Zhenzhen's AI Radar · MOSS Image · 2026-11-18 Morning Report
This issue features 34 new versions: Strongly Recommended 1 · Recommended 7 · Worth Watching 7 · Listed Only 19
Editor's Note This issue reviews 35 new MOSS versions, with 1 rated S, 7 rated A, 7 rated B, and 20 rated C; none were excluded. Top pick: MyGO / Ave Mujica anime screenshot style LoRA: produces cartoon-shaded 3DCG images with pastel color grading and colored outlines that look like frames from an animation. Trigger prefixes, strength, steps, CFG, sampling, and prompt templates are fully documented, and all 10 example images are visible, but the author restricts usage to non-commercial purposes. Skin tone slider, cel-shaded style, Victorian albumen prints, black-and-white charcoal comics, gamebook hatching illustrations, dark fairytale style, and the RDBT integrated base model are rated A; the refusal-removed Qwen3.5-9B text model is merely a third-party quantized conversion and is rated C; standard copyrighted characters and style LoRAs with only one or two sentences of description are rated C.
Strongly Recommended (1)
[S] [MOSS Image] It's MyGO!!!!! / Ave Mujica - anime screenshot style · v1
LoRA · New Model · Author lepetitprince716355 · Published 11-18 01:46 · Downloads 29
BanG Dream! It's MyGO!!!!! / Ave Mujica anime screenshot style LoRA: features cartoon-shaded 3DCG character textures, pastel color grading, colored outlines, soft shadow edges, and ambient lighting, producing images that look like frames captured from an animation.
Suitable for running a single image exactly according to the author's template, then adding 3d render to the negative prompt for comparison, to see how obvious the conflict described by the author is.
Images

Reasons for Recommendation
- New model; serves text-to-image generation for character images with an anime screenshot feel, fan art scenes, and widescreen compositions. The author recommends a canvas size of 1360x768 or 1216x832.
- Fully documented: trigger prefix, strength, steps, CFG, sampler/scheduler, shift, recommended canvas, tested base models (moss_image_sft_bf16 and int8_convrot versions), and a directly copyable prompt template are all provided.
- Pitfalls regarding negative prompts and lighting are noted: 3d render should not be included in the negative prompt (as it conflicts with the cartoon-shaded 3DCG style); even, soft frontal or slightly side lighting is most stable, while backlit night scenes with high contrast weaken this color grading.
- All 10 example images are visible and non-sensitive; free; training parameters are public (3000 steps, 15 epochs, rank 32).
How to Use
- Start the prompt with @mygo_style. Anime screenshot., then describe the character and scene in one natural language sentence (author's template: a woman in her twenties holding an electric bass standing in a warm-lit wooden rehearsal room, three-quarter profile, even lighting).
- Strength 1.0; 25 steps, CFG 2.5, er_sde + simple, shift 5.0; canvas 1360x768 or 1216x832.
- Remove 3d render from the negative prompt; if you do not want this color grading, simply unload the LoRA, as the author states that loading it globally changes the palette, outline colors, and color grading.
Risks and Limitations
- Based on the style of a copyrighted anime (produced by Sanzigen / Bushiroad); the author explicitly states in the model card that it is unofficial and non-commercial, and neither generated images nor weights may be sold or used commercially; the Civitai license section only checks for site rental.
- Approximately 7 hours since release with 29 downloads, so user feedback is still scarce; it was trained only on style, not characters, so specific characters require a separate character LoRA or prompt engineering.
Trigger Words: @mygo_style
Recommended (7)
[A] RDBT | MOSS Cauldron · v0.4 (Kromoss DMD)
Base Model / Checkpoint · New Model · Author reakaakasky · Published 11-17 23:59 · Downloads 79
A ComfyUI int8 integrated base model exported after pre-fusing the RDBT anime style LoRA and DMD acceleration LoRA into lodestones' Kromoss (OPD-priming) base model, capable of directly generating anime illustrations with 8~12 steps and CFG 1.
Images

Reasons for Recommendation
- New model; solves the hassle of having to dequantize and requantize twice when stacking LoRAs on int8 base models: the author first fused the style LoRA and DMD acceleration LoRA on bf16, then converted it to int8 using comfy-kitchen, ready for immediate use upon download.
- Serves quick generation of anime illustrations; the author states that v0.4's RDBT was trained specifically for the OPD base model and further strengthened the default anime style with a batch of official character art data (including many Azur Lane samples).
- Clear usage instructions: trigger phrase "Anime-style illustration of ...", 8~12 steps, CFG 1, euler; the recipe (base model file, DMD LoRA filename) is public, and those who already have a bf16 base model can merge it themselves.
- 8 visible example images; free; 4 comments.
How to Use
- Load the int8 file directly as a base model (rdbt_v0.4_kromoss_OPD-priming_dmd_int8convrot); since DMD is already fused, there is no need to attach an acceleration LoRA.
- Start the prompt with "Anime-style illustration of ..."; 8~12 steps, CFG 1, euler sampler; for other details, the author refers users to the Kromoss model page.
Risks and Limitations
- The author labels this as a beta version: the test LoRA was trained on a test base model, so the standalone RDBT v0.4 LoRA file is not released; the author notes that human figures may appear fuller.
- The license section does not allow derivatives; the author explicitly prohibits sharing fusions made with RDBT and states that special trigger words and latent space watermarks are embedded.
- Only one int8 file is available; 1 sensitive-level example images are hidden.
[A] Dizzle's Charcoal Comic Book (A Dabble Blend) · v1
LoRA · New Model · Author Maridizzle · Published 11-18 03:20 · Downloads 33
Black-and-white charcoal comic style LoRA: features heavy ink charcoal, hard shadows, and cross-hatching, resembling a page torn from an old comic book or sketchbook. Subjects such as ruins, machinery, and monsters are rendered in this gloomy black-and-white sketch style.
Images

Reasons for Recommendation
- New model; serves black-and-white comic storyboards, horror/apocalyptic theme illustrations, and sketch-style concept art; the author claims that any subject matter will be converted into this style.
- Parameters are complete: trigger word dz_sketchil, accompanying magic phrase "a grim black and white charcoal sketch", strength 0.9-1.0, sampler Euler flowmatch, CFG 3, 30-50 steps, with instructions to place the trigger word and magic phrase at the very beginning of the prompt.
- Free; the license section allows image generation and rental.
How to Use
- Start the prompt with "dz_sketchil, a grim black and white charcoal sketch", then describe the scene normally.
- Strength 0.9-1.0; Euler flowmatch, CFG 3, 30-50 steps.
Risks and Limitations
- Only 3 example images are visible (another 2 sensitive-level images are hidden), providing little evidence of stability across different subjects.
- The model card contains only one paragraph, with no mention of training data sources or known issues; approximately 6 hours since release with 33 downloads.
Trigger Words: dz_sketchil
[A] Midnight Fairytale · v1.0 Moss
LoRA · Port · Author Daalis · Published 11-17 23:22 · Downloads 50
MOSS port of the Midnight Fairytale dark fairytale style LoRA: Gothic, slightly eerie fairytale illustrations with deep blue-purple-black shadows and bright highlights, featuring an anime-leaning painting texture. Common subjects include haunted princesses, dangerous elves, and cute yet weird little monsters.
Images

Reasons for Recommendation
- Port: The same page already features versions for Illustrious, Flux.2 Klein 9B, Z Image Turbo, Anima, and Krea 2; the MOSS version's release notes state it closely matches the training set, works well at strength 1, and shows impact at 0.5.
- Serves Gothic fantasy portraits, dark fairytale characters, creature/monster designs, and scene illustrations such as moonlit forests and castles; the model card provides complete example prompts and suggested negative prompts.
- Strength tiers are provided: 0.5-0.8 adds only a touch of atmosphere while preserving the base model's original look, 1.0 is the starting point, and 1.2-1.5 is darker with heavier blue-purple-black tones and a stronger fairytale feel.
- 4 visible example images; free; the entire model page has 1108 downloads and 112 likes (mostly from other base model versions, indicating only general attention).
How to Use
- Start the prompt with MNFairyDaal (the trigger word provided by the author in the description), followed by descriptions of characters and scenes.
- Start strength at 1.0; use 0.5-0.8 for a lighter effect, and 1.2-1.5 for darker and heavier tones; steps, CFG, and sampler are not specified by the author.
Risks and Limitations
- The trigger word MNFairyDaal is only written in the description and is not registered in the MOSS version's trigger word field.
- The strength tiers are general instructions across base models; the MOSS version has only one specific sentence of instruction; 1 sensitive-level example images are hidden.
- The license section only allows image generation and site rental.
[A] MOSS Cel shaded Anime Style · v1.0
LoRA · New Model · Author baowy · Published 11-18 00:07 · Downloads 91
Japanese cel-shaded anime style LoRA: transforms characters in realistic or semi-realistic images into clean cel-shaded 2D anime characters, while keeping the background photorealistic, creating an effect of "anime characters standing in a real-world scene."
Images

Reasons for Recommendation
- New model: The author explicitly states that it only animates characters while preserving the photographic quality of the background, suitable for compositing 2D characters into real scenes, differing from style LoRAs that convert the entire image to anime.
- Serves mixed character + real scene styles in text-to-image generation, anime character portraits, and posters; by the same author as this issue's skin tone slider, who notes that the slider primarily adjusts warm/cool tones on the cel-shaded style.
- Free, license allows selling generated images and fusions; approximately 9 hours since release with 91 downloads, making it the most downloaded item in this group (download count indicates only attention).
How to Use
- Load the LoRA; the author recommends a weight of 1.0; the author has not registered a trigger word.
- Steps, CFG, and sampler are not specified by the author.
Risks and Limitations
- Only 1 visible example images (another 1 sensitive-level images are hidden), making it difficult to assess whether the separation effect between characters and background is stable.
- Example image metadata shows er_sde, CFG 1, 10 steps; the author does not specify if an acceleration LoRA was attached, so confirm before copying settings.
[A] Dizzle's Victorian Albumen Prints · v1
LoRA · New Model · Author Maridizzle · Published 11-18 04:14 · Downloads 26
A LoRA for Victorian-era albumen print effects: warm sepia tones, arched card borders, curled and damaged paper edges, mold spots, and embossed studio stamps at the bottom make any subject look like a photograph from the 19th century.
Images

Reasons to Recommend
- New model: The author lists specific object characteristics of old photos—golden-brown tones, arched cut mounts, curled corners, foxing spots, and embossed studio stamps at the bottom—not just adding a brown filter.
- Suitable for aging services, antique-style photos, historical theme posters, and prop images; the author states it works for portraits, landscapes, and absurd themes.
- Parameters are among the most comprehensive in this group: trigger words, auxiliary prompts, weights, samplers, CFG, and steps are all provided; free.
How to Use
- Trigger word dz_victorian; the author suggests adding a sentence like "a sepia-toned vintage photograph with arched border".
- Weight 0.9 to 1.0; Euler flowmatch sampling, CFG 3, 30 to 50 steps.
Risks and Limitations
- The license does not allow selling generated images (only output and rental are checked); published less than 5 hours ago, no likes yet.
Trigger Word: dz_victorian
【A】MOSS skin-tone slider · v1.0
LoRA · New Model · Author baowy · Published 11-17 23:11 · Downloads 75
A MOSS skin-tone slider LoRA: adjusting the weight from negative to positive shifts character skin tones from dark, warm, and saturated to fair, cool, and pale; the author claims it does not alter facial features, hair, clothing, or background.
Suitable for running a triptych with −3 / 0 / +3 using the same prompt, with one set for realistic styles and one for anime styles for comparison.
Images

Reasons to Recommend
- New model: Turns "changing only skin tone" into a slider that can go positive or negative, without needing to change prompts or swap base portrait models; the author provides weight comparisons: −3 for dark and warm, 0 for neutral, +3 for fair and cool.
- Serves skin tone fine-tuning in text-to-image portraits, usable for both realistic and anime styles; the author notes different behaviors—realistic mainly changes depth (tanning ↔ whitening), while anime/flat colors mainly change temperature (ruddy ↔ pale).
- Parameters are clearly provided: usable weight range, side effects when exceeded, and recommended accompanying acceleration LoRAs are all listed; file size is only about 12 MB; free, license allows selling generated images and merges; published about 10 hours ago with 75 downloads (download count only indicates attention).
How to Use
- Load the LoRA and adjust the weight between −3 and +3: negative values make skin darker and warmer, positive values make it whiter and cooler; no trigger word required (the author did not register a trigger word).
- The author recommends using it with a 8-step DMD acceleration LoRA (strength 1.0); sampler and CFG were not provided by the author.
Risks and Limitations
- The author claims faces begin to distort beyond ±3; "does not alter facial features or background" is the author's claim.
- Only 1 visible examples, with another 1 sensitive-level examples hidden; evidence for stability across skin tones and art styles is limited.
【A】Gamebook illustration [Krea2 + MOSS + Flux1/2 + ZIT + IL] · MOSS
LoRA · Ported · Author ColonelJay · Published 11-18 09:00 · Downloads 1
A MOSS port of a LoRA for 1980s fantasy gamebook illustration styles (Fighting Fantasy, Lone Wolf, etc.): black-and-white pen line art, dense cross-hatching, high contrast, and rough, full compositions.
Images

Reasons to Recommend
- Ported: The same model already has versions for Krea 2, Flux.1, Flux.2 Klein 4B, ZIT, and Illustrious; the MOSS version retains the trigger word CYOA and registers an additional descriptive phrase "traditional comic-style black-and-white line art, thick lines, high contrast, hatching"; the author did not write about the effect differences between the MOSS version and other versions.
- Serves black-and-white fantasy illustrations, tabletop RPG/board game assets, monsters, and battle scenes in retro gamebook styles; the author describes style features specifically (ink line hatching, dark dramatic lighting, tight close-up action compositions) and notes it can also produce color images.
- 11 visible examples; free, license allows selling generated images and merges; cumulative downloads for the entire model are 1853 (download count only indicates attention).
How to Use
- Start prompts with CYOA, optionally followed by the full registered sentence, e.g., "CYOA, a black and white, monochrome illustration in the style of old fantasy gamebooks and vintage fantasy role-playing games."
- The strength range 0.7 to 0.8 provided by the author is for the Flux and Illustrious versions; strength, steps, and CFG for the MOSS version were not provided by the author, but you can start testing from this range.
Risks and Limitations
- The author did not write update notes or parameters for the MOSS version; the filename is epoch-09, and the author did not explain why this epoch was chosen.
- Published less than 1 hours ago, with almost no download feedback; the page contains mixed content, and 9 sensitive-level examples are hidden.
- The style benchmarks against multiple real illustrators; be aware of art style controversies for commercial use.
Trigger Words: CYOA, CYOA, a black and white, monochrome illustration in the style of old fantasy gamebooks and vintage fantasy role-playing games., CYOA, this image is a black and white drawing in a traditional comic book style, characterized by bold lines, stark contrasts and hatching style.
Worth Watching (7)
【B】[Moss] Mystic HHH · Style A
LoRA · New Model · Author alcaitiff · Published 11-18 03:59 · Downloads 82
Style A, the first in the Mystic HHH series art style LoRAs: the author plans to release multiple styles sequentially by letter, sharing the trigger word HHH; MOSS's identity layer was excluded during training to facilitate stacking with character LoRAs.
Images

【B】Moss-Boredom · v2.0
Base Model / Checkpoint · Version Update · Author ukr8b3g201 · Published 11-18 02:27 · Downloads 87
v2.0 (int8-convrot) of the Moss-Boredom merged base model, a dreamy soft-light anime image generation base; the author recommends using it with a DMD acceleration LoRA for 8–10 step generation.
Images

【B】SamDoesArts - Sam Yang - Style MOSS LORA · v1.0
LoRA · New Model · Author Konan · Published 11-18 03:40 · Downloads 62
A LoRA for SamDoesArts (Sam Yang)'s digital painting style, including trigger prompts for the character Kara (white long hair, blue eyes, arm tattoo) frequently drawn by the author.
Images

【B】Fashion and Ad Photography · MOSS 1.0
LoRA · Ported · Author FuenfFingerFaultier · Published 11-18 03:00 · Downloads 53
A MOSS version of a fashion magazine and advertising photography style LoRA: the author retrained on MOSS using the same data (fashion photos from magazines and ads) as the Krea 2 / ZIT / Flux versions.
Images

【B】Painterly - CE · V01b - MossImage Base/SFT
LoRA · Ported · Author CreativeEdge · Published 11-18 03:17 · Downloads 36
A MOSS version of the Painterly artistic texture style LoRA: the author claims it has a strong artistic feel, especially leaning towards watercolor; example prompts are also used to generate thick oil painting brushstrokes.
Images

【B】Johnny The Homicidal Maniac · v1
LoRA · New Model · Author eichenwaldatelier461 · Published 11-18 08:12 · Downloads 7
A MOSS style LoRA for the art style of Jhonen Vasquez's comic "Johnny the Homicidal Maniac": visible examples are mostly full-page comics with rough black-and-white ink lines, handwritten text filling pages, and dialogue boxes, along with a few colored Halloween-themed illustrations.
Images

【B】Hips Shape Slider · v1.0
LoRA · New Model · Author TP_AI_63 · Published 11-17 21:40 · Downloads 3 · Paid
A MOSS hip width slider LoRA: higher intensity results in wider hips for clothed adult characters; the author claims it is only noticeable on tight jeans and back-view compositions.
Images

Included Only (19)
- MoriiMee - Gothic Niji Journey - Style MOSS LORA · v1.0 — LoRA: A MoriiMee-style Gothic Niji Journey art style LoRA. The author's auxiliary prompt is "polished semi-realistic 3D render style," leaning towards a Gothic character illustration with a semi-realistic 3D rendering feel.

- Shikarii - Semi-Realistic digital Pin-up Style - MOSS LORA · v1.0 — LoRA: A Shikarii semi-realistic digital pin-up art style LoRA. The author provided two sets of style enhancement prompts: one leaning towards 3D CGI rendering, and the other towards painterly realism with warm golden lighting.

- Bogna Harkusha - Natural Character LoRA · v1.0 — LoRA: A character LoRA for the fictional realistic person Bogna Harkusha: a woman in her early twenties with short brown bangs and Eastern European features. Defaults to no makeup, capable of generating both studio photography and amateur snapshot styles.

- zzzi_gn Style [Anima/Krea2/MOSS] · MOSS v1.0 — LoRA: A detailed manhua art style LoRA based on the artist zzzi_gn's style, ported from the Anima and Krea 2 versions to MOSS.

- Shokuhou Misaki [Toaru] (MOSS) · v1.0 — LoRA: A character LoRA for Shokuhou Misaki from "A Certain Scientific Railgun" / "A Certain Magical Index," with clothing tags grouped by school uniform, swimsuit, gym clothes, and bikini.

- Tonerico - Hitoner · Moss — LoRA: A character LoRA for Tonerico, a cat-like beastman character from the Shonen Jump+ manga "Hitoner" (by Tomo Iyoi), MOSS version.

- Kiyohime (Fate Grand Order) · Moss Image — LoRA: The MOSS version of a multi-outfit character LoRA for Kiyohime from "Fate/Grand Order," containing multiple outfits in one file, including First and Third Ascension, Lancer class ascension, Heroic Spirit Formal Dress, and Traveling Outfit.

- Olivier Ledroit style · MOSS — LoRA: The MOSS port of an art style LoRA based on artist Olivier Ledroit's style. The author provided no description, so the style characteristics can only be judged from the examples.

- Takehiko Inoue Watercolor · MOSS — LoRA: The MOSS port of a watercolor art style LoRA based on Takehiko Inoue's style. The author provided no description, so the style characteristics can only be judged from the examples.

- Luludiana (An Enchanted Mama's Saga) · v1.0 (moss) — LoRA: A character LoRA for Luludiana from the manga "An Enchanted Mama's Saga," MOSS version.

- Weiss Schnee from RWBY [Krea 2/MOSS Image] · MOSS — LoRA: A character LoRA for Weiss Schnee from "RWBY," MOSS version.

- Qwen3.5-9B-Heretic2.1 (Abliterated TE&PE for Moss Image, Qwen Image 2.1, Minimax H3 Image) · INT8 — Text Encoder: A text model converting the third-party refusal-removed version of Qwen3.5-9B (Heretic 2.1) to INT8 ConvRot. It can serve as a text encoder for MOSS Image, or as a prompt expansion model for Qwen Image 2.1 and MiniMax H3 Image.

- Moss LORA Stripper Platform Mules · v1.0 — Base Model / Checkpoint: A clothing LoRA for drawing platform mule heels. Straps can optionally be transparent PVC, leather, or patent leather, and it can also generate transparent heels. The page type is labeled as a base model, but the file is actually a LoRA.

- Faelynn / g41 / 芙铃 / GIRLS' FRONTLINE 2: EXILIUM · v1.0 moss — LoRA: A MOSS character LoRA for Faelynn from "Girls' Frontline 2: Exilium." Features blonde low twin-tails, fox ears, and heterochromia (red and blue eyes). The author's examples cover four outfits: default combat uniform, gym clothes, sailor uniform, and swimsuit.

- 少把盐 Moss Heal · v2.0 — LoRA: Version 2.0 of the "Shaobayan" MOSS Heal anime art style LoRA. The 2 visible examples are soft, semi-thick-paint illustrations of girls, one of which includes Chinese title typography.

- Nai-mix · v1.0 — LoRA: A thick-paint anime illustration art style LoRA. The author states it has weak line art, strong volumetric lighting, and leans towards an illustrative style.

- Qwen3.5-9B-Heretic2.1 (Abliterated TE&PE for Moss Image, Qwen Image 2.1, Minimax H3 Image) · W4A8 — Text Encoder: The W4A8 ConvRot quantized version of the same refusal-removed Qwen3.5-9B text model. It is smaller than the INT8 version and serves the same purpose: as a text encoder for MOSS Image or a prompt expansion model for Qwen Image 2.1 / H3 Image.

- matsuka (Gamerag) Style · v2.0 — LoRA: Version 2.0 of a MOSS style LoRA based on the artist matsuka's (Gamerag) art style. This version uses a different training dataset.

- FFstyle Anima/Moss Base/Moss Heal · v2.0 Moss Heal — LoRA: An anime style LoRA imitating the NAI5 output art style. This entry is version 2.0 of the Moss Heal branch. The author states that training on Base resulted in a plastic-like, unfaithful look, but it improved significantly when trained on Heal.

Data: Civitai public API (base model MOSS Image), collection window 11-17 20:31 → 11-18 09:01 (Beijing Time). Recommendation reasons have been reviewed by editors; download counts only reflect popularity.