 1920x1920 to 512x512. then the process ran as usual. worked extremely well to change from photo realistic to cartoon style. hope it helps. and also thanks, because this was my first time trying to use img2img and your video was a lifesaver XD
Hi Joe, great video! I'm really impressed with your content. I would love to discuss a potential collaboration with you. Could you please share your contact info or email so we can connect? Thanks!
Hi Joe, great video! I'm really impressed with your content. I would love to discuss a potential collaboration with you. Could you please share your contact info or email so we can connect? Thanks!
random question. I never use clip skip, I've tried it, but it didn't seem to make a difference. I see people still talk about it though. My images look fine. Some great, some not so, but it seem a lot more to do with prompting and sampler settings needing to be adjusted. Do you really think clip skip matters?
Hi Rose, I have only been using CLIP/Clip Skip for the last few weeks, mostly because I couldn't find a default Node for it. I was looking for something named like 'Load CLIP', but eventually found it was called 'Clip Set Last Layer'! What I usually do now is create an image as usual with clip skip set to -1 (or disable the Node). I then keep all the same KSampler/prompt settings but increase Clip Skip to -2, then -3, sometimes up to -5 to see if I get any improved images. I find that although the changes may be subtle, they do quite often (but not always) improve the image, so I think it is worth experimenting. Clip seems (to me) to be something talked about a lot by Python developers, but not so much by end users. I don't think I have seen any other RU-vid videos other than mine where the 'Clip Set Last Layer' Node has even been used, but there are probably some vids out there I am sure. Sometimes I go Skip -24 (on XP models) just for a laugh, as the result can be quite entertaining! :-) Hope this helps, Joe
Thanks for making this workflow for us, I'm using it now and it's great. I feel like I can just use this to generate images for now on and not use Fooocus anymore. I love the presets for the properties, as that gives us a chance to choose from different photography styles and stuff which helps generate the best style of picture. I got a couple of questions, though. What node can we use for LORAs and how can we connect it? Is there a way to generate just one person when I change the height and width to 1024x1024 or beyond? It always comes out generating two different people.
Hi, thanks for your comments, much appreciated. To add a Lora just double left click and type 'Load', you will then get the option to select 'Load Lora'. The Load Lora node will need to sit between the 'Load Checkpoint' and the two Prompt Nodes so you can make the correct connections. For the other issue, I'm no expert but it sounds to me like it may be just a prompt issue. Try adding (for example) '1girl' at the beginning of your positive prompt if you just want one girl (or boy). You can also add emphasis to this by adding brackets - eg ((1girl)). Make sure there is nothing else ambiguous in the positive prompt to make it think you want more than one person in your image. You could also try increasing your CFG a little to make ComfyUI to pay more attention to your prompt. If your still getting say two girls for eg, try adding '2girl' as a negative prompt and see if that helps. It may also be worth changing your Checkpoint model if the problem persists. For photo realistic images I prefer Dreamshaper XL or JuggernautXL. SDXL is not so good for people images I think. Hope this helps.
@@joeconway85 Thank you so much for answering my questions as I really appreciate it. Your suggestions for the two girls appearing it one was fixed in the prompt and fixing CFG levels as well. I agree, 1.5 still works best for people images. I've been using EpicRealism and it's working out pretty great, but Dreamshaper XL and JuggernautXL are pretty good as well. Again thanks for everything, my friend!
Why the Chibbi Simple Resizer? What other Resizer nodes have you tested? I'm always looking to compare and contrast the best nodes. Keen to hear your thoughts. Thank you
Thank you so much. I don't really get where the fox's head thing comes from. I don't want the fox's head per se but I've never scene in my set up. I really liked this and I had been guilty of over egging my cfg and steps when I dialled them down I got much better results. You see other people using much higher numbers and getting good results but that might just be in pony. I saw 75 steps the other day with a nice image!
Hi, the Node icons are called badges in ComfyUI and are useful in identifying who originated the Nodes. Badges can be toggled on or off by going into ComfyUI Manager, then in the left pane, fourth dropdown you will see badges - I selected 'Badge - Nickname' to display the icons. If a Checkpoint model is well documented in Civitai it will include suggested steps and CFG for your images. That's why I use DreamshaperXL a lot, as their page has good recommendations on what to use. If the model doesn't have recommended steps etc I usually look at some of the example images that will be at the bottom of the model page (in Civitai) and try to get a feel for a typical number of steps and CFG as my start point. XL models usually require much less steps and CFG, but not always.
Thanks for the video. I started off using these nodes by myself, but after 5 minutes or so, I knew I had to watch a video, and it would save me so much time. I was right, and your explanation was great. Now I am ready to tackle using them. Thanks Again.
Thanks so much for this tutorial which explained scrolling text so clearly which I could not find in the other videos I watched. Now I will check out your other videos!
Yeah that would have been a nice extra touch. We now have so much choice of colours thanks to this add-on that saving favourites would be a bonus, but now I just make a note of the colour codes I want to use on my Workflow Nodes.
Can you make the cursor larger and make it white when you record? You often say "to this input" or "here", apparently using a cursor, but it's completely invisible to viewers if it's on a dark background.
via Pi CLIP Skip is a technique used in ComfyUI to improve the quality of image generation by incorporating contrastive language-image pre-training (CLIP) into the generation process. This technique helps to better align the generated images with the text prompts provided by users. The "clip clip skip" phrase refers to the specific steps involved in using CLIP Skip with ComfyUI: * CLIP: This step applies the CLIP model to encode both the generated image and the corresponding text prompt into a shared latent space, which helps to ensure that the image aligns with the prompt. * Clip: In this step, the intermediate image generated after the first CLIP step is further refined to better match the text prompt. * Skip: Finally, the refined image from the previous step is "skipped" back into the diffusion process, where it replaces one of the intermediate images generated by the model. This helps to guide the generation process towards producing images that more closely align with the provided text prompt. By using CLIP Skip in ComfyUI, users can expect more accurate and consistent results when generating images based on textual descriptions or prompts.
Hi, thank you for that very detailed description of CLIP and 'Skip'. I do find it strange that there are so few RU-vids on this subject, as it can clearly have an important affect on an image generated in ComfyUI.
@@joeconway85 It was from a FREE application named PI, that is more current on Android than Apple. . via PI Pi AI, also known as "Personal Intelligence," is an AI chatbot developed by Inflection AI, a company founded in early 2022 by Mustafa Suleyman, Reid Hoffman, and Karén Simonyan. Pi AI is designed to be a friendly, empathetic, and engaging conversational AI that users can interact with in a natural and human-like manner. It leverages advanced natural language processing (NLP) techniques to understand user input, generate contextually appropriate responses, and provide assistance on a wide range of topics. Some key features of Pi AI include: * **Personalized Conversations:** Pi AI learns from each user interaction and adapts its responses accordingly, making each conversation unique and personalized. * **Wide Range of Topics:** Pi AI can engage in discussions on a variety of topics, such as entertainment, news, science, history, and personal interests. * **Empathetic and Engaging:** Pi AI aims to provide an emotionally intelligent and engaging conversation experience, recognizing user emotions and responding appropriately. * **Task Assistance:** Pi AI can help users with various tasks, such as setting reminders, providing recommendations, answering questions, and offering suggestions.
be more audible in your next video i did not hear a thing you said in this video i only learnt this through the procedure you listed in the describtion