I really liked trying out all models I downloaded to re-create my favorite Disney Animation characters. I was surprised by how good some models were able to recreate Anna and Else just from my description, but it probably helped A LOT that the models likely already had some knowledge about who the characters are.
Some models support taking a reference image to start with, so that is exactly what I am going to do next. I use the EXACT same prompts and settings that I used in my previous Anna and Elsa post, but also provide the actual image I was trying to recreate. As a reminder, this is the image I am trying to recreate again…
Anima. As far as I could figure out, this model does not support taking an image as input. But I didn’t really like the results of this model anyhow, so moving on…
Dreamshaper 8 DOES support a reference image and it is probably the best model to try this with anyway, because its output by pure description didn’t look anything like what I wanted to achieve. You can spply a “weight” to how strongly the image should influence the output somewhere between 0 and 2. So, here is the exact same run WITH an example image fed to it with the influence set to 1…
Wow, that went from nice images that looked nothing like I wanted to absolute rubbish that looks only slightly like the provided image and prompt. Let’s give it one more try and set the “influence”Weight” slider to the max…
No! Just no! Very disappointing results again. I am probably using this model completely wrong, but let’s just try to forget about this horror and move on…
FLUX.1 dev gave me one of the most satisfying results in the previous run, so I am VERY curious to see how it will perform when I add the actual image to it…
VERY nice results again, BUT… the model is CLEARLY referencing its own knowledge about the characters, because there are details in the dresses that aren’t even visible in the given image and also not specified in the prompt!
All the other models I use to not provide an option to offer a reference image, so I guess I will have to stick with FLUX.1 to generate my Anna and Elsa images for now…
Leave a Reply