我不確定目前是否依舊有人還會使用SD1.5的LORA
如果有需求 可留言 我會繼續更新
I'm not sure if anyone is still using SD1.5 LoRA nowadays.
If there's any demand, feel free to leave a comment — I'll continue updating it.
在此 推薦使用 Run七八造 Cartoon V3
我把系列模型融入 在使用LORA時 效果會更好 後續都會使用V3訓練Lora
H Anime / Hgame Anime Collection
Lora權重:0.7+
CFG scale CFG 7+
Steps 腳步 30+
Sampler 採樣器 Euler A + Karras
Clip skip 2
高分辨率修复 (Hires. fix)
512 x 768 768 x 1024
768 x 512 1024 x 512
放大算法 4x-UltraSharp
高分迭代步数 15+
重绘幅度 0.35+
如果有安裝LORA插件
可以查看訓練標籤 查看有哪些提示詞可以使用


二次元 提示詞起手:
8k, best quality, masterpiece, rule of thirds, superb, high resolution, sharp focus, extremely detailed description, professional, gorgeous and intricate details,
寫實 提示詞起手:
edge quality, perspective silhouette, 8k, best quality, masterpiece, extremely detailed, rule of thirds, photorealistic, superb, HDR, high resolution, sharp focus, photorealistic rendering, extremely detailed description, professional, gorgeous and intricate detail,
負提示詞:
an12,bad-hands-5,bad_pictures,bad_prompt_version2-neg,By bad artist -neg,FastNegativeV2,ng_deepnegative_v1_75t,bhands-neg,EasyNegative,(EasyNegative:1.2),bad-hands-5,
in-touch-with-surroundings,(iris deformation, pupil deformation, cgi,mosaic censoring, rendering, sketch, cartoon, painting, animation),(sketch, mutated hands and fingers:1.4),(deformation, distortion, disfigurement:1.3),text,cropping,out of frame,worst quality,low quality,jpeg artifacts,ugly,repetitive,sick,mutilated,extra fingers,mutated hands,poorly drawn hands,poorly drawn faces,mutation,distortion,blur,dehydration,poor anatomy,poor proportions,extra limbs,cloned faces,disfigurement,overall proportions,misshapen limbs,missing arms,missing legs,extra arms,extra legs,fused fingers,too many fingers,long neck,poorly drawn,bad anatomy,incorrect anatomy,extra limbs,missing limbs,floating limbs,severed limbs,mutation,mutation,ugly,disgusting,amputation,
Description
館熟女
女僕 小文
FAQ
Comments (5)
Work only with Run Cartoon V3 ?
這系列LORA是使用 Run Cartoon V3 進行訓練 所以出圖效果也比較好 您也可以使用其他SD1.5checkpoint 結果我就無法保證了 或許會更好或者變得微妙
@Ng_SowhaT If you train it with a specific checkpoint, it will not work with others. Now I understand why it doesn't work with 5 models I tried with. You should train with a standard model to make it work with other models.
I've gotten good results with a variety of checkpoints especially epicphotogasm_lastUnicorn and Dreamshaper8. A bit of trial and error with support embeds / LORAs, describing the general features (hair style , accessories and body type), and playing with different samplers helps.
SD1.5 checkpoint實在是太多了 理論上是兼容同類型動漫為主的checkpoint 至於run V3是融合了之前所訓練過的Lora 訓練出來的結果也比較接近原動畫風格









