Elevate Your Creativity with my FLUX Cinematic Professional Workflow!
After countless hours of hard work, I have created what I can confidently call the perfect workflow. I encourage all of you to try it and see the difference this new model can make.
FLUX Professional Workflow with txt2img
Refine
Ultimate Upscale (Better than Krea.ai)
Image saver to save the metadata. (civit friendly)
For small GPU and RAM (fp8 + fp16)
PS: I love The Flux Model! The best everyone
The final resolution of the image created with the workflow is 6144x4096. You won't find a workflow anywhere that generates images with a level of detail like my workflow.
IMPORTANT: If you appreciate the workflow, please consider leaving a review and sharing the images generated using this workflow. Your feedback is invaluable to both us and the AI community. Merely downloading without providing any evaluation is not helpful to me or the community. Thank you.
It is strictly prohibited to distribute outside civitai this workflow or hosting services in third-party image generation services, or sell my workflow on other site. You can use this workflow for personal use.
Description
FAQ
Comments (10)
The workflow looks good as a first version, but you need to adjust it to get rid of the "hallucinations" that appear in the background areas. There are additional eyes and faces that appear in the sky, on trees etc.
set denoise by Ultimate SD Upscale to 0.10 and tile by 1024 and Work!
The image from the USDU doesn't have the problem, it seems to be added after then. I can work around it and just thought you should perhaps release a version that works without having to change the settings. Also, there are multiple checkpoint and lora loaders, only one of which says it is for SD1.5, but should they all be SD1.5, or are the unnamed ones for SDXL? I presume they should all be SD1.5, but it's not clear and the workflow/instructions could do with some clarification.
Hi, I tried using this workflow and got an error for a missing node "AV_FaceDetailer" which the node manager couldn't find. I found it after some searching, it belongs to the Comfyui-Art-Venture node, if anyone else is also searching.
It turned out to be the Comfyui-impact-pack for me actually
Have some circular dependency hell with this. A few items require different versions of the same package. (timm 0.6.1.3 and 0.9.16).
version 2.0 with AIO
FLUX + SDXL + Pony + IMG2IMG + Upscale + (Final) AIO
With v2.0, I keep running into crash error after crash error, and troubleshooting and debugging is difficult due to the meatball of piled utility nodes that need to be pulled out from under a locked node and turn into spaghetti to sort through.
The text2image function cannot run without the image2image function also turned on, because it requires the positive prompt to be routed through your Apply Controlnet conditioning node.
Unfortunately this is a huge problem, because order to generate a text2image, the user also needs to have the image2image function enabled and active so that the Apply Controlnet node can be routed through. This forces Controlnet image2image processes that user doesn't need, to be running during a text2image run.
This can be fixed by adding a Fast Groups Bypasser next to your Fast Groups Muter, and disabling the image2image group through a bypass instead of the muter, which will fix the reroute issue.
From here, I believe that I have the text2image part straightened out, however the process itself takes an incredibly long amount of time, and I'm literally stuck in the middle of the 'Simple Custom' node for the last 15 minutes as I write this comment. The command line does show that it is going up slowly, but I feel something must be wrong, since it shouldn't take 15+ minutes just to generate a simple, single text2image image. My GPU has 16gigs of ram and I'm using the exact same checkpoints/prompts that you last ran when you saved your json.
I haven't even been able to try the Flux part out as of yet, as I've been stuck waiting for the SDXL text2image to finish generating. According to the steps counter in the command line, I have another 10 minutes or so before the first image attempt is finally complete.
Update - I canceled the SDXL image generation after about 30 minutes of waiting. I then attempted to try generating a Flux image and ran into more errors. There are no notations and no documentation anywhere to explain what settings can be adjusted or what models and files are needed in order for this workflow to function. After digging around myself, I was eventually able to find all the required models required by the loaders, but continue to run into problems. This workflow unfortunately does not do what I expect it to do without significant investigation and effort on my part.
@kizzi i, update coming soon :)