MOSS Image Evening Report 2026-11-05 | Top Pick: AG: Dantalion - Portraiture Midjourney

Zhenzhen's AI Radar · MOSS Image · 2026-11-05 Evening Report

This issue features 44 new versions: Strongly Recommended 0 · Recommended 7 · Worth Watching 15 · Listed Only 22

Editor's Note This issue reviews 49 new MOSS versions. Top pick: Dark Lines MOSS version from the Velvet fantasy art style series: dark fantasy character illustrations with a black-and-gray palette and sharp lines, clear trigger words, and many visible examples, but the MOSS version lacks its own parameters and was automatically downgraded from S to A. There are 9 items in the A tier, including the cocoamix INT8 Turbo base model, a one-stop workflow with two sampling setups and SeedVR2 upscaling, acrylic markers, Shigeru Sudo lyrical paintings, Jirai Kei fashion, Wai anime style, and forest scenes, among others.

Recommended (7)

[A] AG: Dantalion - Portraiture Midjourney · MOSS-Image

LoRA · Port · Author Azriaelis · Published 11-05 09:10 · Downloads 47

A Dantalion style LoRA from the "Ars Goetia" series. The author states that the training images mix sources such as Midjourney, NijiJourney, and Tumblr, following a refined digital illustration / fantasy portrait route, ported from Krea 2 / Anima to MOSS.

Downgraded from automatic S to A: Parameters and trigger words are complete, and examples are sufficient, but the description does not explain what this specific style looks like, indicating it is part of a batch port of the series.

Images

Reasons for Recommendation

• Port: The same LoRA already has Anima and Krea2Turbo versions; the MOSS version provides a separate set of recommended settings (one set for each of the three base models), but the MOSS version description is empty
• Serves text-to-image generation for fantasy, refined illustration-style character portraits; 9 visible examples, all with prompts
• Actionable: Trigger word @D4NT4L10N; the author claims the style is significantly stronger when the trigger word is included; MOSS settings are Base + 8 step Turbo LoRA, euler + simple, strength 0.8-1, 8-16 steps, CFG 1-3
• Free; this version has 47 downloads and 6 likes (indicating attention only)

How to Use

• Use MOSS Base as the base model and attach the 8 step Turbo LoRA, load DANTALION-MOSS-V1 (1).safetensors, strength 0.8-1
• Include the trigger word "@D4NT4L10N" in the prompt; use euler + simple, 8-16 steps, CFG 1-3

Risks and Limitations

• The model card uses a shared description for the entire series and does not specify the exact stylistic features of this Dantalion version, so judgment relies on examples; the author admits that quality varies across LoRAs in the series and describes them as small experiments
• Training images include outputs from other AI drawing tools, with sources not individually specified
• 11 sensitive-level examples are hidden, exceeding the number of visible examples; the license does not allow publishing derivative / fusion models; this item was not tested

Trigger Word: @D4NT4L10N

[A] Shigeru Sudō Style 須藤 しげる · MOSS

LoRA · Port · Author Bex · Published 11-05 15:09 · Downloads 32

A LoRA imitating the style of Japanese painter Shigeru Sudo (died in 1946): "lyrical paintings" from girls' magazines around 1926–1940 years ago, featuring Taisho Romantic style female characters; the MOSS version is ported from the Krea 2 / Flux.1 D version.

Downgraded from automatic S to A: Usage is clear and examples are plentiful, but it is merely a port of a single artist's style, and parameters like step count can only be inferred from example metadata.

Images

Reasons for Recommendation

• Port: The author previously released Flux.1 D (two versions) and Krea 2 versions; the MOSS version was trained on the same dataset as the Krea 2 version, and the author claims results are better after switching to MOSS and incorporating two months of improvements from the training tool Fizgig
• Serves text-to-image generation for retro illustrations and early Showa era girls' magazine style characters; example prompts include women in kimonos holding lanterns and fans, schoolgirls by desks near windows, and girls roller-skating hand-in-hand, etc.
• Usage is specific: Trigger word shigsustyle is optional according to the author; adding "soft-shaded illustration printed on card stock" to the prompt is helpful; strength 1 is sufficient, increasing to 1.4 gets closer to the original style; 14 examples, free

How to Use

• Load Moss Bex shigsustyle_v1-000037.safetensors (BF16), strength 1, increase to 1.4 for a stronger original style; trigger word shigsustyle is optional
• Write "soft-shaded illustration printed on card stock" in the prompt; examples also use style descriptions like "illustration, soft color charcoal/watercolor shading, printed on card stock"
• The author did not provide step count, CFG, or sampler; see risks for example metadata

Risks and Limitations

• Parameter hints captured in the data, such as "increase to 1.2 or 1.3 on FLUX Dev," come from the Flux version description, not MOSS settings
• Example image metadata (not author description): base model moss_image_sft_int8_convrot, 8 steps, CFG 1, Euler; additionally, two example prompts are black-and-white street photography grids, showing no relation to this style
• The model card mentions that all figures in the training images show their faces, so depicting characters facing away requires more prompt effort; early Flux versions had white borders from training images, and the author did not state if the MOSS version still has this
• Single artist style, narrow use case; 2 sensitive-level examples are hidden; the license's commercial terms only allow on-site rental; this item was not tested

Trigger Word: shigsustyle

[A] Jirai Kei (地雷系) - Fashion LoRA - · MOSS Image

LoRA · Port · Author yakitoritamago · Published 11-05 20:40 · Downloads 14

The MOSS version of a LoRA for the Japanese "Jirai Kei / Ryousangata" fashion style: character styling featuring pink and black color schemes, ruffled shirts, corsets, and twin tails.

Images

Reasons for Recommendation

• Port: Training materials are the same as the Krea 2 version, but the MOSS version uses natural language tagging; the author claims outputs on MOSS tend to look more anime-like but blend naturally with the base model
• Serves text-to-image generation for Jirai Kei fashion in character photos / character designs; the author claims it does not replicate specific individuals but provides flexible character archetypes
• Trigger word yt_jiraikei, and a complete natural language styling description is registered for direct copying; 8 visible examples
• The LoRA file is small, approximately 17 MB; free

How to Use

• Load yt-jiraikei_moss-image_261103.safetensors, write "yt_jiraikei" in the prompt, then append the author's registered description: Her black hair is styled into twin tails, adorned with fluffy white hair ties. She is dressed in a pink, ruffled button-down shirt...
• Because it uses natural language tagging, it is recommended to describe clothing details with full sentences rather than stacking tags; the author did not provide strength, step count, or CFG

Risks and Limitations

• No parameters provided; the author states the MOSS version tends to look more anime-like, so adjustments are needed for a realistic photo style
• 2 sensitive-level examples are hidden; the license does not allow commercial use of outputs or publishing derivative models; this item was not tested

Trigger Words: yt_jiraikei, Her black hair is styled into twin tails, adorned with fluffy white hair ties. She is dressed in a pink, ruffled button-down shirt, a black fabric corset featuring ruffles, and a black ruffled skirt.

[A] V67 Forests · V1 MOSS

LoRA · Port · Author angelomaiota · Published 11-05 14:45 · Downloads 26

The MOSS version of a forest scene LoRA: various forests such as tropical rainforests, primeval ancient forests, cactus forests, coniferous forests, bamboo forests, and glowing mushroom forests, using adjective trigger words to switch forest types, lighting, and elements.

Downgraded from automatic S to A: The trigger word system and example prompts are very complete, but the MOSS version lacks its own parameters, and there are only 3 examples.

Images

Reasons for Recommendation

• Port: The same LoRA already has Flux.1 D, Krea 2, and Z Image Turbo versions; the MOSS version description is empty, and the model card body is almost entirely Flux syntax and parameters
• Serves text-to-image generation for landscapes, fantasy scenes, and nature themes; place the main trigger word (Forests:1.2) at the beginning of the prompt, then add adjectives like Tropical, Ancient, Cactus, Boreal, Glowing, Foggy, Fairy, and Butterfly to select forest type and atmosphere
• The model card includes multiple categorized example prompts (rainforest, redwood forest, cactus forest, bamboo forest pandas, aurora coniferous forest, etc.) that can be directly modified
• Free, license allows outputs, rental, sales, and sales of fusions; this version has 26 downloads (indicating attention only)

How to Use

• Load v67-forests-moss-image-base-epoch-01.safetensors, start the prompt with "(Forests:1.2)", add 2 to 4 forest type / lighting adjectives, then write the subject and scene
• The only strength hint from the author that can be reused: fantasy elements like fairies and glowing effects work best at 0.7–0.8; the author recommends resolutions of 1024x1024 or multiples thereof (banners for panoramas)
• The author did not provide MOSS-specific strength, step count, or CFG

Risks and Limitations

• The strength table, CFG, step count, and sampler in the model card are all written for Flux (e.g., Flux Dev 0.65–0.85, CFG 3.0–4.5, 20–30 steps); the author claims it is "optimized for Flux," so directly applying it to the MOSS version may not be suitable
• Only 3 visible examples, so the effects of dozens of forest type / element combinations cannot be previewed
• The filename includes "base," suggesting it was trained on MOSS Base; the author did not specify performance on Turbo / SFT; this item was not tested

Trigger Word: (Forests:1.2), Trees, Tropical, Ancient, Cactus, Boreal, Temperate, Mushroom, Flower, Glowing, Night, Foggy, Giant, Fairy, Insect, Butterfly, Mossy, flora and fauna, 4 seasons of the woods

[A] MOSS-Image All-in-One Workflow: Two Sampling Setups, Prompt Tools & SeedVR2 Upscaling · v1.0

Workflow · New Model · Author CottonByte · Published 11-05 10:26 · Downloads 47

A MOSS one-stop ComfyUI workflow collection (JSON, not new weights): two sampling setups + three prompt inputs (manual / auto-expansion / image-to-prompt) + SeedVR2 upscaling.

Downgraded from automatic S to A: Descriptions and parameters are complete, but local use requires replacing platform-specific nodes, and it is a workflow JSON combining others' processes.

Images

Reasons for Recommendation

• Two generation methods: Method A uses a single sampling pass with two LoRAs; the author states the images are softer and have more atmospheric grain but weaker prompt adherence. Method B first uses SFT, then hands off to Base; the author states skin texture, expressions, and subjects are clearer, and most examples come from Method B.
• Three ways to input prompts: manual writing, automatic expansion of short phrases into detailed prompts, or analyzing reference images to generate prompts using the built-in MOSS template (the author's example uses DeepSeek V4.1 Flash 1017); SeedVR2 upscaling is applied at the end.
• Specific parameters provided: default 50 steps (SFT 35 + Base 15), plus a fast mode with 8 steps split evenly between the two models; the demo configuration uses MOSS INT8 + Qwen 3.5 9B text encoder + FLUX.2 VAE; there are 10 visible examples, along with a YouTube tutorial.

How to Use

• Import Collection of Moss Image Generation Workflows.json; use MOSS INT8 as the base model (switch to BF16 if you have sufficient VRAM), Qwen 3.5 9B as the text encoder, and FLUX.2 for the VAE.
• Run Method B with SFT 35 steps + Base 15 steps; switch to the 8 step mode for speed (the author notes a significant drop in image quality).
• The output size for SeedVR2 upscaling must be set individually within each generation branch; reference image dimensions are not automatically carried over.

Risks and Limitations

• The prompt processing nodes use RunningHub's built-in tools; for local execution, you need to replace them with compatible nodes or APIs. The specific LoRAs used in Method A are not specified in the description.
• The author states that the original creators of the two generation workflows are different people; this entry merely combines them and adds prompt input methods and upscaling. The model card includes a RunningHub referral link.
• 8 sensitive-level examples have been hidden; dependencies on nodes and models must be verified by yourself, as this configuration has not been tested.

【A】cocoamixMOSS (moss image) · int8_turbo_v1.0 (fast)

Base Model / Checkpoint · Version Update · Author ccooccoo · Published 11-05 10:16 · Downloads 34

The INT8 Turbo version of the cocoamix series MOSS merged base model: it integrates a 8-step DMD2 acceleration LoRA, generating images in approximately 10 steps with a file size of about 5.0 GB; the default cocoamix style does not draw pupils.

Downgraded from S to A: This is the INT8 version of the same recipe as the BF16 Turbo released earlier this morning; choose this version if you want a smaller file size. The update notes are empty, and there are few examples.

Images

Reasons for Recommendation

• Version Update: The same model already has a BF16 standard version (50 steps, CFG 4), a BF16 Turbo version (included in the earlier entry this morning), and an old INT8 10-step version. Based on the version name, this is a new INT8 Turbo; according to the model card's merge list, it consists of MOSS-Image + the author's self-trained model + 8-steps-dmd2-lora, effectively making it a quantized fast version of BF16 Turbo.
• Serves anime-oriented text-to-image generation; tags also include realistic, mecha, and chibi styles. The style switches are the same as in the BF16 version: for pupils, write 'eyes has dark_circle shaped pupils'; for vibrant colors, write 'vivid dark color'; realistic photos have fixed sentence structures.
• Parameters provided: 10 steps, CFG 1 - 4 (CFG 1.0 is fastest; negative prompts only work above 1.5), samplers include ER SDE + simple / Euler + normal / DDIM + normal.

How to Use

• Replace the base model with cocoamixMoss01_int8_10steps.safetensors (full INT8 weights, approx. 5.0 GB); no trigger word required.
• Use 10 steps; set CFG to 1.0 for speed, or above 1.5 if you need negative prompts; use sampler ER SDE + simple or Euler + normal.
• To add pupils, include 'eyes has dark_circle shaped pupils'; for high saturation, add 'vivid dark color'; for realistic photos, use the sentence structure 'photorealistic style has realistic background, realistic hair, realistic face...'.

Risks and Limitations

• The author did not write update notes (the version description field is empty), and the difference in image quality compared to the old INT8 10-step version and the BF16 Turbo version is not specified.
• Only 3 examples are provided; terms like 'cocoart' and 'cocoaface' appear in the example prompts, but the author does not clarify if these are trigger words. Example metadata (not from author's description): 10 steps, CFG 1 or 4, er_sde + simple.
• Strict license: Commercial use is prohibited, and generated images are also limited to non-commercial use; permission from the author is required before merging releases or selling generated images; reposting to unauthorized sites incurs a fee.
• The author warns that sensitive content is easily generated and recommends using prompt_enhance for control; this configuration has not been tested.

【A】MOSS Acrylic Marker · v1.0

LoRA · New Model · Author _Envy_ · Published 11-05 15:52 · Downloads 26

Acrylic marker illustration style LoRA: flat color blocks and thick outlines, applicable to characters, architecture, and interiors.

Images

Reasons for Recommendation

• New Model: Acrylic marker painting style on MOSS; the author provided only a one-sentence description.
• Serves text-to-image generation for illustrations, character designs, and scene concept art; 13 visible examples cover sci-fi/fantasy full-body characters, low-angle architecture, and interior scenes. The wide range of subjects allows you to judge the style accordingly.
• Simple usage: trigger word 'acrylic marker', free, license permits image output.

How to Use

• Include the trigger word 'acrylic marker' in the prompt.
• The author did not provide LoRA strength, step count, CFG, or whether to pair with SFT or Base.

Risks and Limitations

• The description is only one sentence; all inference parameters are not provided by the author.
• 7 sensitive-level examples have been hidden; this configuration has not been tested.

Trigger Word: acrylic marker

Worth Watching (15)

【B】MOSS-FET · v0.4

Base Model / Checkpoint · Version Update · Author aueki4g467 · Published 11-05 10:52 · Downloads 125

The v0.4 full model of the MOSS-FET series: The author performed lightweight full-parameter fine-tuning on MOSS using a small number of realistic images, then integrated LoRAs extracted from their previous anime models. The goal is a general-purpose base model that leans towards realism and can handle various prompts.

Images

【B】Succupon All in One Anima Workflow T2I - BBOX - Style Presets - Turbo - Upscale - Prompt Generator - SLG · v1.0

Workflow · New Model · Author Succupon · Published 11-05 15:59 · Downloads 34

An integrated ComfyUI text-to-image workflow for MOSS Image (JSON zip package, not new weights): It integrates Turbo acceleration, a style library, LLM prompt generation, SLG, bounding box (BBOX) prompts, and three types of upscaling into a single page.

Images

【B】IRC-Galleria // Krea 2, MOSS Image · v1.0 MOSS Image

LoRA · Ported · Author konflikti · Published 11-05 15:39 · Downloads 24

A style LoRA replicating the look of selfies from the Finnish social networking site IRC-Galleria in the 2000s: amateur digital cameras, direct flash, and casual indoor shots of young Finns. Ported from Krea 2 to MOSS.

Images

【B】[MOSS & Anima & IL] Fourth Armor X (Megaman X4) /Rockman X4/ RockmanX4, Fourth Armor X / Force Armor X · MOSS_v1.0

LoRA · Ported · Author flyx3 · Published 11-05 12:18 · Downloads 28

A character LoRA for X's 'Fourth Armor / Force Armor' from Mega Man X4, ported from Anima / Illustrious to MOSS.

Images

【B】Kigurumi Onesie: Giraffe [Krea2/MOSS] · MOSS_SFT_v1.0

LoRA · Ported · Author AuthentiQ · Published 11-05 09:03 · Downloads 36

A clothing LoRA for a giraffe-themed kigurumi onesie, oriented towards realism/cosplay, ported from Krea 2 to MOSS (MOSS SFT version).

Images

【B】Moody Moss Minimal Workflow · v2.0 Standard 25 Steps

Workflow · Version Update · Author catlover1937 · Published 11-05 15:40 · Downloads 167

A minimal ComfyUI workflow (JSON, not new weights) accompanying the author's own Moody Moss series base models; this version is the v2.0 standard 25-step version.

Images

【B】Parcel Sorting Warehouse — Conveyor Belts, Box Mountains, Night Shift | MOSS · v1.0

LoRA · New Model · Author 858654128784 · Published 11-05 09:46 · Downloads 82

A concept LoRA that places scenes in a parcel sorting center: long conveyor belts, mountains of cardboard boxes and woven bags, high shelves, forklifts and barcode scanners, workers in reflective vests, and night shift overhead lights.

Images

【B】MOSS-Image-Nigenmame · MOSS-Image-Nigenmame-v1.0

LoRA · New Model · Author Te2dou_Nin9 · Published 11-05 18:04 · Downloads 27

A Japanese anime illustration style LoRA: clean line art, detailed hair strands and clothing fold highlights, cel shading mixed with soft gradients. Backgrounds retain detail but have a blurred effect; the author states it is suitable for half-body shots, close-ups, and everyday atmosphere images.

Images

【B】semi-realistic anime-半写实二次元(for sft) · v1.0

LoRA · New Model · Author NekoArc17 · Published 11-05 17:51 · Downloads 52

Semi-realistic anime-style LoRA: Anime faces + realistic lighting and materials. The author states it was trained on MOSS SFT and works better when paired with SFT.

Image

【B】Early Camera Phone — The One-Megapixel Look of a 2006 Flip Phone | MOSS · v1.0

LoRA · New Model · Author 858654128784 · Published 11-05 13:57 · Downloads 71

A style LoRA mimicking the photo quality of flip/slide phones from around 2006 years ago: approximately one megapixel, soft focus, high noise, greenish-purple tint, overexposed highlights, compression artifacts in shadows, and often skewed composition.

Image

【B】MOSS-FET LoRAs · v1.0 LoRA

LoRA · New Model · Author aueki4g467 · Published 11-05 18:55 · Downloads 20

The author of MOSS-FET extracted LoRAs from the difference between their three anime-oriented fine-tuned models and the original MOSS, splitting them by layer into L (layers 0–23) and U (layers 24–41) halves. There are 6 files in total, which can be used individually or mixed.

Image

【B】Police Lineup — Behind the One-Way Glass, Height Chart, Number Cards | MOSS · v1.0

LoRA · New Model · Author 858654128784 · Published 11-05 11:51 · Downloads 86

A concept LoRA for drawing "police lineup room" scenes: viewed from behind one-way glass, a row of people holding number cards stand against a white wall with height markings, with the back or profile of a witness in the foreground.

Image

【B】Fourth Wall Interior — A Dollhouse Cutaway Building, One Wall Apart | MOSS · v1.0

LoRA · New Model · Author 858654128784 · Published 11-05 11:44 · Downloads 82

A "dollhouse with one wall removed" composition LoRA: the front wall of the entire building is removed, viewed straight-on at eye level, making all rooms simultaneously visible, each with its own lighting and characters.

Image

【B】Composition Controller for MOSS-Image on Forge NEO · v1.1

Other · New Model · Author mangetsu · Published 11-05 09:01 · Downloads 18

A dedicated extension for MOSS-Image on Forge NEO (a compressed package, not weights): it adjusts noise during generation to produce more varied compositions across different seeds, and also allows toggling SLG without reloading the WebUI.

Image

【B】Event Horizon | MossLoraLab · Anime 80s v1.0

LoRA · Version Update · Author ElectricDreams · Published 11-05 20:39 · Downloads 0 · Paid

The second style LoRA in the author's MOSS LoRA Lab series, "Anime 80s": featuring a Japanese anime style from the 80s, with example subjects including neon cities, mecha, space opera, and fantasy characters; requires paid download.

Image

Included Only (22)

• Selena Corey (Singing Lady) Resident Evil Requiem | illustrious + Krea 2 + MOSS · MOSS — LoRA: A character LoRA for Selena Corey (Singing Lady) from the game Resident Evil Requiem. The MOSS version was ported from the Krea 2 / Illustrious version.
• Haizea - Synthetic AI Character · MOSS v1 — LoRA: A character LoRA for Haizea, a virtual realistic female character (described by the author as a fully synthetic character). Use the trigger word "Haizea" to maintain facial consistency across different images.
• MOSS-Image-TendouAris · MOSS-Image-TendouAris — LoRA: A character enhancement LoRA for Tendou Aris from Blue Archive. The author states that MOSS can already draw this character, so this LoRA is only intended to stabilize facial features and the halo, resulting in minimal changes.
• Tail - MOSS-Image · MOSS Image v1.0 — LoRA: A concept LoRA for adding tails to characters (trigger word: "tailed"). The model card contains only the single word "tail".
• Kigurumi onesie: Spiderman [Krea2/MOSS] · MOSS_SFT_v1.0 — LoRA: A clothing LoRA that puts characters in a Spider-Man kigurumi onesie (with hood and mask eye patterns). The version name indicates it was trained on MOSS SFT.
• Wetside · Moss v1.0 — LoRA: An anime illustration style LoRA featuring cartoonish, vibrant colors. The author only wrote "cartoon, fun, vibrant".
• MOSSIDJM · sft_turbo2_int8 — Base Model / Checkpoint: The second version, sft_turbo2_int8, of the MOSS fusion/fine-tuned base model MOSSIDJM. It is an INT8 full model, and the author states that its settings are identical to MOSS Image Turbo.
• Art style ogipote style 【anima/krea2】 · moss — LoRA: A LoRA imitating the art style of Japanese illustrator ogipote, ported from Anima / Krea 2 to MOSS. No trigger word is required.
• Murazaki style · v0.1 MOSS-image — LoCon: A small LoKr model imitating the watercolor art style of furry artist murazaki. The author states this is a trial version trained at 512 resolution.
• MOSS-Image-TendouKei · MOSS-Image_TendouKei-v1.4 — LoRA: A character enhancement LoRA for Kei from Blue Archive. The author requires loading it together with the Tendou Aris LoRA by the same author and adding a halo description; otherwise, the halo will not be drawn accurately.
• Art style bureedo style 【anima】 · moss — LoRA: The MOSS version of the "bureedo style" art style LoRA, ported from Anima. Based on visible examples, it features a clean, bright, soft flat-color cute anime character style. No trigger word is required.
• meili re:zerokara moss · meili mossv1.0 — LoRA: A character LoRA for Meili Portroute from Re:Zero - Starting Life in Another World. The model card only states "moss sft".
• Female Worker warcraft reforged · v1 — LoRA: A character LoRA for the Human Female Villager from Warcraft III: Reforged. The description simply states it is "taken from the female villager character in the game".
• Shaobayan Anima//Moss Base/Moss Heal · v1.0 Moss heal — LoRA: One of the versions of the "Shaobayan" anime illustration style LoRA, converted from Anima to MOSS. This version is named Moss Heal and, according to example metadata, is intended for use with the MOSS Heal base model.
• [MOSS]Artist:Z3zz4 · v1.0 — LoRA: A LoRA imitating the art style of artist z3zz4, trained on MOSS Base. The author self-evaluates it as poorly done (original quote: "feels like a mess") and plans to retrain it once an anime-oriented MOSS base model is available.
• Public Bathhouse | MOSS · v1.0 — LoRA: A concept LoRA for public bathhouse scenes, generating indoor bathhouse settings such as shower areas, scrubbing stations, hot pools, and steam.
• Real Skin · v1.0 — LoRA: A realistic skin LoRA. The author claims it makes MOSS-generated portraits more realistic while minimizing the weakening of the base model's original capabilities, and recommends pairing it with a 8 step DMD2 acceleration LoRA for fast generation of realistic images.
• Shaobayan Anima//Moss Base/Moss Heal · v1.0 Moss Base — LoRA: The Moss Base version of the same "Shaobayan" anime illustration style LoRA. According to example metadata, it is intended for use with the MOSS Base model.
• Mix2.5D Style | MOSS · v1.0 — LoRA: A 2.5D style LoRA trained by the author using their own art style (the name suggests a style between anime and realism). The model card contains only one sentence.
• Retro Animation · v1.0 — LoRA: A retro animation art style LoRA (tags include "trippy" and "retro"). The trigger word is "retro animation", and the model card contains only one sentence.
• FFstyle Anima/Moss Base/Moss Heal · v1.0 Moss heal — LoRA: An anime style LoRA imitating the art style of NovelAI v5 (nai5). This version was trained on MOSS's heal model.
• FFstyle Anima/Moss Base/Moss Heal · v1.0 Moss Base — LoRA: The MOSS Base version of the LoRA imitating the art style of NovelAI v5 (written by the author as nai5), ported from Anima. The author states that despite long training, the Base version still looks "plastic" and inaccurate, while the simultaneously released Heal version is much better.

---

Data: Civitai public API (base model MOSS Image), collection window 11-05 08:30 → 11-05 21:01 (Beijing Time). Recommendations have been reviewed by editors; download counts only reflect popularity.