Say Goodbye To VRAM Limitations With this FLUX Workflow, Auto Prompts & 1Step Upscale Magic!

[@TheSneakyRobot]

Links For The Video workflow: drive.google.com/file/d/1DcSDEKml4G_BZDwPFkV6jaTEiw2em1uQ/view?usp=sharing comfyui: github.com/comfyanonymous/ComfyUI comfyui manager: github.com/ltdrdata/ComfyUI-Manager Manager install: python_embeded\python.exe -m pip install -r ComfyUI\custom_nodes\ComfyUI-Manager equirements.txt FLUX.1-schnell-gguf: huggingface.co/city96/FLUX.1-schnell-gguf/tree/main FLUX.1-dev-gguf: huggingface.co/city96/FLUX.1-dev-gguf/tree/main clip-vit-large-patch14: huggingface.co/openai/clip-vit-large-patch14/blob/main/model.safetensors t5xxl_fp8_e4m3fn.safetensors: huggingface.co/comfyanonymous/flux_text_encoders/blob/main/t5xxl_fp8_e4m3fn.safetensors t5-v1_1-xxl-encoder-gguf: huggingface.co/city96/t5-v1_1-xxl-encoder-gguf/blob/main/t5-v1_1-xxl-encoder-Q4_K_M.gguf flux vae: huggingface.co/black-forest-labs/FLUX.1-schnell/tree/main/vae xlabs realism lora: huggingface.co/XLabs-AI/flux-RealismLora/tree/main Mistral-ClaudeLimaRP-v3-7B-GGUF: huggingface.co/TheBloke/Mistral-ClaudeLimaRP-v3-7B-GGUF/tree/main?not-for-all-audiences=true

[@Neblina1985]

I really love your manner to explain. Really good tutorial and i can't wait to watch the next one

[@TheSneakyRobot]

Thank you so much!, Iam really happy to hear that.

Thank you for the video, it was very informative and clear, it just made me start playing with ComfyUI again.

[@TheSneakyRobot]

Glad I could help!

[@vfxfisher]

When I try to install the missing nodes - I get an error with VLM_nodes and that it has a conflict with ComfyUI-YOLO. Looking in the shell, I see these two errors: ERROR: Could not find a version that satisfies the requirement llama-cpp-python (from versions: none) ERROR: No matching distribution found for llama-cpp-python

[@mgmike444]

any advices ?

[@juanjosemondragon9838]

Hi after installing missing nodes in Low_VRAM workflow there are some still missing nodes - perfection styler - simple text - LLMLoader and LLMSampler. could help us solve it. THANKS

[@drevstudio]

I had same issue, please help us

[@Radarhacke]

Very good. At 19:55 i prefer a higher denoise Value for more details on skin (30-45).

[@TheSneakyRobot]

Tried Higer denoise, but it changes a lot of features

[@IBooGaGaI]

error when instaling missing nodes: ERROR: Could not find a version that satisfies the requirement llama-cpp-python (from versions: none) ERROR: No matching distribution found for llama-cpp-python UPD: fix it - instaling correct cuda llama_cpp_python

[@mightystudiosvids]

I was able to get past this by returning to the command prompt and installing the missing stuff: python_embeded\python.exe -m pip install scikit_build_core python_embeded\python.exe -m pip install llama-cpp-python In that order. The first one is needed to install the second. Also, the second command will take a long time to complete.

[@superrancel6633]

@@mightystudiosvids Thank you very much, this solved my problem, I just had to install cmake as well.

[@dayemrabi]

Thank you for your effort to help us. my issues: Warning: Missing Node Types When loading the graph, the following node types were not found: PerfectionStyler LLMSampler SimpleText LLMLoader

[@TheSneakyRobot]

check out the new workflow @ kzbin.info/www/bejne/a6u8mqqph6x6f5Y

[@bartosak]

awesome video - can add in post-processing some circles to point values you are talking about ?

[@Geralt26]

BTW: You would chnge anything for the 3080 10GB or run the same workflow?

[@erikschiegg68]

Take Stability Matrix to install ConfyUI and much more easily, then apply the great fun part from this video.

[@titatovernaar6029]

i got PerfectionStyler node missing i get

[@polloloco6353]

And another error: Error occurred when executing UnetLoaderGGUF: module 'comfy.sd' has no attribute 'load_diffusion_model_state_dict'

[@polloloco6353]

Any help..?

[@DIDYOUKNOWHQ9]

Bro it's using CPU not GPU how can I fix it

[@LMahesa]

Unable to install nodes: perfectionstyler workflowFlux Upscale Factor Selector Everything else is installed, just those two left. How do I fix this?

[@TheSneakyRobot]

Installing perfection styler: open the manager click on install via git URL paste the following url and click on ok and wait till its done installing. github.com/TripleHeadedMonkey/ComfyUI_MileHighStyler.git

[@LMahesa]

@@TheSneakyRobot Got perfectionstyler installed okay. Now I'm still missing "workflowFlux Upscale Factor Selector"

[@cives]

​@@LMahesa I had to install a new node Flux upscale factor selector and set it up. The one in the workflow was defective. I am also having some other issues with other nodes that I am fixing. I guess maybe the second version of the workflow damaged some nodes somehow.

[@mayankgupta2937]

Good stuff but do you have the workflow setup and configuration for higher end PC as well? i have rtx 4090 and would love to try your ideas.

[@TheSneakyRobot]

check out the new workflow @ kzbin.info/www/bejne/a6u8mqqph6x6f5Y

[@iamdihan]

does realism lora work?

[@TheSneakyRobot]

yes, it does. check the updated workflow at kzbin.info/www/bejne/a6u8mqqph6x6f5Y

[@Chris-08-15]

ERROR: CLIPTextEncodeFlux - 'NoneType' object has no attribute 'replace' // Any Idea? 🙂

[@panonesia]

i have problem with text to image workflow, especially SDXL promp styler.. whenever i change cr text input switch always stop on SDXL promp styler, let say i want to make model frome image (flowrence) i turn on (image to promp, loader and text to image) switch cr text input to 2, flowrence running well but always stop on SDXL promp styler

[@panonesia]

Failed to validate prompt for output 138: * SDXLPromptStylerPreview 155: - Value not in list: style: '{'content': 'base', 'preview': 'K:\\ComfyUI_windows_portable_nvidia\\ComfyUI_windows_portable\\ComfyUI\\custom_nodes\\ComfyUI-Prompt-Preview\\style-preview\\base.png'}' not in (list of length 107) Output will be ignored Failed to validate prompt for output 137: Output will be ignored A close-up shot of a young woman with auburn skin and dark eyes. Her hair is a vibrant shade of blue, purple, and yellow. Her bangs are pulled back, adding a pop of color to her face. Her eyebrows are a darker shade of brown, while her lips are a lighter shade of pink. The background is blurred, but it is out of focus. Offloading model... Prompt executed in 2.12 seconds

[@TheSneakyRobot]

I think this means there's an issue with the prompt styler, try bypassing it and see if the error is still there

[@cives]

I had the same error, but it's bypassed by selecting another style, other than the base one. I mean, I tried another one and it worked. I don't know yet if the issue is with the base JSON or if there are other possible conflicts.

[@kaafirTamatar]

Awesome workflow! Thanks for sharing. I'm not able to get the Perfection Styler node you're using in the Text to Image group. Can you please tell me where I can get that node?

[@TheSneakyRobot]

the perfection styler is part of the ComfyUI_MileHighStyler Notes. go to the comfyui manager and look for ComfyUI_MileHighStyler

[@djsycoassirati]

same here

[@Latentnaut]

why flux1-schnell-Q4_0.gguf? and not _1? what's the difference?

[@TheSneakyRobot]

The GGUF Model are Lighter on your GPU They Offload Like a Quantized Large Language model, you hardly Notice When they are loading. The Normal Models Are, 1. Larger and, 2. puts more strain on your GPU at least if you're working on a system with 6 to 8 Gigabytes of Vram

[@pablo.montero]

@@TheSneakyRobot In terms of quality and details, is gguf worse? BTW I have a 3090 24GB.

[@Geralt26]

@@pablo.montero I am newbe (will start my journey with AI next week but i watched some videos), but as far I know with 24GB I think you can use original FLUX model (fp16), Quantified (this Q) versions are for lower memory and if you need to use then use Q8 as most similar to original fp16 model.

[@12wsaqw]

I tried this workflow on a mac and it does not work due to issues with Crystools and VLM nodes. I will try to create a working mac solution with your workflows as a base. Thanks.

[@TheSneakyRobot]

let us know if it works when you're done

[@polloloco6353]

Great video, but VLM_nodes IMPORT FAILED, and Try to fix not working

[@polloloco6353]

There's a conflict with the ComfyUI-YOLO custom node

[@romanioamd5319]

Installed missing nodes, but displays Warning: Missing Node Types When loading the graph, the following node types were not found: PerfectionStyler workflowFlux Upscale Factor Selector No selected item Nodes that have failed to load will show as red on the graph.

[@VidayCosmos]

Flux Upscale Factor Selector its a group of nodes. Just create as he show it then group them.

[@HanjLaoye]

Thanks for the great tutorial! The following error appears when running your workflow: Prompt outputs failed validation LLMLoader: - Value not in list: ckpt_name: 'Mistral-7B-Instruct-v0.3.Q4_K_M.gguf' not in [] I downloaded the file Mistral-7B-Instruct-v0.3.Q4_K_M.gguf. Which folder should I put it in?

[@TheSneakyRobot]

You need to place it in the models - LLavacheckpoints Directory. If you don't see the directory create it. ComfyUI_windows_portable\ComfyUI\models\LLavacheckpoints\Mistral-7B-Instruct-v0.3.Q4_K_M.gguf

[@e-shop4435]

i haven't big GPU i have only gtx 550 ti 😓😓 how i run flux unlimited with realism lora as free

[@TheSneakyRobot]

I don't think that will even load the models, but here are some places you can run it for 100% free huggingface.co/spaces/black-forest-labs/FLUX.1-dev huggingface.co/spaces/Nick088/FLUX.1-dev

[@marklandmark3269]

It doesnt work for me naturally

[@ReheatedDonut]

Is Flux possible with a 1050 card and 4GB VRAM?

[@TheSneakyRobot]

I would not advice it. I think that the sweet spot is between 6 to 8 GB of VRam. but you can still use flux for free using the following: huggingface.co/spaces/black-forest-labs/FLUX.1-dev huggingface.co/spaces/Nick088/FLUX.1-dev huggingface.co/spaces/black-forest-labs/FLUX.1-schnell huggingface.co/spaces/NotASI/FLUX.1-dev huggingface.co/spaces/prithivMLmods/FLUX-REALISM

[@Dorizdt]

This video should be titled "Say Goodbye To VRAM Limitations and hello to 20 ways i will make you quit ever relying on workflows via youtubers who run their channel with AI voice" as fk me.. talk about coming up short with everything you need to make this workflow happen as per the video illustrates. eg: "LLM Loader" ... gl trying to unpack that node of knots..

[@TheSneakyRobot]

check out the updated workflow @ kzbin.info/www/bejne/a6u8mqqph6x6f5Y

[@Dorizdt]

@@TheSneakyRobot well..dont i feel like the a-hole now :D... thx

[@Geralt26]

@@TheSneakyRobot One question as I am lost a little. I loved your video but I understand something was wrong? But next video is about different topic (and will be watch it now). But how about this topic. Is there are updated FLUX workflow from this video? There are more models need to download then showed in the video?