V2:Revisited, same dataset as before but I retrained using Ostris V3 adapter + microbatch 8 + compile arguments. The result was a more stable lora, lines are more solid with less distortion. Motion seems fine but let me know if anything bad.It trained around 7 hours, and this was the first time I used a validation dataset! It landed more or less around what I expected and the validation tensorgraph more or less aligned with that thought.I need more time to play around with the checkpoints but the 1800 seemed best (I had trained up to 2400 steps before I stopped). I may find something better and upload an alternate checkpoint.Just like rezero h3 lora, this will get the general look of the characters but a lot of details are wrong. Will probably work great with a reference character sheet for the character you want to make. Again this lora is more for style than creating the characters from the show.V1: Trained on the same dataset as my Krea 2 gurren lagann lora. Images only, 441 screen captures from the show (pre-time skip only). I think I will try video training later, right now video training for H3 is not super great, images is perfectly fine and fast. Characters sort of work, I think this is more useful for style instead of character creation, a separately trained character lora probably would be better. Trained on FLV2A model.The 17k checkpoint is best, but give the 9.9k steps a try if you think its overcooked. Not tested with lightx speed up loras. Recommended euler beta57 20 steps at 1 strength.<h3 id="style-trigger:">Style trigger:</h3>"2d anime style, Gurren Lagann Style"<h3 id="character-trigger:">Character trigger:</h3>"char_firstname"<h3 id="here-is-how-to-recreate-some-characters-(check-the-captions-data-for-more).-some-are-doable-but-a-bunch-not-so-much.">Here is how to recreate some characters (check the captions data for more). Some are doable but a bunch not so much.</h3>Yoko:char_yoko, A woman with long red spiky hair tied in a long pony tail and chopsticks and skull accessory, red flame-patterned black bikini top, light pink scarf, black shorts with a white studded belt, pink thigh-high stockings, fingerless black gloves, and white and red boots. She holds a massive dark grey, hexagonal-barreled rifle.Simon:char_simon, A young man (or “boy”) with spiky dark blue hair, a blue jacket over his bare torso, and red goggles on his head Kamina:char_kamina, A muscular man with spiky blue hair and blue spiral tattoos. He is wearing orange frameless pointed triangle sunglasses and a red tattered cape. He has bandages on his forearmsNia:char_nia, A young woman with wavy blonde and light blue hair and teal eyes with red pupils cross-like in a  floral shape. She wears a pink and white dress with a large gold belt and cuffs. And an elaborate golden collar with red and green gems, a red tie, and a pink and white hair accessory.Gurren &amp; Lagann:A humanoid mecha (might want to describe about the face on his torso etc, but its not captioned much). Be sure to mention the samurai horns on the head if you want that etc. Every form of them is in the data too (flying mode, battle ship etc.)Viral:char_viral, A man with shaggy blonde hair that covers one eye. He wears a jacket with a white fur-lined collar and red shoulder pads.Mecha:All mecha in this are labeled using “mecha” you can say like “whale-like mecha” or “turle-like mecha” etc. to get different types. There is probably all the different ones in the training data. Just use this phrase “mecha” to trigger it.Beastmen = “creature” , ie turtle-like creature, etc.Buta:char_buta, A small, brown, pill-shaped pink pig-mole creature with two long, thin antennae, a curly tail, whiskers and round sunglassesLordgenome:char_lordgenome, A large, extremely muscular man with a shaved head, a dark stylized beard, and intense, light-colored eyes. He is shirtless wearing a dark garment with two large, silver, U-shaped bracelets on his arms. (I guess I missed captioning the beard, try adding that word in too)There is more, I should have covered every character major or minor in the first season. So try yourself to describe them or check the captions. the syntax is char_(firstname) + a description of their outfit and appearance.<h3 id="training">Training</h3>Trained using <a target="_blank" rel="ugc" href="https://github.com/AkaneTendo25/musubi-tuner">Musubi Fork by Akanetendo25</a> with --h3_spatial_density_jitter 0.2 --h3_guidance_distillation_scale 3.5 --h3_base_preservation_loss_weight 0.02 \ and using adam8bit 5e-5. At rank 16, 1376 x 768 repeats 1. On a 5090 locally it used only around 25gb or so of vram and trained around 1.5s/it, very fast. It picks up the style pretty quick, I think around 5k steps even noticeable differences, I think 9.9k was quite usable but I like where 17k steps ended up.

Gurren Lagann Anime Style Lora H3

Revisited, same dataset as before but I retrained using Ostris V3 adapter + microbatch 8 + compile arguments. The result was a more stable lora, lines are more solid with less distortion. Motion seems fine but let me know if anything bad.It trained around 7 hours, and this was the first time I used a validation dataset! It landed more or less around what I expected and the validation tensorgraph more or less aligned with that thought. I need more time to play around with the checkpoints but the 1800 seemed best (I had trained up to 2400 steps before I stopped). I may find something better and upload an alternate checkpoint. Just like rezero h3 lora, this will get the general look of the characters but a lot of details are wrong. Will probably work great with a reference character sheet for the character you want to make. Again this lora is more for style than creating the characters from the show.

Files with SHA256: 727c089d83baac1fc181c3f62a70e90b8858ca564b97aacaff768aee8b3e2f1e