MOSS Image Morning Report 2026-10-28 | Top Pick: MOSS Image Turbo Lite

Zhenzhen's AI Radar · MOSS Image · 2026-10-28 Morning Report

This issue features 47 new versions: Strongly Recommended 1 · Recommended 5 · Worth Watching 10 · Listed Only 31

Editor's Note This issue reviews 48 new MOSS versions, rated as S 1, A 5, B 10, and C 32. Top pick: MOSS Image Turbo Lite, an acceleration LoRA for the original MOSS-Image-SFT, generating images in 16 steps with CFG 1. Strength, sampler, flow shift, and workflow are all provided, maintaining an S rating. Other highlights: Golden Hour warm-tone film LoRA (trigger word, strength, and 6 example prompts included, upgraded from B to A), Realism Enhancer LoRA, Digital Grunge with rough brushstrokes, and retro pixel art style. The auto-S rated "Million Cats" is an overfitted model that memorized 3 copyrighted covers, downgraded to C; styles imitating living artists, ported styles with empty descriptions, and standard character LoRAs are mostly rated C. The FinalCut quantized base model was downgraded to C because the author self-reported canvas issues.

Strongly Recommended (1)

[S] MOSS Image Turbo Lite · 16step-cfg1 Beta

LoRA · New Model · Author Coffmeteor · Published 10-28 00:08 · Downloads 66

An acceleration LoRA (Beta) for the original MOSS-Image-SFT: compresses image generation to 16 steps with CFG 1; the author claims skin oiliness has improved, but multi-person limb and anatomical issues remain unresolved; no trigger word.

Maintaining S rating: Single, clear purpose; complete parameters and workflow; permissive license; the author's explanation of limitations is specific. Beta status and compatibility only with the original SFT are noted in the risks.

Images

Reasons for Recommendation

• New model: A utility acceleration LoRA enabling SFT to generate images in 16 steps at CFG 1; the negative prompt branch is not calculated at CFG 1; the version name indicates it is Beta
• The author clearly states what it does not solve: limb and anatomical issues in multi-person scenes persist; only claims slight improvement in skin oiliness, without exaggeration
• Actionable: Recommended configuration is complete—original MOSS-Image-SFT BF16, strength 1.0, Euler / simple, flow shift 5, negative prompt left empty, skip-layer guidance disabled; includes a ComfyUI workflow zip archive
• All 6 visible example images include prompts; free, license allows image generation, rental, sale, and sale of derivatives; approximately 9 hours since release with 66 downloads (indicating attention only)

How to Use

• Load the original MOSS-Image-SFT BF16 base model, then load MOSS-Image-16step-cfg1-BF16.safetensors (567 MB · BF16), strength 1.0, no trigger word
• 16 steps, CFG 1, Euler / simple, flow shift 5, negative prompt left empty, skip-layer guidance disabled; alternatively, import the included MOSS-Image-16step-cfg1-Workflow.zip

Risks and Limitations

• Beta version; the author states that limb and anatomical issues in multi-person scenes persist
• Designed only for the original MOSS-Image-SFT; compatibility with MOSS Base or other fine-tuned base models is not specified by the author; comparison with other few-step acceleration LoRAs is not provided
• 1 sensitive-level examples are hidden

Recommended (5)

[A] [Moss] Realism Enhancer · v1.0

LoRA · New Model · Author alcaitiff · Published 10-28 01:23 · Downloads 563

A realism enhancement LoRA pushing MOSS towards photorealism (full LoRA, not a slider): the author claims it can suppress MOSS's tendency towards cartoon/anime styles in some cases, and significantly improves skin and textures in already realistic images; no trigger word.

Maintaining A rating: Clear direction, numerous examples with workflows, permissive license, but text parameters only include strength and sampler; the author admits results vary by image.

Images

Reasons for Recommendation

• New model: Trained only on high-quality realistic images above 2048px, targeting photorealism, natural skin, and detailed textures
• Serves realistic portraits and photo-style MOSS text-to-image generation, especially when realistic prompts are still rendered as cartoons by the base model
• The author is restrained: not perfect, results vary by image and settings; recommended strength starts at 1, Euler/simple is usable, but the author claims ER_SDE/Bong_Tangent is better
• 12 visible example images, the author claims all have embedded workflows and can be loaded and reproduced in ComfyUI; free, license allows image generation, rental, sale, and sale of derivatives; approximately 8 hours since release with 563 downloads and 59 likes (indicating attention only)

How to Use

• Load Moss-RealismEnhancer-V1.safetensors (69 MB · BF16), start adjusting strength from 1, no trigger word, describe prompts as realistic photos
• Sampler ER_SDE + Bong_Tangent scheduler (author claims superior to Euler/simple); steps and CFG are not provided in the text, can be read from the workflows embedded in the example images

Risks and Limitations

• The author claims it only suppresses cartoon tendencies in some cases, not guaranteed; training image sources are not specified by the author
• The Bong_Tangent scheduler is not native to ComfyUI, usually requiring third-party node packages
• 8 sensitive-level examples are hidden

[A] [MOSS] sonsp's Art Style — Personal style · v2.0

LoRA · Version Update · Author SnoSp · Published 10-27 21:42 · Downloads 80

The MOSS version v2.0 of the author's personal aesthetic 2D anime illustration style LoRA: expanded training set, adjusted training parameters, the author claims underfitting in V1 has improved; trigger word is sonsp, the model card does not describe style features in text.

Downgraded from S to A: Update notes and MOSS parameters are comprehensive, but the style lacks textual description, visible examples are only 4, and sensitive-level examples constitute the majority.

Images

Reasons for Recommendation

• Version update: Update notes specify expanded training set and adjusted training parameters, aiming to improve style reproduction; the author claims obvious underfitting in V1 is no longer an issue in current tests, recommended strength narrowed from V1's 0.6–2.0 to 0.6-1
• The author declares it is their own aesthetic preference, not a replication of a famous artist; V1 training images were generated using Krea 2, sources of new materials in V2 are not specified by the author
• Serves users wanting this style of 2D illustration in MOSS text-to-image; the author suggests specifying light direction, light source, and lighting atmosphere in prompts, using only "a 2D anime illustration" as the basic style description
• Parameters complete: MOSS_Image_Base, 25–80 steps, CFG 2.5–5.0, Euler + sgm_uniform or er_sde + simple, Shift 5.0–8.0, Skip Layer 22, resolution above 1.5MP; the author claims it seems better at high resolutions around 5MP
• Free; approximately 11 hours since this version's release with 80 downloads and 9 likes, cumulative downloads for the entire model are 277 (indicating attention only)

How to Use

• In the MOSS Base workflow, load Sonsp_Moss_Style_v2.safetensors (142 MB · BF16), include the trigger word "sonsp" in the prompt, add "a 2D anime illustration" and specific lighting descriptions; the author did not use negative prompts during testing
• Strength 0.6-1; 25–80 steps, CFG 2.5–5.0, Euler + sgm_uniform or er_sde + simple, Shift 5.0–8.0, Skip Layer 22, resolution above 1.5MP

Risks and Limitations

• 4 visible examples, 11 sensitive-level examples are hidden, exceeding the number of visible examples
• The style lacks textual description, judgment relies solely on examples; the V1 training set consists of AI-generated images, potentially carrying Krea 2 generation preferences
• The author is unsure if improvements at high resolution come from the LoRA or the base model; the license only allows image generation and on-site rental, the author requests no redistribution to other platforms

Trigger Word: sonsp

[A] Digital Grunge | [MOSS | Krea2 | Anima] · moss_grunge_1.0

LoRA · Ported · Author NoiseWriggler · Published 10-28 03:15 · Downloads 36

The MOSS version of the rough texture style LoRA: adds obvious brushstrokes, sketch lines, thick paint textures, and noise to digital illustrations, making overly clean images appear handmade and mixed-media; no trigger word.

Downgraded from S to A: Style features and MOSS parameters are clearly written, but the MOSS version has no version notes, examples are upscaled without metadata, and training material sources are vague.

Images

Reasons for Recommendation

• Ported: Previously available for Krea 2 and Anima; the MOSS version was trained on MOSS-Image-Base and tested on MOSS-Image-SFT, version notes are empty, the model card lists a separate set of MOSS parameters
• Serves 2D illustration, anime, sketch, oil painting, and mixed-media MOSS text-to-image; the author claims it is best suited for mecha, cybernetics, cyberpunk, armor, fantasy/sci-fi, and dark themes, and also enhances glitch, chromatic aberration, and scanline effects
• Parameters complete: er_sde + simple, 50 steps, CFG 3, strength 1, approximately 2MP, side lengths multiples of 16; the author claims MOSS is very sensitive to distortion and glitch effects, reduce strength or remove corruption-related descriptions in prompts for a lighter effect
• 7 visible example images; free, license allows image generation and rental; approximately 6 hours since this version's release with 36 downloads, cumulative downloads for the entire model are 703 and 80 likes (indicating attention only)

How to Use

• Load Digital_Grunge_NW_.safetensors (138 MB · BF16) in the MOSS-Image-SFT workflow, with no trigger word; the prompt must specify illustration types such as illustration / 2d / anime, as the author explicitly states that this art style does not appear when paired with realistic descriptions
• er_sde + simple, 50 steps, CFG 3, strength 1, approx. 2MP; the author's example uses the default Comfy workflow with PiD upscaling
• The "strength 0.9 – 1.2" in the auto-excerpt refers to parameters for the Krea 2 version, not the MOSS version

Risks and Limitations

• The training materials are a mix of works by multiple artists and images; the specific sources are not listed by the author
• The examples have been upscaled with PiD; the direct output effect is not shown separately; the examples lack generation parameter metadata
• 3 sensitive-level examples are hidden; the license does not allow selling

[A] Nostapixx Retro Pixel Art Style · Nostapixx Retro Pixel Art

LoRA · New Model · Author: trendingvibestar · Published 10-28 04:36 · Downloads 53

Retro pixel art style LoRA: transforms images into a retro pixel art style; the trigger phrase "Nostapixx pixel art style" must be placed at the beginning of the prompt.

Downgraded from S to A: The purpose is clear, and there are many visible examples, but the model card contains only three sentences, the accompanying turbo LoRA is not named, and CFG and sampler information are missing.

Images

Reasons for Recommendation

• New model: A single-style pixel art LoRA, with the trigger phrase and strength specified in the model card
• Suitable for MOSS text-to-image services requiring retro pixel art illustrations or game asset-style visuals
• The author claims excellent results when using the SFT base model with a turbo acceleration LoRA at 8 steps; all 17 example images are visible, with no hidden sensitive-level examples, and the examples include prompts
• Free, with a license allowing image generation, rental, selling, and selling fusions; published approximately 4 hours ago with 53 downloads and 7 likes (representing attention only)

How to Use

• Load Nostapixx_moss.safetensors (66 MB · BF16), strength 0.75-1.0; start the prompt with "Nostapixx pixel art style" (as stated in the description, though not registered as a trigger word)
• Use the MOSS SFT base model, stack a turbo acceleration LoRA, and run for 8 steps; the specific turbo LoRA, CFG, and sampler are not provided by the author

Risks and Limitations

• The model card contains only three sentences; the source of training materials and the relationship between pixel granularity and resolution are not described by the author
• The trigger phrase is not registered as a trigger word and must be manually included in the prompt

[A] Golden Hour - MOSS · MOSS

LoRA · New Model · Author: Ginger_Palace · Published 10-27 21:03 · Downloads 0

Warm-toned film-like "Golden Hour" style LoRA: adds sunset backlighting, peach/apricot warm tones, lifted blacks, and shallow depth of field to any subject; the author claims no additional grain is added; trigger word: goldfilm_style.

Upgraded from B to A: The purpose is clear, trigger word and strength are provided, example prompts can be copied directly, it is free with a permissive license; however, it lacks step count/CFG information, so it does not reach S.

Images

Reasons for Recommendation

• New model: A general lighting + color grading style LoRA, not tied to specific characters; the author lists applicable subjects such as portraits, domestic daily scenes, rural street scenes, still life, indoor window light, animals, and four-season landscapes
• Suitable for MOSS realistic text-to-image scenarios seeking a nostalgic warm-light film photo feel; there is no need to write "golden hour" or "film" in the prompt, as the trigger word handles the style, and you only need to describe the scene
• The author provides 6 example prompts that can be copied directly, along with 5 visible example images (768×1344 vertical images)
• Free, with a license allowing image generation, rental, selling, and selling fusions, making it one of the most permissive licenses in this group

How to Use

• Load goldenHourMOSS_v1.safetensors (123 MB), start the prompt with "goldfilm_style, " followed by the scene description, e.g., "goldfilm_style, a golden retriever running through a meadow with wildflowers"
• Start with a strength of 1.0, lower it for a subtler effect; if the image is too orange, reduce the strength or request cooler tones in the prompt; step count, CFG, and sampler are not provided by the author

Risks and Limitations

• The author states the style is intentionally warm, and some scenes may appear overly orange; text on signs and labels may still be garbled
• All training images are AI-generated, containing only adults and daily clothing scenarios; the effect on out-of-scope subjects (night scenes, urban environments) is not specified by the author
• The download count shows 0, which may be due to statistical delay; 1 examples are hidden by sensitivity level; there are no step count/CFG parameters

Trigger Word: goldfilm_style

Worth Watching (10)

[B] Pokémon Black & White - characters and style · v1.0

LoRA · New Model · Author: Porzo · Published 10-28 06:38 · Downloads 51

A style and character LoRA trained on screenshots from the Pokémon Black & White anime: can be used for an anime screenshot style, and can also generate the 11 female characters listed in the model card; trigger word: zPokemonBW.

Images

[B] [moss image]takamichi style · v1.0

LoRA · New Model · Author: kozue · Published 10-28 01:34 · Downloads 151

A LoRA imitating the art style of artist takamichi (filename: takamichi-comic), trained based on MOSS SFT, with the trigger word "takamichi style"; the author requires it to be used in conjunction with a 4-step acceleration LoRA.

Images

[B] PieMOSS · v0.1

Base Model / Checkpoint · New Model · Author: Imbrium · Published 10-28 05:24 · Downloads 9 · Paid

A realistic photo fine-tuning base model for MOSS-Image (pre-release v0.1): trained on approximately 15k manually curated data points by the author, aiming to remove the overly polished "plastic look" of AI photos, with tags pointing to the photo texture of the 2000s.

Images

[B] Gemini-chan (Gemini-chan) — MOSS-Image Version (MOSS LoRA) · v1.0

LoRA · New Model · Author: kadaizuo604 · Published 10-28 00:28 · Downloads 16

A MOSS character LoRA for the AI anthropomorphic character Gemini-chan (original design by Nopsil): features long blue-purple gradient hair, heterochromia (left red, right gold), cat ears, star-shaped hair accessories, double buns, and a blue fluffy tail; trigger word: gemini musume, strength must be approximately 2.0.

Images

[B] Moebius Style · Moss Image v1.0

LoRA · Ported · Author: feloxxx75 · Published 10-27 22:26 · Downloads 57

The MOSS version of a LoRA imitating the art style of French comic artist Moebius; the model card contains only the sentence "A Lora following the art style of Moebius"; no trigger word.

Images

[B] Neon Cyberpunk Cyborg · MOSS Image

LoRA · Ported · Author: Wackiii · Published 10-28 07:10 · Downloads 32

The MOSS version of a neon cyborg portrait style LoRA: realistic skin paired with facial armor, exposed circuits, mechanical jaw, glowing respirator, and cable headgear, with cyan, pink, purple, and acid green neon lighting; no trigger word.

Images

[B] Kodak Portra 400 Style [MOSS] · Kodak Portra 400

LoRA · New Model · Author: RalFinger · Published 10-27 22:09 · Downloads 110

RalFinger's Kodak Portra 400 film style LoRA (MOSS), trigger word: ral-portra; the name points to the color tone and grain of this portrait color negative film, but the model card contains no text description other than promotional links.

Images

[B] Dragon Princess / GPT-chan — MOSS-Image Version (MOSS LoRA v3) · v3.0

LoRA · New Model · Author: kadaizuo604 · Published 10-27 23:59 · Downloads 14

A MOSS-specific character LoRA v3 for the original character "GPT-chan (Dragon Princess)": a dragon girl with purple curved horns, pointed ears, light purple membrane wings, a white dragon tail, and a white off-shoulder dress; trigger word: gptmsg; the model card fully details the strength, sampling parameters, prompt writing style, and training process.

Images

[B] Nissan Skyline GT-R (R34) · v1.0

LoRA · New Model · Author: Addictedd · Published 10-27 22:49 · Downloads 1 · Paid

A LoRA for the Nissan Skyline GT-R (R34) model: stably renders this car in MOSS realistic images, trigger word: r34gtr; the author provides prompt writing techniques to ensure the GT-R badge on the front of the car appears correctly.

Images

[B] Monster Hunter Rathalos Armor (MOSS) · v1.0

LoCon · New Model · Author: EDG · Published 10-28 01:20 · Downloads 15

A clothing LoCon for the Rathalos armor set from Monster Hunter: enables MOSS to draw Rathalos armor, or modify it to create your own Rathalos-style outfit; trigger word: wearing edgRath_armor.

Images

Included Only (31)

• The Cat That Lived a Million Times · 3in1_moss_v0.0.0 — LoRA: The MOSS version of an overfitted LoRA trained on 3 covers of classic Japanese picture books ("The Cat That Lived a Million Times," "Who's Still Awake," "Guri and Gura"): using the full description registered by the author as the prompt almost exactly reproduces these covers, tagged for parody collage (kusokora) purposes.
• Faria Vogel (Thrud) | Faria Vogel (Thrud) · For MOSS — LoRA: The MOSS version of a character LoRA for Faria Vogel from the game Thrud: features black hair in low twin tails, green eyes, a white headband, headphones around the neck, and a black frayed jacket; trigger word: faniya.
• ✦ Z-Cast: Sennibar O. [OC] · Moss — LoRA: The MOSS version of the character LoRA for the author's original male character Sennibar O: a blond male, with a setting that transforms from a nerd to an arrogant CEO, trigger word Sennibar.
• Dizi930 Style [Krea2/MOSS] · MOSS v1.0 — LoRA: The MOSS version of the Dizi930 art style LoRA: the model card has only two lines of text, with no explanation of whose art style it is or what its characteristics are, so one can only judge by the examples.
• Hayami / Japanese AI girl · MOSS — LoRA: The MOSS version of the character LoRA for a fictional AI-generated realistic portrait of a Japanese woman, trigger word kz_hayami; the author states that the person does not exist in reality.
• Sister Vita Nox · v1 — LoRA: The MOSS character LoRA for the original tattooed battle nun character Sister Vita Nox: black nun habit with slate blue hair ribbon, cross and dagger tattoos and the word 'vita' on the face, Gothic tattoos covering the neck, chest, and arms, crystal cross jewelry; registered trigger word b4dnun.
• Moss Image FinalCut (5GB - 25Step) · Comfy_Quant — Base Model / Checkpoint: The new Comfy_Quant version of the MOSS fine-tuned base model FinalCut: converts the model into NVFP4 (3.1 GB) and INT8 (approx. 5 GB) quantized files specifically for ComfyUI; the author states that both quantized versions have canvas issues and are inferior to the universal format.
• DeepSeek Whale Girl — MOSS-Image Version (MOSS LoRA) · v1.0 — LoRA: The MOSS character LoRA for the community fan-made AI girl 'DeepSeek Whale Girl' (maid outfit version): blue gradient ultra-long hair, whale fin hair accessory, whale tail, maid outfit; trigger word deepseek whale girl, strength must be set to around 2.0.
• VeloriaMix · Moss — LoRA: The MOSS version of a cartoonish anime art style LoRA, with training images from the author's earlier Illustrious base model; optional trigger word veloria, the author admits the results are still being explored.
• Art Style Silver City 3D silverpalace style 【anima】 · moss — LoRA: The MOSS version of the 'Silver City' 3D art style LoRA (umina style series), the model card only contains 'no trigger word needed' and a request for likes.
• Art Style kantoku カントク Supervisor style 【anima/krea2】 · moss — LoRA: The MOSS version of a LoRA imitating the art style of living illustrator Kantoku (kantoku / Supervisor), the model card only contains 'no trigger word needed' and the author's appreciation for this artist.
• Art Style Loom style 【anima】 · moss — LoRA: The MOSS version of the 'Loom' art style LoRA (umina style series), the model card only has one sentence: 'This model does not require a trigger word.'
• MiaoMiao RealSkin · MOSS 1.0 — Base Model / Checkpoint: The MOSS experimental version 1.0 of the MiaoMiao RealSkin series base models: the series focuses on realistic skin textures for anime characters on Illustrious / Anima, the MOSS version model card only provides parameters, with no description of differences from other versions.
• Art Style shigure-s style 【anima】 · moss — LoRA: The MOSS version of the shigure-s art style LoRA (umina style series), the model card only has one sentence: 'This model does not require a trigger word.'
• AI Art Style salt style 【anima】 · moss — LoRA: The MOSS version of the art style LoRA named 'AI Art Style salt style' (umina style series), the model card only contains 'no trigger word needed' and a request for likes.
• Art Style mochirong style 【anima】 · moss — LoRA: The MOSS version of the mochirong art style LoRA (umina style series), the model card only contains 'no trigger word needed' and a request for likes.
• CapStyle // LoRA · Moss v0.1 — LoRA: The MOSS test version v0.1 of a chibi anime art style LoRA created by author NekoCap based on their own CapAnima, trigger words c4psty1e and chibi, the author recommends using it with a MOSS fine-tuned base model.
• MossArt · v2.0 — LoRA: The v2.0 of a general-purpose 'artistic' LoRA trained on hundreds of different art styles (painting, anime, webtoons, etc.), which pushes MOSS outputs towards an illustrative style; the author states the effect is strong, recommending a strength of 0.5 to 0.8.
• Claude Girl / Ors Scholar — MOSS-Image Version (MOSS LoRA) · v1.0 — LoRA: The MOSS character LoRA for the community original character 'Claude Girl' (ors scholar): orange-red long hair, amber eyes, orange chrysanthemum hair accessory, white high-neck ruffled long dress with black corset; trigger word ors scholar, strength must be set to around 2.0.
• Channel Style (caststation) Anima Base · MOSS Image Base v0.2 — LoRA: The second version v0.2 of the MOSS Base art style LoRA imitating the style of artist caststation (@channel); the model card only states that the MOSS version requires no trigger word, with no art style description or parameters.
• Art Style Oni Cat style 【anima】 · moss — LoRA: The MOSS version of the art style LoRA labeled 'Oni Cat', the description is only one sentence: 'a very beautiful art style', with no trigger word, parameters, or art style description.
• Art Style YZR style 【anima】 · moss — LoRA: The MOSS version of the art style LoRA labeled YZR style, the model card is empty, with no trigger word, parameters, or art style description.
• Benes Style · MOSS v1.0 — LoRA: The MOSS version of the art style LoRA imitating American comic artist Ed Benes (tagged Marvel / DC American comics), trigger word @edben; the model card only contains the word 'Temp'.
• Art Style Gwan-E style 【anima】 · moss — LoRA: The MOSS version of the art style LoRA labeled Gwan-E style, the model card only has one sentence: 'This model does not require a trigger word', with no art style description or parameters.
• Art Style Dog Meat muse dash Muse Dash style 【anima】 · moss — LoRA: The MOSS version of the art style LoRA imitating the style of 'Dog Meat', the artist for the rhythm game 'Muse Dash', the description is only one sentence: 'can be paired with colorful if possible'.
• V.L.C. Anime Style · v1 — LoRA: An anime art style LoRA named V.L.C. Anime Style, the model card only has one line: 'Trigger word: ay3agc', with no art style description or parameters.
• Angel Blade Punish Style · v1.0 — LoRA: An art style LoRA replicating the style of the anime 'Angel Blade Punish' by name, the author claims it is their first MOSS LoRA and will update it later; no trigger word, parameters, or art style description.
• daizen · v1 — LoRA: An art style LoRA named daizen, the description is only one sentence: 'daizer art style for moss', with no explanation of whose art style it is or what its characteristics are; trigger word daizen.
• Waist Slider · Moss Image — LoRA: The MOSS version of a body type slider LoRA for narrowing the waist: adjust strength to control waist thickness, no trigger word needed.
• ILLUSTRATORCORE MOSS 💢 BY STX · V1 — LoRA: An illustration art style LoRA named ILLUSTRATORCORE, the author marks it as being in the testing phase, only providing strength 0.8-1.2, with no trigger word or art style description.
• PEAR.Style | TeraWatt 🍐 · MOSS v1.0 — LoRA: The MOSS version of a body type LoRA that fixes a specific female body type (wide hips, narrow waist), trigger word PEARStyle_TeraWatt; no parameters.

---

Data: Civitai public API (base model MOSS Image), inclusion window 10-27 20:31 → 10-28 09:01 (Beijing Time). Recommendations are reviewed by editors; download counts only reflect attention.