CivArchive
    Flux Street V1 - Best Quality with SDXL refiner 8GB friendly - v1.0
    NSFW
    Preview 22584652
    Preview 22584650
    Preview 22587674
    Preview 22587771
    Preview 22587963

    Flux Street V1 8GB Vram/32GB Ram friendly

    I´ve created this workflow based on my Quality Street workflow to get the best quality in best time even with a 8GB GPU.

    This workflow includes:

    • Promts with wildcard support

    • 3 example wildcards

    • basic generation using the Flux model

    • a 2 steps SDXL refiner with upscaling to get the best quality possible

    I have used only important custom nodes. Maybe you have to install the missing ones in the Comfyui manager and also update Comfyui to the latest version.

    Please give me a good review if you like it :-)

    How to add Flux-Loras:

    1. Click right on the Clip Text Encoder and expand it.

    2. Connect the Lora Loader between Loaders and next nodes:

    Description

    FAQ

    Comments (40)

    MemeaterAug 4, 2024· 4 reactions
    CivitAI

    nice, though me thought the reverse gonna be cool, say making concept from 1.5/XL (allowing old loras into flux) then adding detail and upscale via flux (as refiner)

    mrdongtalk92Aug 4, 2024
    CivitAI

    i cant find node Seed Generator and easy showAnything, pls help me

    WhatDaFAI
    Author
    Aug 4, 2024· 1 reaction

    did you try with missing nodes in the manager? but you don't need both. you can right click on sampler and change the seed input to widget. show any just shows the prompt.

    3dforever1969Aug 5, 2024

    @WhatDaFAI yes the manager does not show that missing node.

    emoabstractsAug 7, 2024

    the Seed Generator node is part of the Comfy-UI-NegiTools model (https://github.com/natto-maki/ComfyUI-NegiTools)

    DcaffreyAug 4, 2024
    CivitAI

    Your workflow looks great! Do you know if we can add a Lora? If so, how?

    WhatDaFAI
    Author
    Aug 4, 2024

    Flux doesent support Loras yet.

    Visidious99Oct 10, 2024

    @WhatDaFAI With Flux supporting Loras now, is it possible to add a multi lora loader to this workflow? I have tried to add one but I am a comfyui noob

    WhatDaFAI
    Author
    Oct 11, 2024

    @Visidious99 at the moment im very busy so im not sure i will have time to update it this weekend. But take a look at the description - i just added how you have to add the lora loaders to it.

    Visidious99Oct 11, 2024

    @WhatDaFAI Thanks for replying, I was able to figure out how to do it, all good now thank you again!!

    ZomblexAug 4, 2024
    CivitAI

    Error: "When loading the graph, the following node types were not found:

    ModelSamplingFlux"

    Updated ComfyUI and also nothing in missing nodes Manager.

    WhatDaFAI
    Author
    Aug 4, 2024

    That's strange, that node came yesterday with the comfyui update. But you can remove that node and the workflow should work without it. Just connect the model to the next node.

    bitshifter52425Aug 4, 2024· 1 reaction
    CivitAI

    It doesn't load all the nodes required and there doesn't appear to be a reference on where they can be found. What follows is a summary of the error messages:

    When loading the graph, the following node types were not found:

    JWImageResizeByLongerSide

    JWInteger

    SDXL Prompt Styler (JPS)

    easy showAnything

    WhatDaFAI
    Author
    Aug 5, 2024

    Open Comfyui manager and look for "install missing nodes" thats the easiest way to get them.

    ishadowxxAug 5, 2024

    I second this. None of the missing node are in the manager.

    3dforever1969Aug 5, 2024

    i have one missing "undefined" node and it is not found by the manager. all other nodes are installed by the manager.

    SaftsackAug 7, 2024

    also make sure you have updated everything, both comfyui and all your plugins

    weokrntg1243Aug 6, 2024
    CivitAI

    is there a way to turn off the refiner?

    WhatDaFAI
    Author
    Aug 6, 2024

    yes you can bypass the 2 ksampler or add a groups bypasser like fast groups bypasser from here to switch the whole group off: https://github.com/rgthree/rgthree-comfy

    0l1v1aR0551Aug 6, 2024· 1 reaction

    yes - do not use this workflow at all, just make your own, and adapt it to your needs

    JoinSitesAug 7, 2024
    CivitAI

    i have an error message and the black blank image is generated. "/home/image/ComfyUI/nodes.py:1446: RuntimeWarning: invalid value encountered in cast

    img = Image.fromarray(np.clip(i, 0, 255).astype(np.uint8))

    model weight dtype torch.float16, manual cast: None"

    WhatDaFAI
    Author
    Aug 7, 2024

    sorry i cant help with that - looks like a python programm error. You can just try to update all nodes and comfyui in the manager and check if all models are at the right places. or try th 8fp clip.

    maitruclamAug 7, 2024
    CivitAI

    And we have new way to run Flux ez with 1 click: https://civitai.com/models/628682/flux-1-checkpoint-easy-to-use

    AfronautxAug 7, 2024
    CivitAI

    Where is fluxvae.safetensor located?

    tazztoneAug 8, 2024· 1 reaction

    you need the ae.sft file i think

    WhatDaFAI
    Author
    Aug 8, 2024

    its under ComfyUI\models\unet

    mmdd2543Aug 8, 2024· 1 reaction

    If you don't have it, you can download it from the Huggingface page of the Flux model ( https://huggingface.co/black-forest-labs/FLUX.1-dev/tree/main ). Download the file "ae.safetensor" and place it in the appropriate folder. You can optionally rename it to "fluxvae.safetensor". Then you can pick it from the "Load VAE" node in ComfyUI.

    fawoyo7780Aug 10, 2024

    im getting an error in dualclip loaded. which model do i load there?

    tazztoneAug 8, 2024
    CivitAI

    somehow this workflow uses 23GB of VRAM just to generate 1st pass. so it takes 60 seconds with flux schnell on 3090 while other workflows generate images in 4 secs. something is off

    txhnoAug 13, 2024

    which is the workflow which generates in 4secs

    tazztoneAug 13, 2024· 1 reaction

    @txhno idk was some basic one with flux schnell 4 steps.

    but here i think i just used the wrong weights settings (default, which is fp16 instead of fp8).

    but now that nf4 flux is out i think that's the way to go anyway. and I'm switching from comfy to forge for a while

    EpluchatorAug 10, 2024· 2 reactions
    CivitAI

    This worked great for me. Thank you, great workflow and customization!!!

    parker5288761Aug 11, 2024
    CivitAI

    When loading the graph, the following node types were not found:

    JWImageResizeByLongerSide

    Seed Generator

    JWInteger

    easy showAnything

    I went into the manager and installed missing nodes but it fails import of 126,179,423
    On console it says 'object is not subscriptable'

    Runing on linux not winblows

    WhatDaFAI
    Author
    Aug 13, 2024

    Im running it under window. But you can replace the nodes because they all have very basic functions. Easy show anything you can just remove.

    DreamScapesUnleashedAug 15, 2024
    CivitAI

    Great Workflow! Used the NF4 checkpoint with it and ended up with really nice images in the end.

    Thanks for your hard work!

    modzhahead158Sep 20, 2024
    CivitAI

    Is it possible to use the basic generation by SDXL, and refine using FLUX GGUF?

    WhatDaFAI
    Author
    Sep 20, 2024

    yes it's possible, but you have to replace the sampler and models. But generally that works too.

    SouthtownBluzJan 26, 2025

    This is what I was thinking of doing. Ever figure it out?

    _degenerativeai_Apr 18, 2025· 1 reaction
    CivitAI

    This is my favorite Flux workflow. I hope you someday do a similar one for HiDream. The SDXL refinement helps a lot with the shiny skin texture that Flux and HiDream suffer from.

    symanleedsuk556Jul 19, 2025· 1 reaction
    CivitAI

    Great workflow just the best

    Workflows
    Flux.1 D

    Details

    Downloads
    3,471
    Platform
    CivitAI
    Platform Status
    Available
    Created
    8/3/2024
    Updated
    5/12/2026
    Deleted
    -

    Files

    fluxStreetV1BestQuality_v10.zip

    Mirrors

    HuggingFace (1 mirrors)
    ShakkerAI (1 mirrors)