[LoRA] Métal Hurlant Comics - Moebius, Bilal, Druillet
模型简介：Métal Hurlant (literal translation: "Howling Metal," "Screaming Metal") is a French comics anthology of science fiction and horror comics stories. Originally created in 1974, the anthologies ceased publication in 1987 but revived between 2002 and 2004 in multilingual editions, and then again in 2020.
Update 17/07/23: There will not be another version released for 1.5, all my focus will be dedicated to SDXL. I will train a better version there, with a bigger dataset, higher resolutions, and comprehensive captioning.
⚠️ CAUTION - Alpha Release⚠️
This version is experimental. It can handle portraits quite well but struggles with elaborate compositions containing multiple subjects. I'm working on a more complete version but I felt it was already able to output some good images, so still worth sharing if you're interested.
💥Showcase images 💥
Please note that the showcase images have been upscaled with Multidiffusion (adds sharpness) and lightly edited with Lightroom (enhances colours). You should still be able to get a resembling result by using the prompts provided, but it will not be of the same quality.
How to use
▣ Tested at weight 1: <lora:methurlant:1>
▣ The trigger word is methurlant, additional words are intricate, atmospheric, surreal, gritty, cinematic, stylized, contrast, comics, eerie, cybernetic, futuristic, dystopian, etc.
▣ The resolution should be 512x512 or higher
▣ The best checkpoints are the creative ones (Deliberate, Lyriel, Dreamshaper, etc.)
▣ Works ok with realistic checkpoints (CyberRealistic, ConsistentFactor, RealisticVision, etc.)
▣ Does not work well with Anime checkpoints (AbyssOrangeMix, Anything, etc.)
▣ Good prompting is needed, please see the trigger words and the showcase images
▣ Works better with a colorful VAE
Dataset
▣ A total of 100 images from various Métal Hurlant artists: Bilal, Druillet, Moebius, etc.
▣ All images were curated to remove comics aspects and focus on the art only, with no text, no panels, no speech bubbles
▣ Provides a generic French blend without too much emphasis on any artist
▣ 0 caption
▣ Clip Skip 1 training
~~~~~~~~~~~~~~~~~~~~~~~o(*￣︶￣*)o~~~~~~~~~~~~~~~~~~~~~~~
示例提示词和参数：
Prompt: methurlant, A portrait of a robot, <lora:methurlant:1>, (intricate, atmospheric, surreal, gritty, cinematic, stylized, contrast, comic, eerie, stylized, cybernetic, futuristic, dystopian), Cloudy, Art Nouveau,Door ajar,Gas masks, Desolation, (Simultaneous contrast:1.1), (Color Theory:1.1),(Frottage:1.1), (Reflected light:1.2), Fur, Cylinder, ultra detailed, intricate, oil on canvas, dry brush, (surrealism:1.1), (disturbing:1.1)
Negative prompt: (3D, Adorable, Anime, CGI, Cartoon, Cartoonish, Childish, Computer Generated, Manga, Rendered)
Steps: 20, CFG scale: 7, Sampler: DPM++ 2M Karras, Seed: 1511799189, Size: 960x1920, Model: dreamshaper_6BakedVae, Version: v1.3.1, Model hash: c249d7853b, "methurlant-000010: a6e9ebbee184", Denoising strength: 0, Ultimate SD upscale padding: 32, Ultimate SD upscale upscaler: None, Ultimate SD upscale mask_blur: 8, Ultimate SD upscale tile_width: 512, Ultimate SD upscale tile_height: 512
页:
[1]