https://civitai.com/models/1330314/m2f-transformation

Generates M2F/TSF cartoon sequences. Trained on synthetic images, it should generate more aesthetic images out of the box than most of the existing LoRAs of this type. I'm hoping it also allows better results with less complex prompting.

Post your results! I'd love to see your creations.

Version 2 Update 🎉

Trained on many more images than 1.5, this is the most creative and flexible version yet. I think it's suited to all purposes but if you find it a little too creative you may want to try an earlier version (especially on Pony). Check out the galleries for earlier versions!

Tips:

• After experimenting with a lot of samplers, I think heun is the most likely to keep a consistent order. Regenerating a DPM++ or Euler seed with heun will sometimes fix it, while a seed that fails in heun will usually fail in any sampler. Note that heun is twice as slow as other samplers and needs half the steps. Works well with karras scheduler.

• Lora strength 1 works pretty well. Going lower is more likely to create all female sequences (which can still be fun), potentially more creative sequences, more out of order steps. Going higher can give more masculine starting characters, but also tends to produce longer sequences. At very high levels it will degrade the image.

• To get the right sequence length I find it works best to count people like "(4 people), 2boy, 2girls". Portrait and square mode also tend to generate shorter sequences than landscape.

• For better or worse, realistic and semi-realistic models are not great at androgyny without help. Illustration models do better.

• It likes to hallucinate extra figures into the background. Recommend (blurred background) in the prompt and (people in background) in the negative.