These models makes everything look like it's made out of wool and having been hand-knitted.
I have trained 4 versions of this model to learn more about Flux training. All models are trained with the same training settings (read more about it in my training article), using the same images, but with different captions.
Training Diary Article part 1 (lots of images)
Training Diary Article part 2 (even more images)
âš¡âš¡âš¡ Please support further training and information sharing by downloading this model during early access. âš¡âš¡âš¡
Usage
Based on which model you use, you will either have to use a trigger word, or not.
Example prompt: wizard casting a spell
W00lyW0rldFlux-SingleWord.safetensors: Trigger word is "W00lyW0rld"
W00lyW0rld a wizard casting a spell <lora:W00lyW0rldFlux-SingleWord:1.0>
W00lyW0rldFlux-WD14.safetensors: Trigger word is "W00lyW0rld"
W00lyW0rld a wizard casting a spell <lora:W00lyW0rldFlux-WD14:1.0>
All other models do not require a trigger word, but you still need to "unlock" the knowledge of the model by describing the effect in natural language. For this, I recommend that you use "Made out of wool".
W00lyW0rldFlux-NoCaption.safetensors: No trigger word, but use natural language like:
Made out of wool, a wizard casting a spell <lora:W00lyW0rldFlux-NoCaption:1.0>
I've found that most settings work well with this model. No special samplers or CFG-scale.
Read the training diary to learn more about this Flux training test. (link coming after I've posted)
Note:
I may be using the following extensions in the reference images: FreeU or ADetailer / Face Editor. I may also be using a specific Hires Checkpoint (swaps to another model when doing the hires steps). If your results don't match mine exactly, this may explain why.
Tips and tricks
Below are some tips and tricks for how to get even more out of the model.
You can always add some extra keywords to enhance the style.
made out of wool
knittedLike it?
Description
From the training diary: https://civitai.com/articles/6792
This version of the model was trained using a trigger word and WD14 captions. These captions can be generated by the CivitAI training tool. When you are at the step of uploading images, you can generate captions in this style there. You can also do it using Kohya and other trainers.
I have also released a tool to help you do it quickly, by just entering the images you wish to caption in a folder, and running a script. Check out JoyTag-Batch Github here.
I found that the effect of using this style of captioning was very strong. In many cases when I compared all 4 trained models, my preference was to the WD14 captioned ones. It seem to more often convert more of the whole image into the desired style.
Of course you have to consider that each training brings with it a lot of randomness, so maybe this was just a lucky epoch.