See what Flux can do!
Choose any style and this will output a sample Flux image attempt in that style, along with a filename stamped with that style name and accompanying titled text file containing the prompt used. This is helpful in A) Identifying which styles Flux identifies and B) troubleshooting your prompts to better apply said styles.
Sample gallery of styles I've tested:
https://enragedantelope.github.io/Styles-FluxDev/
Tips:
Use ComfyManager to identify missing nodes easily!
FULLY update Comfy and associated nodes. This is still new so things are changing rapidly.
Description
Initial version. Instructions included in the workflow. I highly recommend Comfy Manager to identify and download any missing nodes!
FAQ
Comments (16)
Is it possible to automate the generation of the same prompt with different styles? Example: Choose 10 styles and have it generate 10 different images based on same prompt.
I will look more into doing this. I did try exactly that with an XY Plot, but while it kept stepping through styles correctly, it left the prompt information and the filenames static (stuck on the first style chosen). So it was limited usefulness as it made it hard to identify which style gave which output.
Hi again. I have been testing this. My previous attempt was with SDXL (quicker). It appears the XY plot nodes can't yet handle Flux's extra encoder. I filed an issue here. Hopefully will have a working solution if that gets updated.
https://github.com/TinyTerra/ComfyUI_tinyterraNodes/issues/140
"Without guidance, creativity remains elusive." - Z
Thank you for the ideas / LORA! 馃挏
Is there anyway to set it up loke an x/y/z/ grid and run every style without needing to manually choosing each one? I like your example page you made but I would like to try with some other prompts also..
Love this workflow! Thanks for sharing
Thank you for trying the workflow, and for the kind words!
I have a request in to the Tiny Terra Node dev for exactly that grid functionality. Tiny Terra Nodes work great for XY grids, but don't yet support a dual encoder setup (or at least the Flux T5 encoder). So for now, I'm in a holding pattern on being able to do this.
Issue is posted here in case you want to add in your comment or request to it:
https://github.com/TinyTerra/ComfyUI_tinyterraNodes/issues/140
@EnragedAntelope聽Ok thank you! I will make a comment there also.. I also noticed on the "new" version of comfy some of their nodes don't work. I actually had that happen with other custom nodes as well. I sure hope the creators of custom nodes can find the time to update them when the official launch of the new ui hits on the 15th which is just around the corner!
@EnragedAntelope聽They responded!!
@SouthbayJay聽wow you have the magic touch :). I look forward to their updates.
And yeah I feel bad for all of the node devs who are in a crunch now with the 15th looming. Plus a major new model just landed that has everyone scrambling... it's a busy time.
@EnragedAntelope聽I hope we see an update soon! I'm glad they responded..
@EnragedAntelope聽Looks like they pushed the update! That's fantastic!! How can we setup the grid?
@SouthbayJay聽I caught that and I set up a grid that works for image output in each style, and that works great. But so far I can't get the image names to reflect the specific style, or capture individual prompts. They stay static. I feel like those are incredibly helpful to include... if not the exact prompts then definitely the style names, so you can easily find what you want to replicate. So I am trying different things still and asking around Discords.
This is amazing, how can we as users interrogate and explore even more tags?
Since Flux uses natural language but does not always recognize specific artist names, you can ask Claude, copilot, etc. to describe an art style you like without referencing the specific artist/style name. Then ask it to create several different sample prompts for Stable Diffusion T5 text encoder.
I'd love to see a Style Comparison being made using CFG 6/Guidance Neg 10 with either the Tonemap or Dynamic Thresholding workarounds to get negative conditioning to work. For a quick example, here's how the same abstract painting prompt generated using negative conditioning.
Abstract Painting Winston Churchill looking at iPhone., Painting in abstract styles, often using non-representational shapes, colors, and forms.
That is pretty cool! I have not experimented with those methods yet. My understanding is that the dynamic thresholding method breaks down rather quickly (at fairly low steps) so isn't ideal for most workflows. I am not familiar with the tonemap method.
There are several styler nodes, and some also style the negative. If you have a workflow with the methods you described, you could use one of those stylers to also style the negative and reinforce the correct intended art style.