An anime illustration style LoRA for MOSS-Image: larger eyes with heavy lashes, a soft blush under the eyes, glossy hair and warm, deep shading.How to prompt. Start the prompt with <code>d3style.</code> and name the medium in a sentence, for example <code>d3style. Anime-style digital illustration of ...</code>. MOSS-Image renders a photograph when no sentence says how the picture is drawn, and this LoRA does not change that: the trigger alone keeps the result photographic.Strength. At 0.6 the picture keeps the base model's outfit and figure and takes on the face and shading. At 1.0 the full look comes through, and figures also turn curvier. Composition stays with the seed at every strength.Tested with ComfyUI 0.42.1, MOSS-Image SFT (int8 ConvRot DiT), the Qwen3.5 9B text encoder, 25 steps, CFG 2.5, er_sde / simple, shift 5, 832x1248, loaded with LoraLoaderModelOnly. Trained on MOSS-Image-Base with ai-toolkit; it works on SFT.Training. 40 anime illustrations, rank 32, 2000 steps, captions that describe the content with the trigger first.

[MOSS Image] d3style anime illustration style

First release: the step 2000 checkpoint. Start the prompt with <code>d3style.</code> and name the medium in a sentence. Strength 0.6 keeps the base model's outfit and figure; 1.0 gives the full look.

Files with SHA256: dd013537b18bd37ddf587134bd55051712cbfaf62f2d82ecdd453fa02c227d95