Zhenzhen's AI Radar · MOSS Image · 2026-10-28 Evening Report
This issue features 42 new versions: Strongly Recommended 1 · Recommended 4 · Worth Watching 14 · Included Only 23
Editor's Note This issue reviewed 44 new MOSS versions, rated as S 1, A 4, B 14, C 24, with 1 excluded. Top pick: MOSS Pixel Art v0.3: A LoRA for pixel art / 2D game assets that outputs a strict 8x grid; shrinking the output by 8x yields native pixel art. Both standard and 8-step accelerated parameter sets are fully documented, maintaining an S rating. Other highlights: Vine's monochrome neon style from Ars Goetia, colorful line art concept style, Fran Bow-style adventure game backgrounds, and grotesque creatures Weirdies. The INT4 version of Realism By Stable Yogi is merely a quantized file and was downgraded from S to B; imitations of living artists' styles, ports with blank descriptions, and standard character LoRAs were mostly rated C; Cinematic DarkRays shares the same weight hash as Slavic Dark Fantasy by the same author and was excluded as a duplicate upload.
Strongly Recommended (1)
[S] MOSS Pixel Art · v0.3
LoRA · New Model · Author robl · Released 10-28 18:57 · Downloads 49
A LoRA for pixel art / 2D game assets that outputs a strict 8x pixel grid: every 8×8 output pixel corresponds to one logical pixel; shrinking the image by 8x after generation yields a native-resolution pixel art image; trigger word pxl8x.
Maintaining S: Specific capabilities (fixed 8x grid, scalable back to native pixels), fully documented standard and accelerated parameters, numerous examples, and permissive license; the need for post-processing of mixels is noted in the risks.
Images
Reasons for Recommendation
- New model: Trained on MOSS-Image-SFT, specifically for grid-aligned pixel art and 2D game assets; the author provided a comparison between generation resolution and pixel grid: 512×512 (training resolution) corresponds to 64×64, 768×768 corresponds to 96×96, and 1024×1024 corresponds to 128×128
- Serves MOSS text-to-image users needing game assets like sprite sheets, tiles, isometric scenes, and pixel-art bosses; the grid is fixed, and shrinking to 12.5% yields native pixel dimensions, eliminating manual alignment
- Complete parameters: Euler (flowmatch / sgm_uniform), 50 steps (30–50), CFG 3.0 (3.0–6.0), Flow Shift 5.0, SLG skips layer 22, LoRA strength 1.0; the author claims it can be stacked with 8-step Turbo / DMD acceleration LoRAs and provided 4 example prompts
- 15 example images; free, license allows generation, renting, selling, and selling fusions; released about 2 hours ago with 49 downloads and 8 likes (indicating attention only)
How to Use
- Load moss-pixelart-v.0.3.safetensors (138 MB · BF16), strength 1.0, start the prompt with pxl8x (or [pxl8x]); the trigger word is only mentioned in the description and not registered as a trigger word
- Standard generation: Euler + sgm_uniform, 30–50 steps, CFG 3.0–6.0, Flow Shift 5.0, set SLG uncond_skip_layers [22], slg_start 0.01, slg_stop 1.0
- Accelerated generation: Stack a 8-step Turbo / DMD LoRA, change steps to 8, still use Euler, Flow Shift 5.0, and set both LoRA strengths to approximately 1.0
- Use side lengths like 512 / 768 / 1024 for resolution; when assets ready for direct game use are needed, the author recommends using Sprite Fusion Pixel Snapper for grid snapping and cleanup
Risks and Limitations
- The author states that diffusion decoding occasionally produces sub-pixel edge bleeding or grid drift (mixels), requiring post-processing for game assets
- At 512×512, the logical canvas is only 64×64; the author recommends keeping text within 1–2 words or switching to 1024×1024
- Version v0.3 is still early; the author explicitly excluded layer 22 during training, and SLG skips the same layer during inference; the effect of changing SLG settings is not documented
- 2 sensitive-level examples have been hidden
Recommended (4)
[A] AG: Vine - Monochromatic Dream · MOSS-Image
LoRA · Port · Author Azriaelis · Released 10-28 11:00 · Downloads 65
The MOSS version of the Vine style LoRA from the Ars Goetia series: Based on the name and tags, it appears to be a monochrome, high-contrast, halftone dot plus neon-feel anime illustration style; the model card does not describe the style item by item; trigger word @V1N3ST0RM, the author claims the LoRA effect is significantly stronger when the trigger word is included.
Downgraded from S to A: MOSS parameters are clearly documented and visible examples are plentiful, but the style lacks textual description, the accompanying turbo LoRA is not specified, and training material sources are mixed.
Images
Reasons for Recommendation
- Port: Previous versions existed for Krea 2 and Anima; the MOSS version's release notes are empty, but the model card lists a set of MOSS parameters (Base model + 8-step turbo acceleration LoRA)
- Serves users wanting monochrome, high-contrast, halftone-textured illustrations in MOSS text-to-image; the author claims this series is a whimsical small experiment, with varying quality across versions
- Complete parameters: Stack a 8-step turbo LoRA on the Base model, Euler + simple, strength 0.8-1, 8-16 steps, CFG 1-3
- 9 visible example images, all included examples have prompts; free, license allows generation and on-site renting; this version was released about 10 hours ago with 65 downloads and 10 likes, the entire model has accumulated 515 downloads (indicating attention only)
How to Use
- In the MOSS Base workflow, first stack a 8-step turbo acceleration LoRA, then load VINE-MOSS-V1 (1).safetensors (246 MB · BF16), include the trigger word "@V1N3ST0RM" in the prompt
- Strength 0.8-1, 8-16 steps, CFG 1-3, sampler Euler, scheduler simple; the author did not specify which turbo LoRA to use
- The "0.6-0.8 / 25 steps / CFG 4" in the auto-excerpt are Anima version parameters, and "0.8-1.4 / 8-10 steps / CFG 1" are Krea 2 Turbo version parameters; neither applies to the MOSS version
Risks and Limitations
- Training images mix generated images from Midjourney and NijiJourney with web images from Tumblr and elsewhere; the author did not specify the exact sources
- The author self-describes it as a small experiment, with unstable series quality; the style can only be judged by looking at the examples
- 7 sensitive-level examples have been hidden; the license does not allow selling, nor does it allow publishing derivative models
Trigger Word: @V1N3ST0RM
[A] Franbow style (Moss/Krea2) · v2.0
LoRA · Port · Author deygdrixe221 · Released 10-28 17:02 · Downloads 20
The MOSS version of a LoRA for 2D background art style in point-and-click adventure games: The author claims it references the atmosphere and composition of games like Fran Bow, producing flat hand-drawn, picture-book-like, layered stage-cutout front-facing scenes with darkened corners; trigger word FranBowStyle.
Maintaining A: Specific use case (game scene backgrounds), numerous visible examples, copyable recommended prompts, and permissive license, but lacking strength and sampling parameters, and the style directly targets a commercial game.
Images
Reasons for Recommendation
- Port: A previous Krea 2 version existed (with a full composition description as the trigger word); the MOSS version v2.0 changed to the short trigger word FranBowStyle, with empty release notes
- Serves background art / scene concept art for point-and-click adventure games and dark fairy tale themes: strict front view, orthographic projection, backgrounds only without characters
- The author provided a full recommended positive prompt (dark fairy tale picture book, front view, stage set, layered paper-cut parallax, low saturation, heavy vignette, etc.), which can be copied directly
- 16 visible example images; free, license allows generation, renting, selling, and selling fusions; released about 4 hours ago with 20 downloads (indicating attention only)
How to Use
- Load MossImage_LoRA__FranBowStyle.safetensors (36 MB · BF16), include "FranBowStyle" in the prompt, followed by the recommended positive prompt from the model card
- The author did not provide strength, steps, CFG, or sampler
Risks and Limitations
- The author self-describes it as an experimental LoRA; the name directly points to the art style of the commercial game Fran Bow, with no authorization statement seen; the license checking "sellable" does not imply consent from the original rights holder
- No generation parameters provided; the author did not specify the training material sources
- 3 sensitive-level examples have been hidden
Trigger Word: FranBowStyle
[A] [MOSS & IL]【flyx3】Colorful Line Art / Colorful Line Art · MOSS_v1.0
LoRA · Port · Author flyx3 · Released 10-28 17:13 · Downloads 59
The MOSS version of a colorful line art style LoRA: Mimics the color setting sheet effect from line art to color art in animation production, adding partial hand-drawn coloring (the author claims mostly colored pencil texture) to black-and-white line art; trigger words Colorful Line Art / partially colored / color trace.
Downgraded from S to A: The style's use case is clear, trigger words are fully registered, and visible examples are plentiful, but the MOSS version has only a one-sentence description, with no steps, CFG, or sampler provided besides the weights.
Images
Reasons for Recommendation
- Port: An Illustrious version existed previously; the MOSS version's release notes are empty, and the model card only adds a note stating that the author claims the MOSS version performs well but requires more natural language descriptions, and users can copy the author's preset key prompts if they don't know how to write them.
- For services wanting a "line art + partial coloring" setting with a draft quality for MOSS text-to-image, such as character design sheets and anime art style illustrations.
- All three trigger words are registered; recommended weight 0.8~1.5 (general model card notation, no specific MOSS parameters listed).
- 11 visible example images; free; this version released about 4 hours ago with 59 downloads and 9 likes; the entire model has accumulated 7400 downloads and 1041 likes, mainly from the Illustrious version (indicating attention only).
How to Use
- Load Colorful Line Art_MOSS.safetensors (15 MB · BF16), include "Colorful Line Art", "partially colored", and "color trace" in the prompt, then describe the scene using natural language.
- Strength 0.8~1.5; steps, CFG, and sampler are not provided by the author.
Risks and Limitations
- 9 sensitive-level examples have been hidden.
- The MOSS version lacks separate release notes and generation parameters; it is unclear if the weight range follows the IL version, and the author has not specified the training data source.
- The license only allows on-site rental; commercial use of generated images and sales are prohibited, as is secondary fusion.
Trigger Words: Colorful Line Art, partially colored, color trace
[A] The Weirdies · Weirdies [Moss]
LoRA · Port · Author TheP3NGU1N · Released 10-28 15:48 · Downloads 30
The MOSS version of the "Weirdies" LoRA for grotesque creatures: generates fantastical monsters and surreal creatures with large eyes and neon tones; when applied to characters, it introduces a biomechanical feel resembling bone armor; trigger word W31RD13S.
Downgraded from S to A: Unique use case, complete trigger words and strength settings, and many visible examples, but the author claims it is a beta version, examples have been upscaled, and steps and CFG are missing.
Images
Reasons for Recommendation
- Port: Previous versions existed for Flux.1 D, SDXL 1.0, and Pony; the author states that this time Weirdies and their own Biotacean dataset were combined for training, resulting in a biomechanical appearance resembling bone armor on characters, which is the main difference from older versions.
- Serves text-to-image scenarios such as monster design, surreal illustrations, and fantasy creature concept art; the author claims MOSS responds well to weird prompts, so this version was remade.
- Trigger words and strength are provided: strength 0.8-1.0, use around .5 for slight variations.
- 9 visible example images with prompts; free; this version released about 5 hours ago with 30 downloads and 5 likes; the entire model has accumulated 2993 downloads (indicating attention only).
How to Use
- Load WeirdiesMoss_V1_Final.safetensors (123 MB · BF16), include "W31RD13S" in the prompt; strength 0.8-1.0, lower to around .5 for subtle effects.
- The author reminds users to always include the background in the prompt, otherwise the background tends to repeat; steps, CFG, and sampler are not provided by the author.
Risks and Limitations
- The author claims it is a "beta" version and admits they did not write prompts specifically for MOSS.
- All example images have been upscaled using SeedVR2, so they do not represent the resolution and details of direct LoRA output.
- 3 sensitive-level examples have been hidden; the license only allows on-site rental, prohibits derivatives, and does not include commercial use of generated images.
Trigger Word: W31RD13S
Worth Watching (14)
[B] Realism By Stable Yogi Moss · v1.5 INT4
Checkpoint · Version Update · Author Stable_Yogi · Released 10-28 20:30 · Downloads 22
The INT4-ConvRot quantized file (3.0 GB) newly added to the realistic photo checkpoint Realism by Stable Yogi (MOSS) V1.5, intended for RTX 20 and newer NVIDIA GPUs with small VRAM; the author claims it produces slightly darker images than other formats.
Images
[B] NijiReol | Semi-realistic Niji Style [Anima/Krea2/MOSS/NAI/Z-Image] · MOSS v1.0
LoRA · Port · Author Imagine__That · Released 10-28 19:48 · Downloads 41
The MOSS version of a semi-realistic style LoRA trained on curated Niji v7 generated images: the author describes it as a cute-leaning semi-realistic style with a slight painterly feel; trigger word nijireol, strength and sampling parameters for the MOSS version are not provided by the author.
Images
[B] V67 Negative Space · V1 MOOS
LoRA · Port · Author angelomaiota · Released 10-28 16:01 · Downloads 24
The MOSS version of a composition LoRA that controls the "negative space" ratio in the image: adjusts the amount of whitespace between "subject engulfed by environment" and "subject filling the frame", supports positive and negative weights; the model card was written for FLUX, and the MOSS version itself has no description.
Images
[B] Moss-image - Betm illustration | Style · v1.0
LoRA · Version Update · Author shigjfg · Released 10-28 17:46 · Downloads 19
The official MOSS version v1.0 of the "Betm illustration" illustration style LoRA: the author claims it was trained only on illustration materials without anime materials, enhancing skin texture; trigger word artist:betmillustrationstyle.
Images
[B] Moss Image BBOX Prompter · v1.0.11
Workflow · Version Update · Author ukr8b3g201 · Released 10-28 10:51 · Downloads 79
A ComfyUI visual bounding box prompt node (adapted from Krea2 BBOX Prompter Suite for MOSS-Image): draw boxes on the canvas to place characters, objects, and text, write descriptions for each box, and export the layout to the MOSS workflow; this version v1.0.11 is a feature update.
Images
[B] Vintage Travel Poster Style [MOSS] · Vintage Travel Poster
LoRA · New Model · Author RalFinger · Released 10-28 19:05 · Downloads 30
RalFinger's vintage travel poster style LoRA (MOSS), trigger word ral-vista; the name points to the visual style of old-fashioned travel promotional posters, but the model card has no text description other than promotional links.
Images
[B] Risograph Print Style [MOSS] · Risograph Print
LoRA · New Model · Author RalFinger · Released 10-28 18:43 · Downloads 25
A style LoRA named Risograph Print, trigger word ral-rizo; the model card body only contains training platform promotional links and a Discord invitation, with no description of use cases or parameters.
Images
[B] Realistic Background Enchanted | Moss image · v1.0
LoRA · New Model · Author Akai_azusa · Released 10-28 17:14 · Downloads 75
A style LoRA that integrates anime characters into realistic scenes: the author claims it enhances the tendency of "anime character + realistic textured environment and lighting" that MOSS can already do, and also improves lighting; no trigger word.
Images
[B] Natural Skin Realism [MOSS] · Natural Skin Realism
LoRA · New Model · Author RalFinger · Released 10-28 18:47 · Downloads 74
A realistic skin LoRA named "Natural Skin Realism", which from the name appears to be used to make portrait skin more natural; the model card only contains training platform promotions and Discord links, with no description of effects or parameters, trigger word ral-truskin.
Images
[B] Genishin lumine | Genshin Lumine · v1.0
LoRA · New Model · Author Suzukanaoko · Released 10-28 17:22 · Downloads 26
A character LoRA for Lumine from "Genshin Impact": trained on MOSS Base, the author claims it is suitable for general SFT; includes three switchable outfits (Traveler outfit, Fated Stars outfit, Shadow outfit), and outfits can be lent to other characters; parameters and known issues are well-documented, but there are no visible examples.
Images
[B] Deadite - Evil Dead - Moss · v1.0
LoRA · New Model · Author EvilPeterPan · Released 10-28 11:28 · Downloads 51
A LoRA for the "Deadite" possessed demon look from the "Evil Dead" series: generates horror characters with white eyes and demonic features, training images come from Evil Dead 1, 2, and Ash Vs Evil Dead; trigger words Deadite, White Eyes, Evil-Dead.
Images
[B] Slavic Dark Fantasy Artstyle · MOSS Image v.1.0
LoRA · Port · Author GreenRazor · Released 10-28 18:31 · Downloads 14
The MOSS version of a Slavic dark fantasy art style LoRA: the author claims it is aimed at themes of ancient legends, black magic, harsh nature, and mythological creatures, adding unique textures and heavy shadows; the filename is darkrays style; no trigger word.
Images
[B] Regional character LoRAs for MOSS-Image using the layout bounding boxes (two-pass workflow + small custom node) · v1.0
Workflow · New Model · Author collectedtooth4908 · Released 10-28 13:29 · Downloads 65
A ComfyUI workflow + small custom node that applies character LoRAs only within MOSS layout boxes: first generate the overall composition and global style, then locally resample each box using its respective character LoRA and prompt to prevent multiple character LoRAs from bleeding into each other.
Image
【B】1111 - MOSS Image Asian Utopian · UBW
Checkpoint · New Model · Author hinablue · Published 10-28 16:29 · Downloads 2 · Paid
A MOSS fused checkpoint named "Asian Utopian": the author performed full fine-tuning with LoRA on the Base using a curated dataset, adjusting limbs, poses, and expressions, then fused it back into the model with layered weights; the trigger word is "asian woman", and the name refers to Asian female portraits.
Image
Listed Only (23)
- Rin Nightmare Style · Rin Nightmare MOSS V1 — LoRA: A MOSS version of an anime-style LoRA imitating the artist Rin Nightmare; the author claims similar works were mixed in; the model card has only two lines of text and no style description.
- Arknights: Endfield Xiaocong Si (スイ/si) arknights · moss — LoRA: A MOSS version of the character LoRA for "Si" from Arknights: Endfield: low twin tails with double buns, an Eastern dragon tail, black detached wings, a gold headband, and a white dress with detached sleeves; the registered trigger word is the Chinese "祀（明日方舟）".
- Moss East Asian Girl · v1.0 SFT — LoRA: A portrait LoRA that changes female faces to large-eyed East Asian female faces (version name v1.0 SFT): trained on AI-generated fictional character images, no trigger word; the author claims the hit rate is low and requires changing seeds frequently.
- Lia 4 MOSS · v1.0 — LoRA: A test LoRA trained by the author on-site (Lia 4 MOSS): the model card does not specify what style or character it learns; the author claims local results are only "whitish, yellowish", and the upload is for testing on-site.
- [MOSS & Anima & IL] NARUTO / Naruto, Masashi Kishimoto, Tetsuya Nishio - Artist Style · MOSS_v1.0 — LoRA: A MOSS version of a LoRA imitating the styles of Naruto original creator Masashi Kishimoto (manga style) and animation key animator Tetsuya Nishio (anime style), each with a trigger word; the author claims it can also draw most Naruto characters; permanent paid download.
- Moss-style-nai5 (Part 1) · v1.0 — LoRA: An anime-style LoRA trained on a dataset of images generated by NovelAI 5 (nai5), based on MOSS Base; the trigger word is @neoanime, with an optional set of auxiliary words for eye highlights and PBR skin texture.
- Original Manga Style Green (Pokemon Adventures) · v1.0 — LoRA: A character LoRA for Green from the manga Pokemon Adventures (Pokémon Special), targeting the MOSS Base checkpoint, with the trigger word pkmn_green_adv; the model card has no appearance description.
- chiffon hijab MOSS · v1 — LoRA: A clothing LoRA that adds a chiffon hijab to characters, with the trigger word hijab; the model card only says "Hijab generation for MOSS".
- RadiantDream · MOSS v1.0 — LoRA: A MOSS version of the author's "favorite style" LoRA: the model card has only two sentences, does not specify the style, and has no trigger words or parameters.
- Zenless Zone Zero | Eldreda Sunbringer · moss — LoRA: A MOSS version of the character LoRA for Eldreda Sunbringer from Zenless Zone Zero, restoring features like heterochromia, silver-gray long hair, and horn-like headwear; the MOSS version trigger word is a full paragraph of English appearance description.
- WHISK FACE Black [Krea2/MOSS] · V2 MOSS — LoRA: A MOSS SFT version of the "WHISK FACE Black" face LoRA, with the trigger word whiskface-black; the model card does not describe the features or source of this face, only stating the training set size.
- MOSS Murata Range (Renji) Style · v1 — LoRA: A LoRA imitating the dieselpunk style of anime/manga artist Range Murata; the author claims it was trained on 50 images from this artist; the trigger word is rmurata, recommended at full strength; permanent paid download.
- Apone3D - Style - Illustrious XL / Anima / MOSS Image · MOSS Image — LoRA: A MOSS version of a LoRA imitating the rendering style of Blender 3D artist Apone3D: the author claims they switched to a newer dataset and re-tagged using an LLM; no visible examples.
- mikozin style · v1.0 — LoRA: A style LoRA named mikozin, with the description only saying "a style lora", without specifying whose style it is or its features; no trigger word.
- scale for lamia/mermaid · moss v1 — LoRA: A MOSS version of a concept LoRA enhancing scale representation for lamias (snake women) and mermaids: the author provided a weight of 1, and 0.8~1 when stacked with other LoRAs; the MOSS version has no registered trigger word.
- MOSS-style-Glossy Skin · Moss-base-V1.0 — LoRA: A MOSS version of an anime-style LoRA (Glossy Skin): retains the trigger word @dream craft and auxiliary prompts from the Anima version; the author claims it emphasizes skin gloss and fabric fold texture; the author has not tested the MOSS version themselves.
- Cyber-Punk Line Art Moss Base Image · v1.0 — LoRA: A cyberpunk line art style LoRA (for use with MOSS Base): the author claims it can draw robots, cyborgs, and flesh-and-blood characters; no trigger word.
- Hakuro (Onmyoji) LoRA - Illustrious · Moss-image V1.0 — LoRA: A MOSS version of the character LoRA for Hakuro from Onmyoji: trained with bilingual Chinese-English descriptions, both Chinese and English prompts work; the trigger word is 白狼Hakuro (or just 白狼 or Hakuro), with a default outfit prompt attached.
- Linglong Bai Yuekui · v1.0 — LoRA: A character LoRA for Bai Yuekui from the domestic animation Linglong: based on MOSS SFT, trained on 200 images, with the trigger word byk plus a description of silver-white short hair, a light gray fur-collared long coat, etc.
- Corisande Vael v2 (scar) · v1 — LoRA: A character LoRA for the scarred version (v2) of the character Corisande Vael; the model card contains only the character "6", with no information on the character's source, appearance, or trigger word.
- CAM/@camtheartguy - Style - Anima / Illustrious XL / MOSS Image · MOSS Image — LoRA: A MOSS version of a LoRA imitating the style of artist CAM (@camtheartguy); the model card only has a danbooru tag link and the phrase "clean data, no watermarks", with no trigger words or parameters.
- single above knee amputation · Moss - forearm crutches — LoRA: The second MOSS version of a concept LoRA for characters with single above-knee amputation: this version corresponds to forearm crutches, while the previous version was for axillary crutches; no trigger word, no description.
- Channel Style (caststation) · MOSS Image Base v1.0 — LoRA: The third version v1.0 of a style LoRA imitating the style of artist caststation (@channel) for MOSS Base; the model card only states that the MOSS version requires no trigger word, with no style description or parameters.
Data: Civitai Public API (Base Model MOSS Image), inclusion window 10-28 08:31 → 10-28 21:01 (Beijing Time). Recommendations reviewed by editors; download counts only reflect attention.