Type | |
Stats | 828 0 |
Reviews | (100) |
Published | Feb 27, 2023 |
Base Model | |
Training Images | Download |
Hash | AutoV2 B57B06BAB3 |
Creating these models takes time, money, and effort and it's great when that effort is rewarded. If you really like it you can buy me a coffee ☕.
If you like this model please consider checking out some of my other projects. I've made LoRAs to emulate the style of Super Crooks, Fullmetal Alchemist, Death Note, and Chainsaw Man.
This is Just a LORA that emulates the style of the movie Spider Man: Into the Spider-Verse. Depending on the subject you're trying to create, you'll need to play around with the strength. All these images were created with Dreamshaper and Deliberate as the models. X-formers is on.
I've found this model is a lot less fire and forget than my others. It may be because of the unique animation style. I may train a checkpoint later down the road and see if it's better. As you can see from the examples it is quite good.
One known issue: It seems to really have problems with people of color sometimes. Other times it generates them with no problems. It may be because it's overtrained on Miles' face, but they often end up looking like him to one degree or another. If I decide to retrain this model I will likely prune the dataset significantly. It is possible to fiddle with the settings and get a good PoC, but it's harder than it probably should be. Frankly though, this is an issue with all AI models to one extent or another and something everyone in the AI community should be thoughtful about.
I trained it on around 4,000 still images from the movie, so it's quite good at emulating the style. No need to give it a particularly lengthy prompt these images were generated with really simple prompts.
This LORA mixes well with others as well. So if you want a trained character/person in the Spider Man: Into the Spider-Verse style that's possible. If you have any questions, comments, or concerns feel free to send me a message. I'm happy to chat about my methodology.
Going forward I'm going to upload the training data for all my models. Gathering these images in this quantity isn't something everyone in the community can do, so providing them will give model makers stronger tools.