A style LoRA for creating a rough ink brush painting style inspired by Sumi-e and other east asian ink wash painting styles. Thank you to the people helping me test the H3 version, especially to @Foxntail who also made the showcase videos for me. Check out their <a target="_blank" rel="ugc" href="https://civarchive.com/user/Foxntail/videos">other work</a>! I couldn't test it myself but it seems to work best starting the prompt with "A sumiink painting style animation". For more complex prompts you maybe need additional style description or use prompt enhancement for the style. Using a Turbo LoRA seems to decrease the visual quality and prompt adherence. In my tests with Minimax H3 I found you should use at least 30 steps without Turbo LoRA to get a high quality video. Therefore I remade two of the showcase videos without using a Turbo LoRA. Thank you to @FreijaFoxy and @SenatorMakenzie for using their LoRAs and to @Takah0x for using some prompt remixes in the showcase images of the Krea 2 version. The MOSS Image version seems a bit undertrained, but you can compensate by adding more style description, for reference look at the showcase images' prompts.trigger words H3: A sumiink painting style animation trigger words Krea 2: SumiInk, traditional media trigger words MOSS Image: sumiink, ink splatter, brush strokes, rice paper background, ink wash painting<div data-type="blurb" data-id="70">If you like the model please leave some reactions on the showcase images or videos. Also I would like to see the images or videos you generated. So if you really like the model please click "Add Review", leave a like and add some images, so I get notified you did. If you like you can also write a short review.</div>

SumiInk - Sumi-e and Ink wash painting inspired style [MOSS Image/MiniMax H3/Krea 2]

Trained with Moss Image Base, use with Moss Image SFT. Works best when adding additional style description, look at the showcase image for reference. I wanted to train further, but civitai trainer did not let me, so I probably will do a version 2 once the trainer is fixed. You can recover the Comfy workflow by dropping the images into ComfyUI. I used a custom LoRA node from ComfyUI-Lora-Manager extension, alternatively you can replace it by a default LoRA loader node.

Files with SHA256: 85bd78cc8b8a0ac214d606da1486bf25aa83e3ae249936a2d3b9849730714c33