- 11
- 72 066
Benji’s AI Playground
Hong Kong
Приєднався 6 січ 2025
Welcome to Benji’s AI Playground! I’m Benji, a tech entrepreneur who like to create content as my hobby tube about AI and tech here. This is your go-to space for exploring the world of Open Source AI and cutting-edge technology. From tutorials and deep dives into AI frameworks to discussions on the latest tech trends, I’m here to make complex concepts simple and accessible. Let’s build, learn, and innovate together in the open-source community!
Lumina Image 2.0 In ComfyUI Native - Small Size With Big Power Image Diffusion Model
Lumina Image 2.0 In ComfyUI Native - Small Size With Big Power Image Diffusion Model
In this video, we explore Lumina Image 2.0 , the latest diffusion-based AI image generation model now natively integrated into ComfyUI. This powerful yet lightweight tool allows creators to generate high-quality images using simple text-to-image workflows, without the need for complex node connections or heavy computing resources. Whether you're designing anime-style characters, hyper-realistic portraits, or cyberpunk landscapes, Lumina Image 2.0 delivers versatile results with minimal effort. We’ll walk through how to set up and use this model, including tips for crafting structured prompts to maximize output quality. From hands-on demonstrations to practical use cases, this video is your guide to mastering Lumina Image 2.0.
Who Is This Content Suitable For?
This content is ideal for digital artists, graphic designers, and AI enthusiasts who want to experiment with cutting-edge image generation tools. It’s also suited for beginners exploring AI art creation and professionals seeking efficient workflows for commercial projects. If you’re passionate about creating diverse visual styles-whether for social media, marketing campaigns, or personal projects-this tutorial will provide valuable insights and actionable techniques.
Why Does It Matter?
Lumina Image 2.0 stands out because of its compact size (just 2 billion parameters) and ability to run smoothly on consumer-grade PCs, making advanced AI image generation accessible to a wider audience. Unlike larger models like Flux, which require significant computational power, Lumina strikes a balance between performance and accessibility. Its versatility in generating various styles-from anime to hyper-realism-makes it a game-changer for creatives looking to produce professional-grade visuals without investing in expensive hardware. By mastering Lumina Image 2.0, you can streamline your creative process while achieving impressive results.
Resources:
ComfyUI Example for Lumina Image 2.0
comfyanonymous.github.io/ComfyUI_examples/lumina2/
Lumina Image 2.0 AI Model
huggingface.co/Comfy-Org/Lumina_Image_2.0_Repackaged/blob/main/all_in_one/lumina_2.safetensors
If You Like tutorial like this, You Can Support Our Work In Patreon:
www.patreon.com/c/aifuturetech
Discord : discord.com/invite/BTXWX4vVTS
#comfyui #LuminaImage2.0 #diffusionmodels #aiimages
In this video, we explore Lumina Image 2.0 , the latest diffusion-based AI image generation model now natively integrated into ComfyUI. This powerful yet lightweight tool allows creators to generate high-quality images using simple text-to-image workflows, without the need for complex node connections or heavy computing resources. Whether you're designing anime-style characters, hyper-realistic portraits, or cyberpunk landscapes, Lumina Image 2.0 delivers versatile results with minimal effort. We’ll walk through how to set up and use this model, including tips for crafting structured prompts to maximize output quality. From hands-on demonstrations to practical use cases, this video is your guide to mastering Lumina Image 2.0.
Who Is This Content Suitable For?
This content is ideal for digital artists, graphic designers, and AI enthusiasts who want to experiment with cutting-edge image generation tools. It’s also suited for beginners exploring AI art creation and professionals seeking efficient workflows for commercial projects. If you’re passionate about creating diverse visual styles-whether for social media, marketing campaigns, or personal projects-this tutorial will provide valuable insights and actionable techniques.
Why Does It Matter?
Lumina Image 2.0 stands out because of its compact size (just 2 billion parameters) and ability to run smoothly on consumer-grade PCs, making advanced AI image generation accessible to a wider audience. Unlike larger models like Flux, which require significant computational power, Lumina strikes a balance between performance and accessibility. Its versatility in generating various styles-from anime to hyper-realism-makes it a game-changer for creatives looking to produce professional-grade visuals without investing in expensive hardware. By mastering Lumina Image 2.0, you can streamline your creative process while achieving impressive results.
Resources:
ComfyUI Example for Lumina Image 2.0
comfyanonymous.github.io/ComfyUI_examples/lumina2/
Lumina Image 2.0 AI Model
huggingface.co/Comfy-Org/Lumina_Image_2.0_Repackaged/blob/main/all_in_one/lumina_2.safetensors
If You Like tutorial like this, You Can Support Our Work In Patreon:
www.patreon.com/c/aifuturetech
Discord : discord.com/invite/BTXWX4vVTS
#comfyui #LuminaImage2.0 #diffusionmodels #aiimages
Переглядів: 2 699
Відео
Hunyuan Video In ComfyUI With MultiLora For Txt2Vid and V2V
Переглядів 3,2 тис.9 годин тому
In this video, we dive into the exciting world of Hanyuan Video , a powerful AI tool that allows you to transform ordinary video clips into extraordinary creations using multiple LoRA models. Whether you're turning a scene from John Wick 4 into a Star Wars-inspired laser gunfight or animating dance movements with custom character styles, Hanyuan Video makes it possible to reskin videos and gene...
ComfyUI Janus Pro - Integrate With AI Image And Video Workflow - Tutorial Guide
Переглядів 13 тис.21 годину тому
ComfyUI DeepSeek AI Janus Pro - Tutorial Guide Integrate With AI Image And Video Workflow Unlock the full potential of DeepSeek AI's Janus Pro in ComfyUI with this comprehensive tutorial! In this video, we’ll guide you step-by-step on how to install and use the Janus Pro custom node for both image generation and image understanding tasks. Whether you’re a beginner or an advanced user, you'll le...
DeepSeek AI Janus - Multi-Model AI With Vision And Image Generation
Переглядів 3,6 тис.День тому
DeepSeek AI Janus - Multi Model AI With Vision And Image Generation In this video, we dive into DeepSeek AI, a leading Chinese AI company making waves with its advanced multi-model AI systems like Janus. Learn how DeepSeek's models, including the Janus Pro 7B, are pushing the boundaries of AI by combining text, image, and video understanding. From generating detailed image descriptions to creat...
ComfyUI PuLID Flux ll - New Better AI Image Character Face - Bye To ReActor
Переглядів 11 тис.14 днів тому
ComfyUI PuLID Flux ll - New Better AI Image Character Face - Bye To ReActor (Tutorial Guide) Discover PuLID Flux 2, the ultimate tool for face identification and character transformation in AI-generated images! In this video, we explore how to use PuLID Flux to seamlessly integrate faces into new images, creating natural-looking results without the awkward edges or pixelation of traditional fac...
Kokoro TTS in ComfyUI - A Lightweight Text To Speech AI Model Running Locally
Переглядів 4,3 тис.14 днів тому
Kokoro TTS in ComfyUI - A Lightweight Text To Speech AI Model Running Locally Unlock the power of Kokoro TTS with ComfyUI custom nodes! In this video, we dive into how to set up and use the Kokoro Text-to-Speech (TTS) framework within ComfyUI to generate high-quality AI voiceovers. Whether you're creating educational content, entertainment, or integrating TTS with AI video workflows, this tutor...
Nvidia Cosmos In ComfyUI - AI Diffusion Model For Video Generation - Setup Tutorial
Переглядів 14 тис.21 день тому
Nvidia Cosmos In ComfyUI - AI Diffusion Model For Video Generation - Setup Tutorial In this video, we dive into Nvidia Cosmo, the groundbreaking Text-to-World and Video-to-World diffusion models announced at CES 2025. Discover how to run Nvidia Cosmo in ComfyUI for Text-to-Video, Image-to-Video, and Video-to-Video workflows, and explore the future of AI video generation. Updated from Comfy.org ...
Hunyuan Video Video-to-Video In ComfyUI - With Flow Edit And Native Node Easily!
Переглядів 10 тис.21 день тому
Hunyuan Video Video-to-Video In ComfyUI With Flow Edit Discover the power of Hunyuan Video with the Video-to-Video method using Flow Edit in ComfyUI! In this tutorial, we dive deep into how to transform video motions and styles seamlessly, creating stunning AI-generated videos with ease. Whether you're a beginner or an advanced user, this guide will show you how to leverage Flow Edit for effici...
Hunyuan Video GGUF In ComfyUI - Low VRam Optimization For AI Video Generation
Переглядів 4,4 тис.28 днів тому
Hunyuan Video GGUF In ComfyUI - Low VRam Optimization For AI Video Generation In this video, we explore a faster method for AI video generation using Hunyuan Video's GGUF quantization models. Perfect for running locally on lower VRAM GPUs! Learn how to optimize workflows, use face swaps, and upscale videos with ComfyUI. Plus, check out how MM Audios adds sound effects for a complete AI video ex...
HuggingFace Smolagents Open Source AI Agent Framework Full Setup Tutorial Guide
Переглядів 1,6 тис.Місяць тому
HuggingFace Smolagents Open Source AI Agent Framework Full Setup Tutorial Guide Unlock the power of AI agents for content creation! In this comprehensive guide, learn how to automate your creative projects using Hugging Face's Smolagents AI framework. Discover the step-by-step process of integrating AI agents with language models like LLaMA to streamline tasks such as writing movie scripts, gen...
Hunyuan Video Lora In ComfyUI - Generate AI Video With Specific Character Style
Переглядів 6 тис.Місяць тому
Hunyuan Video Lora In ComfyUI - Generate AI Video With Specific Character Style Unlock the full potential of Hunyuan Video by integrating Lora Models without installing any additional custom nodes in ComfyUI! In this tutorial, we show you how to seamlessly incorporate Lora models into your Hunyuan Video workflows, enhancing your AI-generated videos with custom styles and characters. More Inform...
Thanks, I had no idea how to add a lora with ComfyUI native support
Good but, pretty useless, flux is still my number 1 image generation as of now...
hehe.. great, then keep using with your image XD
10gb.......um nah......
thanks for your work. I´m not convinced of this model - its fast but producing pics in an "anime style" - Flux works much better for me !
What is the point in the model? She is no better than XL and does not know how to do what Flux or SD3.5 does. We need a model who can generate Chinese, Japanese, Russian and Arabic text.
Looks good ….will It work with LORAs?
The Lunmina older version have Lora , so i think yes. Just wait for it, there shoupd be some lora trainer for it.
@@BenjisAIPlayground Am trying with power lora loader. CKPT ---PLL---MSAF ---Sampler. also connecting between checkpoint ---PLL---Clip---. I changed the supplied prompt but still wants to create anime cartoon. ** trying FLUX LORA FLUX_REALISM_SDXL lora
@@RDUBTutorial oh cool, it this work with Flux lora?
@@BenjisAIPlayground don’t think so …can’t get an image that shows Lora influence yet. Not sure where to put the Lora yet so trying bunch of combinations.
*Saaar, with full respect and as a proud Indian gentleman of culture and class, let me tell you-many Western women whisper the same about us: "Small size with BIG" just like you say about this model, Saaar. It’s not about the size, it’s about the power, the presence, the ancient Vedic energy we carry!*
I remembered when SD3 released and the same time this model version 1. People were trying to moving from SD to this one in Open source community.
Yes, they said Lumina, Hunyuan AI ,Pixart , trying to build nodes and ecosystem around it. Then Flux appeared
Resources: ComfyUI Example for Lumina Image 2.0 comfyanonymous.github.io/ComfyUI_examples/lumina2/ Lumina Image 2.0 AI Model huggingface.co/Comfy-Org/Lumina_Image_2.0_Repackaged/blob/main/all_in_one/lumina_2.safetensors
how do i get inside the (comfy) environment when you're first at the command prompt.... i see a .venv subfolder should i run scripts\activate.bat or something first? update: i think i got it C:\ComfyUI> c:\comfyui\.venv\scripts\activate.bat (ComfyUI) C:\ComfyUI>
I use Conda virtual environment. You can check it out.
ON MAC_ keep getting stuck at sampler: SamplerCustomAdvanced Trying to convert Float8_e4m3fn to the MPS backend but it does not have support for that dtype ANYONE with some great ...helpful ideas?
Nice sails pitch, my bet is in no more than 3 years, Microsoft or NVidia will bundle this stuff for a couple extra bucks per month for their users, and since AI training happens with every single character that is added to the internet, it may happen even faster. Think of the history of the mega corps doing this with a broadcast tv signal when the tech was switching from those that had the money to build a high-end antenna system to watch broadcast television on their tiny black and tv's in the late 60's to early 70's for free or could subscribed to one or two channels from your local antenna Co, that rebroadcasted a signal that you could not receive good enough and wanted a perfect picture. Then Broadcast corporations created a business MODEL to run a wire called co-axial cable that was installed right to the back of your TV from the local head end. The path between them is filled by many active, passive devices and human Maintenace crews to ensure a clean signal to the end customer all for a monthly bill. One of the strongest business models is called lek billing, you will have to check the spelling, but it is a reoccurring bill that is a fee for service. The system was fed by huge satellite receiving dishes to the head end and or point of distribution in your local city directly fed from the Broadcast Network station via very powerfully satellite transmission dishes that can even transmit 120v 60Hz signals along with radio frequency waves to satellites that were installed 22,500 miles above the earth placed in geosynchronous orbits at strategic places above the earth. All designed and installed in a few short years and by the mid to late 70's installed to many parts of the Globe. we will more than likely see a transition that wipes out the small business models by the mega corps again with Ai content generation web site that charge money to use AI apps and or hardware. It will happen so fast the average person that is running these businesses will never see it until their model is gone. Just like when all these ma and pa dial up internet Co's were wiped out in one swoop when the Cable Co's engineers said wait just a sec, we already have a signal to everyone's home that has a very weak return path signal, we just have to change out every active with amps that have return path amplifiers to generate the high quality return path frequencies needed for a stronger data connection and add a few fiber nodes (that the industry calls Star gates), and we can charge even more money to transmit data to and from the subscribers/ (SUBS) home. 25 years ago, internet junkies did not care that they had to sign a contract that locked them to the cable Co for only two years for a data package that cost 100 bucks per month and that was on top of the signal for TV that was all on that same Coaxial wire. and just like that every single dial up internet provider was pretty much gone. when the Ai market is said to be a 500 trillion dollar industry looking forward, you can guaranty that Software Operating system Co's like Microsoft and hardware Co's like Nvidia are creating and or have already created businesses models that will be taking a huge piece of that pie and crushing those small to large AI generating Co's. Same old news that happens with any form of tech.
Emma Watson I believe 🙂
can we use multiple lora in ltxv
I don’t think ltx supports Lora yet… it why I keep coming back to trying to make HyV to work better …on my system ltx is way faster than hyv
@@RDUBTutorial yes, LTX smaller , faster and lighter weight
So Hunyauan is the best right now ? Better than VLX Studio ?
preformence wise in open source local, yes.
Can i use this workflow in 8 gb 4060 with 16gb ram ????????
yes, with limited VRam it's better to use lower sampling and Hunyuan Fast.
I didnt tried MultiLora, but PowerLoraLoader lets you keep your loras in one node and seems to work well with hunyuan, still have to compare them
Let us know how it preform 👍
Great Video, as always! That multilora by fok saves my pixels. Are you planning on making a video about recent img2vid Lora by LeapFusion workflow?
I think I will wait for the img2vid models weights.
Custom Nodes Mentioned: Flow Edit: github.com/logtd/ComfyUI-HunyuanLoom Hunyuan Video MultiLora: github.com/facok/ComfyUI-HunyuanVideoMultiLora TeaCache node: github.com/lldacing/ComfyUI_Patches_ll Workflow And Research For Freebie: www.patreon.com/posts/hunyuan-video-in-121487379? Additional Research and Content For Patreon Supporters: www.patreon.com/posts/121487753?
i need a videocard bro,
While this version of PuLID workflow is improved, it is still not nearly as good as Reactor/Insight-Face when it comes to face swaps of people you know. It's 'ok' for other people since it just does a resemblance, versus a true representation like ReActor does... too bad...
It's not working for men's face swapping, I am getting for any woman, it's 80% close, but for men its way off
ok.. so kinda embarasing but im stuck on the very first thing you do.. "in ComfyUI you search for Janus" Where tf do i search for Janus in ComfyUI?!
Use Comfy UI Manager. U need to install it
This model is not currently available via any of the supported third-party Inference Providers
Is comfy ui free ?
Nice tutorial Bro, can we run this on mimic pc
Yes you can
i was under the impression this had a branch already that doesnt need gpu...
Is there also new pulid version for sdxl?
Is it possible to use this to understand an aerial photo with buildings and roads, in conjunction with img to img at low noise reduction, to outline the buildings and label roads?
Does it better than flux.1 dev? Don't want to waste my time and space for another me too model
ComfyUI-Manager is supposed to make "pip install requirements" already, you don't need to do it manually... What the asian github is saying probably is "either use manager or install manually" and you have done both ;) Manual "pip install" are required sometimes in addition to the Manager because of some specific versions that are needed... you would have seen specific "pip install package@version" commands and not a generic "requirement.txt" what manager already use...
Don't think so, if you really did install some custom nodes. Some requirements.txt do not go through from auto restart.
better than flux?
No
When you talk video to video, can you output the last frame on the first video to automate content in a better way?
Yes image from batch node
How long can the video be? Any time limit? Big problem create automated content is the length of the video in all apps…
Hi. Thank you for this video. You are a good teacher. You are going step by step in order to help us download an important platform . My question is : Is it possible for Windows 10 ,8 Ram,AMM CPU ?Thank you in advance.
J-Anus😂... Liked and subscribed
Great Tutorial.need quantized gguf models of this asap!
soon it will be
U know any way to stop the model from hogging all the vram? Using this for video prompts doubles the vram use while rendering... :(
Purge VRAM after some node have done its job
where is the link for the cosmosimagetovideolatent node? it doesnt show in the missing custom nodes :(
Looks like you are skipping. Its a native node in Comfy.
@@BenjisAIPlayground not sure I understand sorry, how / where do I get it?
"created" aka illegally distilled?
They said, it Open source bitch , see if you can do it 😂
@BenjisAIPlayground Nobody can do it legally. It's another example of CCP "innovation" aka IPTHEFT .....deliberately timed with Stargate as "Unrestricted Warfare".
@BenjisAIPlayground 😄👍
Stop crying Muricans.
TogetherVision is better
will this able to use for video vision to text?
Great tutorial 😊 thanks
:) first after the jerk above me :P
hahhaa thanks
ComfyUI Workflow Created In This Tutorial (Freebie): www.patreon.com/posts/comfyui-janus-ai-121111621? For Patreon Supporters, the V2V workflows updates with Janus Pro www.patreon.com/posts/121112042
👋 hi 👋
좋은 정보 감사합니다 :)😊
Good info bro.. really looking forward to using this. Is there a safetensor format dl somewhere, those .bin files from a china company is not what i want.
@@aivideos322 stay tune another video how to use in Comfy 😉 If not bin file , theres none currently. Or don't use it. LOL
Does it support file uploads? For example, can I upload a PDF file and have the AI explain to me what that PDF is about? I want to have discussions with the AI to help me understand the content of the PDF better. When discussing documents, I’d prefer to communicate with the AI both through text and through spoken words like in a normal conversation.
For that case, you have to use Deepseek v3, or R1 , running locally using Open Web UI , you can attach documents in chat
Can it solve ‘Math & Physics’ problems, and show and answers step by step?
you can try it, take a picture of math question and ask it to tell you step by step.
The vision model will be useful for image to image on comfy ui