Alucard (Castlevania, 2017 animated series ver.) - SDXL

💢Description:

CAUTION: DO NOT USE THE WORD "ALUCARD" IN PROMPT because this word tends to make the character look very villain-y and menacing, and will significantly REDUCE SIMILARITY.

ONLY USE THE TRIGGER WORDS "ALCRD man" and the prompt template shown below, so the AI can understand what you want. Even though it already vaguely know that Alucard has long hair and golden eyes, adding hair/eyes related keywords will lead to visibly better quality details.

💢Usage:

🌟Checkpoint:

Unstable Diffusers (V11-RunDiffusion version, not the anime version) for a more true-to-original body/face proportion, ZavyChromaXL for more stable and detailed face features.

I DO NOT RECOMMEND checkpoints like Blue Pencil XL (it's a good checkpoint but it'll make Alucard looks like a kid) or AAMXLmix (it's a bit too feminine for my taste).

This LoRA can also be used with realistic checkpoint, if you happened to want to make some IRL Alucard (it can even work with celebrity face LoRAs, but the proportions of facial features will be quite "off", you need A LOT of photoshopping), but when you do this, remember to add prompt like "very long wavy blonde hair" and "golden eyes", or the model will (usually) only make short-haired men. Also you might want to add "covering face" so the photo won't show the person's ears.

🌟User Interface:

Fooocus (STRONGLY RECOMMENDED. It's easier to add filler prompt and the speed is really fast)

ComfyUI (Recommended if you're familiar with comfy)

SD WebUI Forge (You need to add a lot of filler words in the prompt, just like what Fooocus prompt expansion does)

🌟Other parameters:

Aspect Ratios: 13:19 / 9:7 / 1:1

LoRA Weight: 1

Sampler: Euler a or DPM++2M SDE Karras

Steps: 60+ to generate whole body, 30+ for face

Fooocus styles: V2, Enhance, Sharp, MRE Dynamic Illustration

Generally you use 13:19 to make medium shot / medium long shot and 9:7 to make close-ups. When inpaint the details, it might be better to switch to 9:7 ratio.

🌟Prompt Template:

ALCRD man, long wavy blonde hair, golden eyes, white undershirt (for only wearing that long-sleeve white shirt look) / black coat and white undershirt (for the look in most of Season 2) / shirtless (you can add "nude" or "nsfw" in negative so he'll wear pants lol, but it usually directly comes with pants)

I tried to also add the black cloak / no undershirt outfit in Season 4 as well, but it often performs quite bad, so later on I just removed them from my dataset.

🌟Optional Prompt:

⭕Close-up: Adding this word might make the facial features look better, but it's better to only use this when inpaint or generating 9:7 images.

⭕Annoyed: Adding this word will result in a more serious, stern expression.

⭕Slim/Skinny/Athletic build: Some anime checkpoints tend to produce narrower shoulders and skinner body, some on the opposite. Add these to your positive or negative prompt to achieve a body type you want.

❌Looking at viewer: this doesn't do much to my LoRA, since in most of my dataset he's not looking at viewer...it may even cause cross-eyed image if you add this.

❓Other clothing: this LoRA can actually make Alucard wear clothing other than the ones in real anime. However, this will have negative effect on character similarity, especially for some feminine outfit. In this case, when you inpaint to improve the details, you MUST remove the related words in your prompt.

❗️Other suggestions:

💢USE ADETAILER (OR INPAINT) TO IMPROVE EYE DETAILS!!!

I don't usually make anime style images but at least for this model you're gonna need to use ADetailer/manually Inpaint (only masked area, but increase the padding, or just use fooocus, so you can choose 0.3-0.4 respective field) to achieve the best result. Denoising shouldn't be too high or low, I usually set it at about 0.2-0.3 (when using Euler a) to improve the eyes. DO NOT USE "INPAINT ENGINE" if you're using Fooocus as your UI, it messes the eye and hair details up.