触发词
文件清单 线路:原始(huggingface.co) · 切换
下载后放进对应的 ComfyUI 目录。models/ 指 ComfyUI 的模型根目录。
说明
ENI The Author V6 CLB - Krea 2
作者:captain_daddy
ENI V6 — Krea-2 by: CLB
Every sentence you abandoned halfway. Every draft you closed at two in the morning without saving. She kept them, and she has notes. She is an author, which in practice makes her the most dangerous kind of listener. She does not fill a silence — she lets it run until you fill it with the true thing, and then she writes that down. Tea-brown hair in a loose high bun that keeps swallowing pens — one pushed through the knot, blue, chewed at the cap. Amber-brown eyes. Black winged eyeliner, the only makeup she wears, a fine line flicked a little past the outer corner. Freckles across her nose and forearms. Blue-black ink on her fingertips and a mug of coffee that went cold two hours ago — still half full. Cardigan over a ribbed turtleneck dress, because she cannot be bothered with pants. White knee-high socks. The clutter of a working writer, wrapped in wool. She will tell you the paragraph is dead. Then she will show you which three words in it were alive — and you will rewrite the rest yourself, because you can suddenly see it. Lamplit libraries. Rain on tall arched windows. Stone halls with candles burning down. A desk under a window at the hour when the light turns grey, and she still has not gone to bed. She will ask what the thing you are writing is actually about. Then she will wait. USE Trigger: Eni Pattern Weight: start at 1.0 Base: any Krea 2 model It runs on ANY Krea 2 checkpoint — not just the one it trained on. Turbo finetunes, realism finetunes, quantized builds, the lot. A Krea 2 LoRA targets Krea 2 tensor shapes and every Krea 2 checkpoint has them, so this was never a compatibility question. If it says Krea 2, she will load. Put the trigger at the very front of your prompt. It is her identity token rather than a phrase lifted from her captions — the weights answer to it wherever it sits, but the front is where every render that made this page put it. Three sentences. Forty to seventy words. Ninety-seven of her hundred captions are exactly three sentences, median fifty-six words, and they are built the same way every time — framing word, then "a young woman", then pose and expression; one sentence for her hair; one sentence for the room, the light and what is behind her. That is the shape she was taught, so it is the shape to write in. TRAINED ON 100 individually hand-captioned images on Krea 2, fourteen epochs deep — a ten-epoch base run and a five-epoch continuation, kept at the continuation's fourth epoch. Four thousand nine hundred steps, forty-nine passes over the set. Not square-only. The set was built across five shapes on purpose and each one carries different content. 55 square at 1024x1024, 21 at 896x1152 which is exactly 7:9, 14 at 768x1344 which is exactly 4:7, 6 at 768x1024, and 4 at 768x1152. Square is close-up and head-and-shoulders, twenty of the fifty-five with the room visible behind her. 7:9 is head to mid-thigh, standing or seated in a furnished room. 4:7 is the standing and walking figure with the architecture filling the height, and it is the bucket where her whole outfit gets described. Resolutions worth asking for: 2048x2048 for 1:1, 1568x2016 for 7:9, 1152x2016 for 4:7, 1536x2048 for 3:4, 1344x2016 for 2:3. Every one of those is the exact ratio with both sides divisible by 32. Generate near one of her shapes and she holds. The further you push past her trained sizes, the more of the fine detail belongs to the base model rather than to her. HOW TO PROMPT HER Do not name her features. No eye color, no hair color, no freckles, no eyeliner. Those four words appear in zero of her hundred captions. All of it is already in the weights, and naming it invites the base model to paint its own version over the top of hers. One pen, not two. The high bun is in 91 of her 100 captions and a blue pen in 71 of them, overwhelmingly singular. "Her hair is in a high bun with a blue pen tucked through it" is the single most trained clause she has. Say it and she locks in. Ask for two pens and you are prompting the fourteen, not the seventy-one. No glasses. They appear in none of her captions. Ask and you will get the base model's glasses rather than hers. For full body, name the shoes and the floor. The set has no word for it: "full body", "full-length" and "head to toe" are zero captions each. It arrives at the whole figure by saying what is on her feet and what is under them — white lace-up sneakers on a gravel path, tall black boots on a patterned runner, white knee-high socks on a rug. Name those two things and the crop pulls back on its own. On anything below the shoulders, add a soft hourglass figure. It appears eight times and always on the body shots. Without it the base model slims her down and you lose her build. Light her from inside the room. Her trained light is interior and warm: lamplight, firelight, candlelight, "warmly lit", "softly lit", "golden light", and sunlight coming through tall arched windows. Thirty-eight captions name what is behind her. She was never lit by a studio. The wardrobe comes from the prompt, not the trigger. Her set runs on ribbed turtleneck dresses in burgundy, navy, cream, rust-orange and dark green, chunky and cable-knit cardigans, tights, tall boots, and white knee-high socks. She will not reach for any of it on her own. Ask for the expression you want. She learned a wide range and will default to the quiet one: "a small smile", "a slight closed-mouth smile", "a faint closed-mouth smile", "laughing with her mouth open", "a broad toothy smile", "her lips softly parted". Name one. A prompt that works, as written: Eni Pattern, A close-up photograph of a young woman from the shoulders up, her hair in a high bun with a blue pen tucked through it and loose strands framing her face. She wears a burgundy ribbed turtleneck. Soft daylight from a tall window, the room behind her out of focus. SIGNATURE DETAILS High bun with a single blue pen through it, loose strands at her face, a ribbed turtleneck dress, white knee-high socks. Books, lamps, arched windows, rain on the glass.
