Since the movie is an anthology of different stories animated by different studios, I let this LoRa be influenced by a combination of all the syles to hopefully create it's own somewhat unique style while still giving a general nod and homage to this great movie.
Z Image Turbo -
Strength of .7-.9 seem to work best. "HeavyMetalZit style of" <your prompt> will get the style working for you.
Hunyuan Video
This was created using an existing database of just 512x512 images I had on hand using Diffusion Pipe.
example prompt: HeavyMetalstyle animation of a green alien with a long face, skinny features, orange vest, tan shirt, dark green pants is sitting against the control panel of his spaceship. Lots of switches and light panels and a monitor in the background. The alien is looking at the viewer and has his left hand raised.
SDXL
Based on the popular animation styles used in Heavy Metal. Might need to increase weight to 1.2 if using a realistic model. Also helps to use tags like flat art, or 2D flat art. My first attempt at this so be nice.
Flux
Not perfect by any means but it's a good starting point. Hopefully we'll be able to train loras locally with 12gb sooner than later and I can spend more time on it. I tagged the training data in natural language but it seems to respond well to both types of prompts. Normal guidance seems to work fine. The model can get a little horny sometimes with the boobies but so did the movie so it's fitting.
Description
FAQ
Comments (6)
nice job been hopeful for more stuff like this trying to get some nice prompts but struggling right now , im doing widescreen wallpapers though so it is going to take some tinkering , some decent results so far though.
yeah far from a polished product. I think I'm pretty aggressive in my prompting style so it probably helps here. Honestly, this was more of a test run, but I was surprised at how many good results I got without much cherry picking and no editing/inpainting at all. I figured it was worth sharing for those of us that liked the old school animation style.
@Lokitsar its still the only model that can do this style so congratulations. Im revisiting some of these old prompts now that SDXL is in a better place overall. I wonder do you have your source material id love to train a pony version of this because it does struggle to work in pony models , or hell id even donate you the buzz if you want to train it.
Might wanna put "XL" in the title, would probably help you to get more downloads
Q - did you train on only the style shown in the pictures? Or did you train on the other scenes too? No wrong answer, just curious
Awesome idea, excited to try this out
Thanks for the suggestion. Very good point. I tried to train on all different types of scenes to make sure it understood the sci-fi/fantasy concept and to be a little more well rounded. Wasn't a lot of training images but I used the best and highest quality images I could find. And I made sure to train on some of the nudes as well because I think that's a big part of the movie.
This lora is awesome!! Super versitile and pretty easy to get something looking about right with minimal prompt engineering. This is one of the better trained loras I've come across. Really nice work.