Anima ReStyler

You probably know about this LoRA that turns Anima model into a character focused image-edit model. It's great and it allows you to edit existing images using vast knowledge of Anima Base model. But unfortunately it's too rigid, can change outfit and replace background but changing pose is almost impossible - feels like ControlNet lineart.

One day I've been testing this LoRA and discovered one interesting behavior: it can generate new content in the areas that are empty. This got me thinking what would happen if I

• Attach solid color block to the input image

• Run inpaint workflow with mask covering only solid color area

• Add (split screen, multiple views:1.2) to the prompt

This is what happened:

So basically solid white rectangle on the right is the canvas for the model to work with, mask limits generation to this solid color area, split screen in prompt makes model pay attention(!) to the left half of the image, while edit LoRA handles Cosmos-Reference image condition to achieve character consistency. What is Cosmos-Reference you ask? It's a custom node that enables image conditioning in Anima - you need special LoRAs for it to work and Anima Edit is one of them.

So I've made an easy to use workflow that handles image stitching and cropping for you. You only need to specify target image size and write your prompt - change style for your characters or make them perform in the new environment. Edit your image using all concepts knowledge of Anima models in other words.

I named this workflow "Anima ReStyler" but it's actually more general - "Create new image featuring character from the input image" - something like that.

Some tips:

• Not all seeds are equal - some seeds will ruin the generation while others will work perfectly. Anima has this instability

• You don't really need to describe your character design for this to work but sometime it helps with trickier generation

• Perfect input image is a character in the neutral pose at simple background. If your input image has characters in the difficult pose or background is too flashy you can use Flux and Wan to untangle it before using this workflow

• This workflow sometimes struggles with monochrome images/sketches - colorize them first using original Anime-Edit workflow for example

• It's easy to change style but hard to maintain it. If you want to change pose of your character and keep it style 100% consistent you'd better use Wan 2.2. I have the workflow for this task specifically

• Anima can handle prompt weights like (at night:2.0) without breaking - use them to push your generation when needed. Also my workflow use Schedule Prompt so you can use extended syntax for prompting: [:close eyes:0.3] - here closed eyes will only activate after 30% of generation steps have finished.