CivArchive
    Hassaku Anima by Ikena - v1
    Preview 1
    Preview 2
    Preview 3
    Checkpoint/Model created by Ikena (please credit them and support them whenever you can): https://civitai.com/models/2641326/hassaku-anima Description by Ikena: Hassaku aims to be an anime model with a bright, distinct, slightly 2.5D anime style. Join my Discord https://discord.gg/zSR5FcYWWE for everything related to anime models and AI art. If you’d like to support my work, you can do so on SubscribeStar https://subscribestar.adult/citrus-models Every bit helps me continue making AI models and even an AI VTuber, which is currently in the works. How to Use 30-50 steps, CFG 4-6, recommended sampler er-sde. - Put @ in front of artist tags - Use lowercase for tags and spaces instead of underscores. - Recommended positive: "masterpiece, best quality, score_7,", but since masterpiece is altered, in some cases masterpiece is enough - Recommended negative: worst quality, low quality, score_1, score_2, score_3, blurry, jpeg artifacts, sepia," - When using a tag that is different between Danbooru and Gelbooru, prefer the Gelbooru version. Supporters Special thanks to my supporters: pttcot, SETI, Kodokuna and dataset captioners mariotheknight and tamashiicolle Recommended Resolutions for Anima Here are some resolution suggestions for Anima: 1536 × 640 1344 × 768 1216 × 832 1152 × 896 1024 × 1024 896 × 1152 832 × 1216 — most recommended 768 × 1344 640 × 1536 Larger resolutions are possible. Note: It is a Diffusion Transformer (DiT) model, it is less restrictive on sizes in compare to a Unet like Illustrious.

    Description

    This is the V1 release of my Anima-based style fine-tune. The 0.1 version was a small internal style-training experiment and was not initially planned for release. V1 focuses on better use of the style training set with staying closer to the original Anima tagging logic. Unlike a style LoRA merge, this is a model natively trained with a style, so tags such as traditional media, watercolor, or specific artist/style tags should still work normally. The style will vary, especially when generating characters. It is trained to “uplift” the generations, more or less noticeable through details such as the characteristic hair shine. If a result differs strongly in style, it is usually also a strong style outlier in Anima Base on that same seed. For this version, I tested character training with NTE characters and Femme Soule from pukey goddess shot trick. Some NTE characters had a low image amount to train on, so those characters will not work. The training was done using a mix of tags, long natural-language captions and short natural-language captions. The ratio is around 3/5 normal booru tags to 2/5 natural-language captions. The main test was to see how to train completely unrelated franchises into one model and get both working together in a single natural-language image. Character forgetting is a problem in Anima. My training makes that decay extremely slow, but it is still there. Extremely weak, but still there. V1 Notes Compared to V0.1, this version should be more stable in backgrounds, with a more consistent use of the trained in style. Disclaimer This model is still experimental, but I am more confident that V1 produces more cohesive backgrounds with better-looking natural-language generation compared to V0.1.

    FAQ