Zhenzhen's AI Radar · MOSS Image · 2026-10-30 Morning Report
This issue features 22 new versions: Strongly Recommended 0 · Recommended 3 · Worth Watching 8 · Listed Only 11
Editor's Note This issue reviews 25 new versions. Top pick: MeMaMoss Flat Anime v2.0: a flat anime style LoRA trained on magazine illustrations, retrained in V2 with higher rank and training resolution. The author clearly specifies positive and negative prompt usage but does not provide trigger words or strength settings, hence rated A instead of S; also rated A are the Indian miniature painting style and the Japanese maternity mark concept LoRAs. KreaMossMix is a workflow linking MOSS and Krea 2 rather than new weights, listed as B; imitations of living artists and style transfers distilled from other models are downgraded to B, while thin character LoRAs and limited-time paid sliders are mostly downgraded to C.
Recommended (3)
[A] MeMaMoss Flat Anime Style - Moss · v2.0 [Moss-Base]
LoRA · Version Update · Author Fish788 · Published 10-30 07:46 · Downloads 23
Version 2 of the flat anime style LoRA: trained on Megami Magazine illustrations (with Japanese text manually removed by the author), aiming to enable MOSS to consistently produce 2D flat coloring without drifting into 3D or realistic styles; this version increases rank and training resolution, and provides a fixed set of positive and negative prompts.
Downgraded from S to A: The changes and prompt usage are clearly explained with 5 examples, but trigger words and strength settings are missing, and the training material consists of magazine illustrations with unspecified licensing.
Images
Reasons for Recommendation
- Version update: In V1, the author was dissatisfied with the 2D style collapsing when switching to new prompts; V2 was retrained with rank 128 and 2048 resolution (V1 used 1792), resulting in a file size approximately twice that of V1; the author claims significant quality improvement, though the style loss issue persisted at the time
- The author later identified quality tags as the main cause, particularly 'Masterpiece' overriding the LoRA, and thus established a fixed set of prompts: positive prompts should only include 'high quality', while negative prompts list 'best quality', 'photorealistic', 3d, 'film grain', 'depth of field', 'thick lineart', 'painterly', etc.
- Serves scenarios in MOSS text-to-image generation requiring a magazine illustration-style flat anime look; trained on Base with examples generated using SFT, the author states it can produce images from 2048×2048 up to 1792×1792 and below, personally using 1600×1984
- 68 training images; 5 visible examples; free, license allows image output, on-site rental, and rental; this version received 23 downloads in approximately 1 hours since release, with the entire model accumulating 745 downloads (indicating popularity only)
How to Use
- Load MeMaMoss V2.safetensors (567 MB · BF16) in the MOSS-Image SFT workflow (trained on Base); no trigger words
- In positive prompts, keep only 'high quality' for quality tags, do not write 'masterpiece'; copy the author's negative prompt list exactly (best quality, normal quality, low quality, worst quality, photorealistic, realistic, 3d, film grain, bad anatomy, watermark, blurry, depth of field, thick lineart, painterly, etc.)
- Resolution at the 1792–2048 level (author uses 1600×1984); LoRA strength not specified by the author (V1 notes suggest increasing strength if the image appears too 3D)
Risks and Limitations
- Training material consists of Megami Magazine illustrations from 2018–2026, with copyright held by the magazine and original artists; the author does not specify licensing
- LoRA strength and sampling parameters are not provided; the author notes in update logs that style loss may still occur when changing prompts, requiring fixed positive and negative prompts to suppress it
- File size is approximately 567 MB, double that of V1 (approx. 284 MB)
- 15 sensitive-level examples are hidden; the license does not allow merging or selling generated images
[A] Indian Miniature Painting Style · MOSS
LoRA · Port · Author vizsumit · Published 10-30 02:54 · Downloads 33
MOSS version of the Indian miniature painting (Mughal miniature) style LoRA: the author describes this style as known for intricate miniatures and grand architecture, blending Persian, Indian, and later European influences; the MOSS version changes the trigger word to 'miniature painting'.
Maintained at A: Clear subject matter, 7 examples, and the most permissive license; missing are strength settings and MOSS-specific parameters.
Images
Reasons for Recommendation
- Port: Previously available as SDXL 1.0 and Illustrious versions (trigger word 'mughal_art'); the MOSS version has an empty description, only showing that the trigger word was changed to 'miniature painting', with a smaller file size than older versions
- Serves users wanting to create Mughal/Indian miniature-style illustrations, court figures, and architectural subjects in MOSS text-to-image generation; this is a historical art style, not an imitation of a living artist
- 7 visible examples; free, with a permissive license: allows image output, rental, sales, selling merged models, and attribution is optional
- This version received 33 downloads and 10 likes in approximately 6 hours since release, with the entire model accumulating 182 downloads (indicating popularity only)
How to Use
- Load indian_miniature_painting_moss_v1.safetensors (123 MB · BF16) in the MOSS workflow, and include 'miniature painting' in the prompt
- Strength, sampler, and steps are not provided by the author; unclear whether trained on Base or SFT
Risks and Limitations
- The MOSS version lacks any specific documentation; strength and parameters must be determined through experimentation
- The model card begins with an advertisement for the author's prompt website, followed by only two sentences describing the style
- 1 sensitive-level examples are hidden
Trigger Words: miniature painting
[A] Maternity Mark (Moss-image) · v1.0
LoRA · New Model · Author cannabis · Published 10-29 21:28 · Downloads 45
Concept LoRA for the Japanese 'Maternity Mark' (a pink heart-shaped official icon with blue outlines): used to generate keychains, stickers, and badges featuring this mark; the author states that MOSS's Japanese capabilities can be leveraged to add the text 'Baby in my belly' to the mark.
Maintained at A: Trigger phrase, optional words, and strength are complete, with an explanation of why the trigger word was chosen; however, the use case is very narrow, and the license does not allow commercial image output.
Images
Reasons for Recommendation
- New model: MOSS-specific object/icon concept LoRA, serving scenarios in merchandise design and illustrations where characters wear the maternity mark, requiring accurate reproduction of this specific icon
- The author provides complete usage instructions: trigger phrase, optional prompts (Japanese text, keychain charm, badge, sticker), and strength around 1
- The author explains the reason for changing the trigger word: initially using 'm4ternity mark', the text encoder recognized it as 'maternity', often enlarging characters' bellies, so it was changed to 'mtnt-mark'
- 3 visible examples; free; this version received 45 downloads and 13 likes in approximately 12 hours since release (indicating popularity only)
How to Use
- Load maternity_mark_moss_v1-000070.safetensors (142 MB · BF16) in the MOSS workflow, with strength around 1
- Include the trigger phrase 'mtnt-mark, featuring a pink heart-shaped design of a simple illustration with blue outlines,' in the prompt, then add keychain charm / badge / sticker as needed
- To add text, write 'Japanese text on the mark reading "Baby in my belly"'; the author warns that if the mark is drawn too small, the text may become blurry or misplaced
Risks and Limitations
- Very narrow use case, targeting only this specific mark
- The license only allows on-site rental, not commercial image output or selling generated images; caution is needed for actual merchandise production
- Only 3 visible examples, with 5 sensitive-level examples hidden
- The maternity mark is an official Japanese icon; its usage regulations must be checked separately before commercial use
Trigger Words: mtnt-mark, featuring a pink heart-shaped design of a simple illustration with blue outlines,, Japanese text on the mark reading "おなかに赤ちゃんがいます",, keychain charm,, badge,, sticker
Worth Watching (8)
[B] [style] hellbert · Moss e08
LoRA · Port · Author artyclaw · Published 10-30 05:43 · Downloads 28
MOSS version of the hellbert style LoRA: the author describes it as a mix of anime and American comic styles, with heavy shadows, rich details, dynamic poses, and elaborate backgrounds; training images were generated using Artemis v1.2 with style prompts like 'by dilbert' and 'by helltaker'.
Images
[B] WLOP_Illustrations · MOSS v1.0
LoRA · Port · Author speach1sdef178 · Published 10-30 04:18 · Downloads 57
MOSS version of the illustration style LoRA for artist WLOP (Wang Ling): trigger word 'wlop', with the author providing a weight range.
Images
[B] dora style · v0.9 for moss base
LoRA · Port · Author ibara0608 · Published 10-30 01:32 · Downloads 40
MOSS Base version of the dora style LoRA: the author describes its characteristics as grayish-black tones, slightly rough lines, mixed with film grain and chromatic aberration; the style was distilled from another style LoRA (wangxi205 style).
Images
[B] DuCreux_Style · Ducreux-MOSS
LoRA · Port · Author pope_phred · Published 10-29 22:33 · Downloads 48
MOSS version of the style LoRA for French classical painter Joseph Ducreux: the author describes his style as quite realistic for its time, focusing on various exaggerated facial expressions (notably his self-portrait making a funny face); trigger word 'Ducreux'.
Images
[B] MossAnime · V1
LoRA · New Model · Author Adel_AI · Published 10-30 01:09 · Downloads 9 · Paid
An anime-style LoRA named MossAnime, with the trigger word "Bradhamel art style"; the author only provided the sampler and weights, and it is a paid download.
Image
[B] 70s Horror Movie · MOSS
LoRA · Port · Author ArsMachina · Published 10-29 21:51 · Downloads 5 · Paid
The MOSS version of a LoRA for 70s horror movie stills: the author claims it specifically recreates the atmosphere of horror films from that era, generating images styled as "movie stills"; the trigger word is ArsMovieStill, with separate files provided for MOSS Base and SFT, available as a paid download.
Image
[B] KreaMossMix Workflow · v1.0
Workflow · New Model · Author NicoGJ · Published 10-30 08:55 · Downloads 3
A ComfyUI workflow, not new weights: it runs MOSS-Image and Krea 2 simultaneously from a single prompt, generating up to four images at once—pure MOSS, pure Krea, MOSS composition + Krea finishing, and Krea composition + MOSS repaint—with sliders to adjust the contribution ratio of each model.
Image
[B] [LuisaP❤️] MOSS - UNMODEL SLIDER · MOSS
LoRA · New Model · Author luisa_caotica · Published 10-30 07:16 · Downloads 58
A slider LoRA to remove the "doll face" effect: the author claims it specifically suppresses the unnatural, fake-doll-like faces in generated portraits, invoked using negative strength (e.g., -2, -3), with no trigger word.
Image
Included Only (11)
- ✦ Z-Cast: Xeimi S. [OC] · Moss — LoRA: A MOSS version character LoRA for the original character Xeimi S. (a petite Asian woman with black hair, set as an "international superstar"), with the trigger word Xeimi.
- Niji Oil Painting style · v1.0 — LoRA: An oil painting-style digital art LoRA named Niji Oil Painting, with the trigger word nijioil; the model card contains only the phrase "Just something simple."
- Chat GPT White Dragon Girl-White Dragon Girl Chat GPT · Moss-Young White Dragon Girl — LoRA: The MOSS version of the "Young White Dragon Girl" character LoRA: training images were produced by modifying GPT Image 2.5, totaling 30 images covering front, side, back, top-down, and low-angle views; the trigger word is Young White Dragon Girl.
- Samantha NordhavenLove · MossImage — LoRA: A MOSS character LoRA for the fictional character Samantha NordhavenLove (set as a maritime archivist on the west coast of Sweden, often wearing a wool skirt and glasses); the model card consists entirely of character backstory, with no trigger word or parameters.
- Hollow Out Dress (MOSS) · v1.0 — LoRA: A hollow-out mesh dress clothing LoRA: using the trigger word "wearing edgHollowOut_(fishnet dress)" puts a hole-filled mesh skirt on the character, with training prompts pairing it with a tube top and shorts.
- Shiny Umbreon (Illustrious/moss) · v1.0 Moss — LoRA: The MOSS version of a character LoRA for the Pokémon Shiny Umbreon; the trigger word is "Shiny Umbreon," and the author notes that the MOSS version may require adding "black fur" to the prompt.
- urotsuki-moss · v1.0 — LoRA: A LoRA named urotsuki, tagged as an art style; the author only wrote "no trigger word, trained on moss-base-int8," without specifying whether it is a character or art style, or what it looks like.
- [Moss] Umbreon (The regular kind with yellow rings) · v1.0 — LoRA: An anthropomorphic character LoRA for the Pokémon Umbreon (regular yellow ring color scheme), the author's first MOSS character LoRA; the trigger word is "Umbreon," and the author states the output tends toward sensitive levels, requiring well-crafted negative prompts.
- Moss_Turbo_with_Lora · v1.0 — Workflow: A ComfyUI workflow, not new weights: it adds LoRA loading and an acceleration switch to the common MOSS text-to-image base workflow, allowing users to quickly test with acceleration mode first, then disable it for final renders.
- Frieren: Beyond Journey's End Style · v1.0 — LoRA: A LoRA mimicking the manga art style of "Frieren: Beyond Journey's End"; the model card contains only the words "Manga style," with no trigger word or parameters, and it is a paid download.
- Abs Slider · Moss-image — LoRA: The MOSS version of a slider LoRA for adjusting abdominal muscle definition, with no trigger word; higher strength results in more defined abs; it is a paid download, and the MOSS version has no specific documentation.
Data: Civitai public API (base model MOSS Image), inclusion window 10-29 20:31 → 10-30 09:01 (Beijing Time). Recommendations have been reviewed by editors; download counts reflect attention only.