V5 - Added Latent Noise Injection section to the workflow - Improves the quality and details of the image
Added the power lora loader back in for ease of use
The Red Nodes are the model selection nodes
The Yellow Nodes are settings nodes
The Green Nodes are the prompt, LORA loader and image chooser
The Purple Nodes at the bottom are The Face and Eye detailers and the Ultimate SD upscaler. These have been collapsed to keep the workflow clean, however you can click on the little grey dot on the left of the node or right click and expand to change any settings in these.
I have included the upscaler model (4xFFHQDAT) i am using and the face and eye detection models in the zip file. Save these in the relevant places
Upscaler - \ComfyUI\models\upscale_models
Face and Eye Detection - \ComfyUI\models\ultralytics\segm and \ComfyUI\models\ultralytics\bbox
Sam Loader Model - \ComfyUI\models\sams
V4.5 - Added a second pass for the Face and Eye Detailer
Made the workflow more compact so that it is easier to see and follow along
Re-introduced the fast bypasser/switcher as there no longer appears to be any issues with it.
The Red Nodes are the model selection nodes
The Yellow Nodes are settings nodes
The Green Nodes are the prompt and image chooser
The Purple Nodes at the bottom are The Face and Eye detailers and the Ultimate SD upscaler. These have been collapsed to keep the workflow clean, however you can click on the little grey dot on the left of the node or right click and expand to change any settings in these.
I have included the upscaler model (4xFFHQDAT) i am using and the face and eye detection models in the zip file. Save these in the relevant places
Upscaler - \ComfyUI\models\upscale_models
Face and Eye Detection - \ComfyUI\models\ultralytics\segm and \ComfyUI\models\ultralytics\bbox
Sam Loader Model - \ComfyUI\models\sams
V4- Added a Face Detailer and Eye Detailer nodes
Face Detailer Model - https://huggingface.co/Bingsu/adetailer/blob/main/face_yolov8m.pt
Eye Detailer - https://civarchive.com/models/178518/eyeful-or-robust-eye-detection-for-adetailer-comfyui
V3(GGUF)
https://huggingface.co/city96/FLUX.1-dev-gguf/tree/main
https://huggingface.co/city96/t5-v1_1-xxl-encoder-gguf/tree/main
Added an image chooser before the up scaling starts - By default the workflow will produce 4 images initially. You can then choose which one(s) you want to progress with to upscale and save. That first run will take longer as its creating 4 images. You can alter how many it runs in the parameters group.
The prompt generator has been updated to a Searge LLM Node which works a little better than the previous one. It still requires a Groq token like the previous version
Removed the switch as it was causing some issues for some people. You can manually bypass the groups to avoid running the prompt generator each time you try to run the image generator.
V2(GGUF) - Updated to accommodate the GGUF models and encoders found here
https://huggingface.co/city96/FLUX.1-dev-gguf/tree/main
https://huggingface.co/city96/t5-v1_1-xxl-encoder-gguf/tree/main
Prompt Generator
Added a LLM prompt Generator using Groq. Type in a simple prompt and the model will give you an expanded and more detailed prompt for you to paste into the image gen prompt box to edit to your liking before running.
PLEASE NOTE: The prompt generator does require a Groq api key (free) that you can get at https://groq.com/. to apply the key, simply add a TaraApiKeySaver node into your workflow, paste the key into it and run the generator. You can then delete the TaraApiKeySaver node
A switch node has been added so that only the prompt gen or the image gen is ran at any given time
V1.1: Added in Max shift and Base shift Node that was missed from V1
A simple Flux Dev workflow that allows Multiple LORA selections, a Resolution chooser, Ultimate SD Upscaling and an image comparer
The workflow will save a copy of the normal image and Upscaled image in separate folders within your ComfyUI output folders
Very much still a comfyUI beginner so this is just a mashup of multiple Workflows that I've used and picked out the useful sections of each that i needed.
Description
Changed the Unet Loader and Dual CLIP Loader to (GGUF) to accommodate GGUF Flux Models
Added in a prompt generator
Added a switch so the prompt gen and image gen can be ran separately
FAQ
Comments (11)
Are gguf faster, or just allow it to run with lower VRAM?
Fantastic workflow! just what I was looking for! I notice the switcher seems to lock up when turning off image gen and going to prompt and have to refresh. Is this normal or have I just not figured out how it works yet? Thank you!
thank you, very impressive workflow!
Incredible but i have a suggestion, can you do something similar but instead of text pompts, the let us to choose the ability to upload existing images ?
fantastic explanation of where everything goes, look forward to trying it out,
Thank you!
Could you also add hand-feet detailer?
Can you add a hand and feet detailer as well?
workflow doesn't connected nodes properly - https://ibb.co/8YWPdpb
Thank you, this is the best workflow that I've ever worked with to get real quality image generations and getting them upscaled with the right details.
I have been running into an issue with the workflow, and have found that the .zip file does not include the upscaler:
ComfyUI\models\ultralytics shows both the bbox and segm folders, bbox only includes the face_yolov8m and hand_yolov8s .pt files while segm has the person_yolov8m-seg .pt file,
\models\sams has sam_vit_b_01ec64, but \models\upscale_models doesn't have the upscaler.
Items were moved automatically by drag-and-dropping the simpleFluxDevWorkflowLorasFace_v5 .json file into the comfyUI workflow space.
Checking for the model directly, the 4xF... model doesn't appear in the ComfyUI Model Manager, neither does the eye detailer, but the models listed above do.
Hope this helps solves some problems for others, and I'd love to know if/when you can fix that: I love your work.
This is really the best I've found so far! Gives you options right away, choose and continue with the best! Magnifies it perfectly. Thank you very much!
Here's exactly what I would add here for a perfect workflow:
1. Improving hands
2. Leg improvement
3. Give the option to use regular flux models optionally not gguf, need safetensors two.
Ideally, an example of a picture would be used to accurately convey the pose of a person wearing a mask.
4. Optionally, for example, add a block to choose from img2img so that the original image is recognized, an LLM description is written about to it, and then everything according to the workflow is as it is. See fluxSUPERWorkflow_v211
If you combine all this and give the control to turn on and off the optional pieces, it will be a fairy tale! Dude, please don't quit! We believe in you!
Looks like we don't have an active mirror for this file right now.
CivArchive is a community-maintained index — we catalog mirrors that volunteers upload to HuggingFace, torrents, and other public hosts. Looks like no one has uploaded a copy of this file yet.
Some files do get recovered over time through contributions. If you're looking for this one, feel free to ask in Discord, or help preserve it if you have a copy.