⚠️ NOTICE ⚠️
This workflow has now morphed into the UniFlex-Workflow 9.0!!!
Description
Significant restructuring and additions.
FAQ
Comments (13)
Wow thank you so much this workflow is neat, clean, easy to understand. Awesome
You are welcome. 😎
💪 Flexi-Workflow v4 just released.
SamplerCustomAdvanced
mat1 and mat2 shapes cannot be multiplied (4096x64 and 128x3072)
It executed normally once then I had these errors.
Can you share more? What variant and what recipe? If it ran once, then the error does not seem to be with the workflow itself. Unfortunately, there are some glitchy things with ComfyUI and a few of the custom nodes I have no control over.
* Check that you are running the recipe EXACTLY as provided, until you understand the workflow.
* Can you run the Lite version, and does it work?
* Try right-clicking on the load image node (02a) and "Reload node".
* If using Canny or Depth, try changing the resolution (101a).
* If using Redux, try changing the crop position (101a)...or disconnect the mask altogether.
What does the terminal (or command window) show?
the error usually happens because of the mask node if you are not applying the mast node remove it and it will work
@MaxFir3 I presume with using the Redux model? ...it's pretty glitchy, to be honest, but I think the problems are with ComfyUI and/or the custom nodes used. Disconnecting the mask does seem to fix the errors you are encountering, but I'll continue to investigate and hopefully figure out a better solution.
@kaptainkory Thanks and your workflows are lovely
💪 Flexi-Workflow v4 just released.
For anyone encountering errors running the workflow with Ultimate SD Upscaler:
* Follow the given—slightly revised—instructions carefully! Notice you must disable 03e or tiling will not work correctly:
💪 :: Toggle to "yes" 01 01a (or 01c); 02 02a and 02c; 03 03b–d (not 03e!); 05; and 1004 1004c (and opt. 1004a)
03b :: Conditioning switch = 1 (no ControlNets +), or 3 (to run with other ControlNet)
05 :: Image switch = 4 (save upscaled image Ultimate SD)
1004c :: Image switch = 1 (upscale loaded image) or 3 (upscaled image basic)
OPT :: Load Flux Sigma Vision Alpha1 (03c) and/or Lumiere alpha model (01a) recommended for realistic
OPT :: Load GGUF (01a) for low VRAM
OPT :: Toggle to "yes" 101 101b to run with ControlNet, such as upscaler (tiled) and/or canny
* For the above in the SDXL variant, toggle to "yes" 03a and set the latent to 2 (image-to-image).
* In limited testing incorporating ControlNets (not Flux Tools), these seem to be compatible for Flux: Jasper AI Upscaler and XLAbs AI Canny v3...and these for SDXL: TTPLanet SDXL Controlnet Tile Realistic v2, Xinsir Сontrolnet Tile SDXL 1.0, diffusers Canny, sai XL Canny, and t2i XL Canny.
* However, these seem to be INCOMPATIBLE for SDXL: bdsqlsz Tile Realistic, bdsqlsz Canny, and kohya Canny. It seems these may require image dimensions in multiples of 32.
* The Flux Tools Canny model also works with the checkpoint, but not LoRA.
Having issues installing the Depthflow nodes, any ideas? Is the nodes github page down?
This is the corrected link:
https://github.com/akatz-ai/ComfyUI-Depthflow-Nodes
If you have issues installing via Manager, try to the manual install instructions.
💪 Flexi-Workflow v4 just released.