!!! PLEASE READ INSTRUCTIONS BEFORE USING !!!

More character models are coming soon, please be patient since quality takes time!

Tired of inaccurate, inflexible Overwatch LoRA's that look nothing like the actual character? Then it's about time you try a model that was trained on actual gameplay assets for maximum canonical accuracy!

Model usage instructions and recommendations:

• There are actually two files total for each character, a LoRA (this one) and a Dreambooth model for advanced users. I recommend beginners to use the LoRA alone, and then use the dreambooth model together if they want to maximize accuracy. More details below.

• The LoRA should work with both realism and 2D models, but given that it was trained on game assets only, don't expect fantastic amounts of detail without some intervention.

• This isn't a one-click wonder model, and as far as I know, none exist. These are a set of tools for artists and should be treated as such. So please, don't come yelling at me if you can't reproduce the sample images with a single click. CivitAI generation data doesn't capture the entire process needed to make good images, although I might upload a video tutorial detailing the process sometime.

• Useful LoRA strength should be anywhere from 0.15 - 0.65 depending on how little or how much you want the resemblance to Widowmaker to be. SETTING THE STRENGTH TOO HIGH WILL RESULT IN BLURRINESS, especially to the face.

• You can find all the keywords the model was trained on by inspecting the LoRA in A1111, but I'm going to list the most important ones anyways that you will want to experiment with, either by putting them in your prompt or in your negatives:

((from_behind)), ((from_side)), ((from_below)), overwatch widowmaker, purple skin, 3D render, cgi, photo portrait, closeup photo portrait, full body photo portrait, shiny pink jumpsuit with combat armor on shoulder, armor stripes on face, long purple ponytail, long purple braided ponytail, black cybernetic visor helmet with red dots on head, light smile, angry expression, natural expression, pleased expression, gun

• Therefore, if you want a canonically accurate Widowmaker, you final prompt might look something like this:

high quality photo portrait of overwatch widowmaker, purple skin, shiny pink jumpsuit with combat armor on shoulder, armor stripes on face, black cybernetic visor helmet with red dots on head, angry expression, holding big gun <lora:Overwatch - Widowmaker:0.4>

---

[Advanced Tips]: Once you are familiar with the LoRA, and if you are looking to create the most true-to-source-material character, go ahead an download yourself the Dreambooth model. Be aware, it is a full model, so it is 5gb+, which is one of the main reasons I don't recommend it for beginners. The model itself was trained on my own SenBlend 1.9. Therefore, be aware that this model will not excel at any style that SB1.9 is not proficient on. Check out the model page for details.

Using the dreambooth and the LoRA in combination should result in best possible outcome for accuracy, but it can be tricky. Personally, I recommend first creating your base prompt with your own model of choice with the LoRA only, and for anything beyond text2image (like inpainting) use the model in conjunction with a small amount of the LoRA.

The trigger word for the dreambooth model is: sbwidowmakerxx person. So using the example above, the prompt you would use with the model loaded, for example, in img2img, would be:

high quality photo portrait of sbwidowmakerxx person, purple skin, shiny pink jumpsuit with combat armor on shoulder, armor stripes on face, black cybernetic visor helmet with red dots on head, angry expression, holding big gun <lora:Overwatch - Widowmaker:0.15>

If you are using this model and posting your art elsewhere, please at least link them to this civitai page so other people can use it too. Happy diffusing and stay posted for more characters!

描述:

训练词语: see description