moeblack/moss-image-prompt-skill
ResearchThis page was translated automatically. Technical details may differ from the original. View the original
MOSS-Image Phram Compilation Rulebook (Agent Skill format, CC BY-NC-ND 4.0)
Overview
Rules for the production of image tips for MOSS-Image (OpenMSS, Qwen3.5-9B text encoder + 5B single-flow DiT): three modes of writing, binary, cross-diverse, with multiscopy comics and JSON layout. A group of 12 groups covered the 1990s Hong Kong Film, Dune, Joker, Incaption, Titanic, etc. At the time of writing**, only §3.2 is used to describe the organization: initial optical and visible parameters (film model, lens, colour), the body writing space and protection, the light position of the lamp form and face, the age color and particles, and the daily prop to break the sense of precision. The same set of rules replaces the initial style phrases, and can be transposed between hard and straight, Rembrandt cutting, neon frames, and orange screens. The following are all generated by MOSS-Image-SFT + FKEY Style Lora, with the hints written by SKILL.md in this repository (sampling: euler + sgm uniform, Steps 50, CFG 3.0, Shift 5.0, LoRA 0.6). A rank 4, 18.66 MB, a very light-size ductor trained with Muon Optimizers: five officers and hairy pens, clothing and background with loose water chalk; low saturated peach colour, clean line, not locking each generation to the same role or clothing. Recommended strength 0.4 - 0.8 (from 0.6). If you want to switch to another painter style, you can just press SKILL.md §17.2 to replace the four-part skeletal trigger. All documents in this warehouse (SKILL.md, README.md, CHANGELOG.md, examples/) are available. Official Chinese Abstract: [T0] : Take it, no problem; don't change the hair, don't sell it. Would you like to change it? Open up Issue and say, the written permits are mostly granted。
Overview (machine translated)
The rule book for the production of image tips to [[T0] [OpenMOSS, Qwen3.5-9B text encoder + 5B single-stream DiT]: three modes of writing, binary, cross-diverse, with multiscopy comics and JSON layouts.
📌 Update the record
| Date |:-: |: |: |: | v2.0.0 | 2026-10-21 ** manual ** **: reordering of the entire volume to actual presentation process to harmonize evidence levels and terminology; extension of the binary windpath, four-part thick coating skeletons, full colour multiple lens Layout belong to the air bubble-talker; completion of the default parameters and self-check list. Zenium | v1.0.1 | 2026-10-18 | Tighten the first row of the solid base and negative pool and amend the §0 mode start description. Zenium v1.0.0 | 2026-10-18 | Initial open source, single file Skill format. Zenium
Zoo, the movie’s back
** Upline [[T1] is the original screenshot, downline [[T2] is the result of a repetition of this [[T3]. ** Group 12 covers Hong Kong films of the 1990s, Dune, Joker, Incaption, Titanic, etc.
At the time of writing,** do not write the name of the film,** do not write the name of the actor, only as described in § 3.2: initial optical and visible parameters (film model, lens, colour), body writing space and cover, light form and face luminous position, age-temperature and granules, and then add daily small objects that break a good sense. The same set of rules replaces the initial style phrases, and can be transposed between hard and straight, Rembrandt cutting, neon frames, and orange screens.
🎨 Example
The following are all generated by [[T4] + FKEY Style LoRA, with the introduction prepared by this repository SKILL.md [sampling: [(T7]] + [(T8], Steps][[T9], CFG [[T10], Shift][[T11], LoRA]0.6]).
Double-System, hair-systolic, bone-systolic, blonde and light-haired characters, white skirts, double-syllable lace, high-profile exposure. Film Four-Geek Narrative Pages + + + + + + + + + +
The inner monologue page, no tail bubble. Main image + local inset with character and utensils with object to avoid growing humans Head
Symmetrical double position + emoticon close-up two independent BBox pulls distance, innet only eyes and cheeks Six Germontage Tunetian “Close” wooden plates and seals, with a Quest on the waist.
# To the LoRA
The main examples of this set of rules come from a companion FKEY Style — Painterly Anime LoRA for MOSS-Image:
One rank 4, 18.
This page was translated automatically. Technical details may differ from the original.
Recommended Tools
Try a different keyword or remove a filter.
Install
npx skillfish add moeblack/moss-image-prompt-skill