Thanks, Yeah, someone asked for it and I thought it would be cool to do. She was my childhood crush too :)

I mostly run it .95 and I prompt for 80s style 2d animation. I must have put the wrong strengths in the model description

@darionk , steal my prompt so you can have some fun. Here's an example:

H0LL1\/\/0UlD, a 2d cartoon video in the style of classic animation of a naked sexy fair young blonde woman facing the camera, blue eye shadow, she is showing off her genitals shyly, flirty.

The background is dark with some neon lights, Transform this video into the style of old-school animation, reminiscent of classics. Apply a hand-drawn 2D animation aesthetic with smooth, flowing character movements and slightly exaggerated expressions for a whimsical, lively feel.

Seed: 969426619930980
Shift: 9
Guidance: 7.5
Euler
Lora: .95 / All Blocks
Hunyuan BF16
FP8
Steps: 50
Length: 109

Which results in this: https://civitai.com/posts/13937749
Give it a try, let me know what you think.

When it doesn't work, check the server logs, if it's failing to load it should tell you that. You may also be missing something, or have a bad node somewhere (for example, if you have teacache disable that and it will probably fix it). If all else fails install a new blank comfy install somewhere else, copy over the couple models, and then try your workflow there. If that doesn't work then you are probably trying to use the model it's not designed for.

@Suppressor Thanks, this prompt works better and I been testing some ideas! Unfortunately, I can't do 50 steps, without taking an eternity lol
I'm using a 3060 laptop, 20 steps already takes me 10 mins with only 61 frames at 320x480 + Upscaling. Either way, I'm getting better results, and the character appears more consistent now.
You did an excel work on this Lora, the results you showed on your clips look exactly like the movie which is impressive. Mine are still not there but definitely closer now, but I will play around more with your Lora.
I do have some questions, but in regards the training if you dont mind me asking:
1. Wouldn't it had been better to just use the character name as the keyword or like Danbooru use it, for example holly_would? Or did your testing lead you to use the current Trigger word?
2. Also did you train only on Doubles or in full? I have learned recently that "supposedly" just training the double blocks can make the results more consistent and also the training faster, so I was wondering if you used it or not.
3. Have you tried just training with minimal prompts or just the trigger word with maybe 1 or 3 extra prompts to see how the model works? I wonder if Hunyuan can benefit from simple words in a Lora for more precision, instead of full sentences. 🤔

Either way, thank you for indulging me with my questions and for giving me a better start in the prompting. I'm definitely curious to see what else you plan to create when you have time!

@darionk 
Answers:

1. Wouldn't it had been better to just use the character name as the keyword or like Danbooru use it, for example holly_would? Or did your testing lead you to use the current Trigger word?

A: With Hunyuan you want to use a trigger word that the model cannot confuse for real words contained within it's natural language captioning. The more obtuse the word is, the more likely that it will work as a trigger.

2. Also did you train only on Doubles or in full? I have learned recently that "supposedly" just training the double blocks can make the results more consistent and also the training faster, so I was wondering if you used it or not.

A: I used double blocks.

3. Have you tried just training with minimal prompts or just the trigger word with maybe 1 or 3 extra prompts to see how the model works? I wonder if Hunyuan can benefit from simple words in a Lora for more precision, instead of full sentences.

A: In my experience 260 token natural language captioning is superior. That doesn't mean this is true in every instance, but for most things I try to accomplish, this is the way.

@Suppressor Thanks for the answer, the answers were as I expected based on my knowledge, but I was still curious to ask, since Hunyuan might act different than SDXL or Flux. But it seems it follows most of the rules. Although the first question comes from a theory of mine, I have thought that if a model has already data from a certain concept/character, just having the trigger word been the same as the one in the Database might just add to it. For example, Hunyuan does have data from the Disney Princess by itself and I feel your DP model added more and just helped.

Either way, it's just a thought process I had, not sure how accurate or feasible such method is in reality, I don't have the machine capable to produce these and test, the most I been able to do is Model Mergers and play heavily with Weight Blocks to produce a result I wanted.

Well again, thanks for the answers, definitely curious what else you can produce, and I will try to upload something ones I get a result I'm happy with, inside my machine limitations lol

@darionk Thanks for the buzz! I wish you luck!

A man after my own heart.

Gee, I'm sorry you don't like the trigger on the free model your downloading.

Of all the things this guy could say while downloading a free model, he chooses to complain. Probably doesn't know how to copy and paste.