MOSS Image Morning Report 2026-11-03 | Top Pick: North American Steam Locomotives | MOSS-Image LoRA

This issue reviews 53 new MOSS versions. Top pick: North American Steam Locomotives LoRA: generates famous preserved locomotives and railway scenes by name and model, no trigger word required, with complete settings for strength, steps, CFG, and sampler, and the author clearly documents known defects. Tier A includes 80s fantasy movie stills, the gothic dark scene Black Rain, the grey-teal dry-brush anime style Sabikaze, the "calm chaos" surreal scenes, and the Inkpoet watercolor ink-line style. Auto-S entries for realistic random faces, personal fusion base models, standard game characters, and sports car ports lacking parameters are downgraded to B; CitronLegacy's batch-ported Zelda characters, AuthentiQ's onesie series, and character LoRAs with overly brief descriptions are rated C.

Zhenzhen's AI Radar · MOSS Image · 2026-11-03 Morning Report

This issue features 50 new versions: Strongly Recommended 1 · Recommended 5 · Worth Watching 19 · Included Only 25

Editor's Note This issue reviews 53 new MOSS versions. Top pick: North American Steam Locomotives LoRA: generates famous preserved locomotives and railway scenes by name and model, no trigger word required, with complete settings for strength, steps, CFG, and sampler, and the author clearly documents known defects. Tier A includes 80s fantasy movie stills, the gothic dark scene Black Rain, the grey-teal dry-brush anime style Sabikaze, the "calm chaos" surreal scenes, and the Inkpoet watercolor ink-line style. Auto-S entries for realistic random faces, personal fusion base models, standard game characters, and sports car ports lacking parameters are downgraded to B; CitronLegacy's batch-ported Zelda characters, AuthentiQ's onesie series, and character LoRAs with overly brief descriptions are rated C.

Strongly Recommended (1)

[S] North American Steam Locomotives | MOSS-Image LoRA · v0.5

LoRA · New Model · Author colinu · Published 11-03 01:23 · Downloads 27

North American steam locomotive photo LoRA: can generate famous preserved locomotives by name (Big Boy 4014, N&W 611, SP 4449 Daylight, etc.), common wheel arrangements (2-8-2, 4-8-4, etc.), narrow-gauge locomotives, as well as railway scenes such as smoke, bridge crossings, hill climbs, stations, museums, cabs, and black-and-white archival photos.

Maintained at S: specific subject matter, ability to specify locomotives by name, clear documentation of parameters and known defects, all examples visible, and permissive license; this is the most rigorously documented MOSS entry this issue.

Images

Example 1216x832

Reasons for Recommendation

  • New model: The MOSS base model struggles to accurately recognize specific locomotive models and liveries; this LoRA incorporates 46 locomotive categories, and the author lists the locomotives that can be summoned by name and the supported wheel arrangements
  • Serves railway photography-style images, nostalgic archival photos, transportation history illustrations, and scene images for model enthusiasts; prompt writing is simple: start with the locomotive name and model, then describe the scene, and short prompts work well
  • Complete parameters are provided for easy replication: strength 1.0, example images use 40 steps, CFG 3, euler / sgm_uniform, shift 5, 1216×832; the author states that both Base and SFT work, with SFT producing sharper results and better composition
  • The author proactively documents known defects and training details (approximately 2,900 photos primarily from Wikimedia Commons, each with three lengths of MOSS-style captions, rank 64 / alpha 16, approximately 13,500 steps); all 13 example images are visible; free, license allows image generation and rental; published approximately 8 hours ago with 27 downloads (indicating attention only)

How to Use

  • Load na_steam_locomotives_moss_v0.5.safetensors (573 MB), no trigger word required, start the prompt with the locomotive name and model, e.g., "Union Pacific Big Boy, a 4-8-8-4 articulated steam locomotive.", followed by camera angle and scene description
  • For image generation, use MOSS-Image SFT (Base also works), strength 1.0, 40 steps, CFG 3, euler + sgm_uniform, shift 5

Risks and Limitations

  • The author states that v0.5 still has several inaccuracies: geared logging locomotives like Shay, Heisler, and Climax are generated as standard rod-driven locomotives, and names are often printed on the cab; SP's cab-forward locomotives have the cab drawn at the rear; Challengers often resemble single-unit 4-8-4
  • Locomotive numbers and road name lettering are often garbled, which the author attributes partly to base model limitations
  • Weights are FP32, making the file much larger than similar LoRAs; this entry was not tested

Recommended (5)

[A] 80s Fantasy Movie · MOSS

LoRA · Port · Author ArsMachina · Published 11-02 21:17 · Downloads 171

MOSS SFT version of the "80s Fantasy Movie" film still style LoRA: mimics the visual texture of early 80s fantasy films, generating "old movie screenshots" featuring swordsmen, sorcerers, kings, fantasy creatures, and forest scenes.

Maintained at A after being capped from the S tier by automatic scoring: established style, example prompts are copy-paste ready, key MOSS usage points (skip layer, strength cap) are documented, but complete sampling parameters are missing. Choose between this and the MOSS Base version on the same page.

Images

Example 1024x1024

Reasons for Recommendation

  • Port: This style already has over a dozen base model versions including Krea 2, MiniMax H3, Flux.3 Klein, and Anima; there are two MOSS versions, and this entry is the one trained on SFT, which the author states is based on the V2 dataset (V2 includes a large number of fantasy characters, monsters, and scenes)
  • Key usage difference for the MOSS version: the author states that Skip Layer (skip layer guidance) must be enabled, otherwise results will be poor, and all examples use a base workflow with skip layer; SFT version strength should not exceed 0.9 unless paired with a Turbo LoRA
  • Serves fantasy-themed cinematic stills, retro posters, and board game/novel illustrations; example prompts consist of a trigger word followed by a natural language description, covering portraits, group shots, and animal scenes without people
  • Free; 8 visible examples; this version has 171 downloads, with the entire model accumulating 44939 downloads (indicating attention only)

How to Use

  • Load 80sFantasyMovieMossSft.safetensors (138 MB), start the prompt with "ArsMovieStill, 80s Fantasy Movie Still,", then describe the scene in natural language
  • Skip Layer Guidance must be enabled in the workflow; strength should not exceed 0.9 (can be higher when paired with Turbo LoRA)
  • Steps, CFG, and sampler are not documented in the description by the author, but are available in the example image metadata

Risks and Limitations

  • Training images are from Midjourney generations (both tags and update notes mention the MJ dataset), so the style is "80s fantasy films as seen by AI," not actual film stills
  • Relies on skip layer guidance, but the author does not specify which layers to skip; example image metadata shows 25 steps, CFG 2.5, er_sde
  • License does not allow fusion; 9 sensitive-level examples are hidden, more than the visible examples; this entry was not tested

Trigger Word: ArsMovieStill, 80s Fantasy Movie Still

[A] City Worlds : Black Rain - MOSS Image · V1

LoRA · New Model · Author Z1os · Published 11-03 01:55 · Downloads 50

"Black Rain," a gothic dark scene LoRA from the "City Worlds" series: castle-like interiors with black walls, carved wooden furniture, and deep red window light, as well as gothic cathedral-style city street scenes in red torrential rain at night.

Downgraded from S to A: many examples, replicable parameters, clear direction; however, the description is sarcastic, the style must be inferred entirely from examples, and the license prohibits selling generated images.

Images

Example 896x1152

Reasons for Recommendation

  • New model: The author's description of "colorful and joyful" is sarcastic; based on tags (gothic, dark, creepy, baroque, vampire, tower, bridge) and example prompts, it generates dark environment images such as vampire castle living rooms, bedrooms, bars, and red neon rainy night cities
  • Serves scene concept art, gothic scene settings for games or novels, and dark wallpapers; characters can be placed in the scenes, and the author states that both indoor and outdoor scenes can be generated
  • Parameters are replicable: strength 1; 50 steps with MOSS SFT, or 8 steps with a 8-step Turbo LoRA; 19 examples with prompts included; published approximately 7 hours ago with 50 downloads (indicating attention only)

How to Use

  • Load CW_Black_Rain_Preview_Moss.safetensors (43 MB), include the trigger word "z1os" in the prompt, followed by a long natural language description of the scene (materials, light source colors, furnishings)
  • Strength 1; use 50 steps with MOSS SFT, or 8 steps when stacking a Turbo acceleration LoRA; example image metadata indicates the use of MOSS SFT int8 + an acceleration LoRA named MOSS_Image_SFT_8_steps_dmd

Risks and Limitations

  • Strict license: the author prohibits commercial use, fusion, and using it to train other LoRAs, and generated images cannot be sold; contact the author for commercial use
  • The author does not positively describe the style in text, so it must be judged from tags and examples; the filename includes "Preview," suggesting it may not be the final version
  • The acceleration LoRA configuration in the example image metadata is 10 steps, CFG 3, Euler, which slightly differs from the 8 steps in the description; 1 sensitive-level examples are hidden; this entry was not tested

Trigger Word: z1os

[A] [MOSS Image] Sabikaze - grey-teal dry-brush anime style · v1

LoRA · New Model · Author lepetitprince716355 · Published 11-03 03:23 · Downloads 32

MOSS version of the "Sabikaze" quiet anime illustration style LoRA: low-saturation grey-blue / grey-teal tones, with only a small area of warm color accent in the image, thick ink lines, flat shading blocks, and dry-brush texture on clothing and backgrounds.

Downgraded from S to A: description, parameters, and pitfalls are all complete and replicable; however, this is a port of the author's existing style, with only 6 examples, leaving the top spot for the issue's only entirely new capability entry.

Images

Example 832x1248

Reasons for Recommendation

  • Port: The author previously released a LoRA with the same name on Anima and Krea 2; the MOSS version was retrained on MOSS-Image SFT using the same batch of 57 images generated by the author (ai-toolkit, rank 32, 2000 steps), not a conversion
  • The style is described in text, and MOSS-specific usage is provided: after the trigger word, a sentence must explicitly state "this is an anime-style digital illustration," otherwise MOSS will generate photos; characters work best in realistic environments like streets, rooms, stalls, and ports, and a warm color should be specified for an object in the scene
  • Parameters are replicable: strength 1.0, 25 steps, CFG 2.5, er_sde / simple, shift 5, 832x1248, loaded with LoraLoaderModelOnly; all 6 examples are visible; free; published approximately 6 hours ago with 32 downloads (indicating attention only)

How to Use

  • Load moss_sabikaze_v1.safetensors (142 MB), start the prompt with "@z3-sabikaze.", immediately followed by a sentence specifying the style, e.g., "Anime-style digital illustration of ...", then describe the scene in complete sentences, and specify a warm color for an object in the scene
  • Author's test environment: MOSS-Image SFT (int8 ConvRot DiT) + Qwen3.5 9B text encoder, strength 1.0, 25 steps, CFG 2.5, er_sde + simple, shift 5
  • The author states that the style takes effect even without the trigger word after loading the LoRA

Risks and Limitations

  • The author notes that scenes with storefronts or walls occasionally produce signs with garbled text; even when the prompt specifies no text, this issue persists, requiring a seed change.
  • The training images were generated by the author themselves 57 images (the generation model is unspecified); the style ceiling is limited by the quality of these original generated images.
  • License requires attribution and prohibits selling; only 6 examples are provided; this entry was not run.

Trigger Word: @z3-sabikaze

【A】CalmСhaos · v1.0

LoRA · New Model · Author Reversoma · Published 11-03 04:56 · Downloads 38

“Calm Chaos” surreal scene LoRA: Ordinary people go about their daily routines (eating noodles, reading newspapers, making breakfast, taking selfies) as if nothing is wrong, while giant monsters, tornadoes, or UFO invasions occur simultaneously around them. The visuals have a realistic cinematic feel.

Maintained at A: Fresh concept, easily explained in one sentence, examples match the description, usable without trigger words; lacks sampling parameters and has few examples, preventing an S rating.

Images

Example 832x1248

Reasons for Recommendation

  • New Model: Turns the contrast trope of “expressionless protagonist amidst apocalyptic chaos” into a LoRA. The author clearly states the goal in one sentence: calm daily actions + bizarre surrounding chaos + completely indifferent protagonist.
  • Serves text-to-image generation for absurd posters, memes, social media images, and short film storyboards; visible examples align with the description: taking a peace-sign selfie while UFOs loom outside the window, eating instant noodles while clothes fly around in a laundromat, reading a newspaper at a street-side café before a tornado, pouring drinks in pajamas while kitchen items float mid-air.
  • Simple usage: No trigger word needed; start adjusting strength from 1.0; free, with an open license (including fusion and selling fusions); published about 4 hours ago with 38 downloads (indicating attention only).

How to Use

  • Load CalmChaos_v1_step1500.safetensors (138 MB), set strength to 1.0 initially and adjust as needed; no trigger word required.
  • Prompts must include two parts: the calm daily action the protagonist is performing, and the chaotic events occurring around them.
  • Steps, CFG, sampler, and whether to use Base or SFT are not specified by the author.

Risks and Limitations

  • Only 3 visible examples (plus 1 sensitive-level examples hidden); stability across different subjects is unknown.
  • No sampling parameters or base model specifications; this entry was not run.

【A】Inkpoet · v1.0 moss image

LoRA · Ported · Author JaneDoe01 · Published 11-03 04:23 · Downloads 27

MOSS version of the Inkpoet watercolor + ink line style LoRA: Flowing watercolor washes, expressive ink tones and hand-drawn lines with paper texture; applicable to characters, animals, landscapes, and architecture.

Maintained at A: Specific style description, prompt writing methods and strength tiers are actionable, license is most permissive; but the MOSS version lacks dedicated documentation and sampling parameters, preventing an S rating.

Images

Example 720x1280

Reasons for Recommendation

  • Ported: Previously only available for Krea 2; the MOSS version’s release note is just one sentence, “MOSS IMAGE version of Inkpoet,” with no author explanation of differences from the Krea 2 version. The usage instructions on the model card are shared between both versions.
  • Serves illustrations, concept art, and book imagery requiring a “slightly rough, hand-drawn” watercolor ink line effect; the author claims it is particularly suitable for dramatic compositions and cinematic scenes. Visible examples include crows, long-haired women, canyon rivers, white wolves on snowy cliffs, pine forests in mist, and girls holding knives, all with a unified style.
  • Usage instructions are actionable: Trigger word, strength range, and effects at each strength level are provided, along with advice not to stack other style words; free, open license; this version has 27 downloads, while the entire model has accumulated 485 downloads (indicating attention only).

How to Use

  • Load inkpoet-v1.0 MOSS IMAGE.safetensors (138 MB), start prompts with “inkpoet,” then directly describe the scene. The author recommends not adding other style words initially.
  • Strength 0.6–1.0: 0.6–0.7 retains more base model features, 0.8 is the author’s recommended starting point, and at 1.0 the watercolor and ink lines are most pronounced.
  • For a cinematic feel, describe composition and moments (low angle, wide-angle, backlighting, wind/rain/fog, character poses, foreground/background) rather than stacking style tags; steps, CFG, and sampler are not specified by the author.

Risks and Limitations

  • The trigger word “inkpoet” is only registered for the Krea 2 version; it is not registered for the MOSS version and appears only in the model card body text.
  • Strength recommendations are shared between versions; whether the MOSS version requires different strengths is not specified by the author; no sampling parameters provided.
  • 3 sensitive-level examples are hidden; this entry was not run.

Worth Watching (19)

【B】Asian Face Mix - Moss · v1.0

LoRA · New Model · Author tjjjjjjj842 · Published 11-02 22:03 · Downloads 80

Asian female “random face” LoRA: Generates different faces each time to break the repetition of the same face, rather than depicting a fixed character.

Images

Example 1504x2656

【B】MOSSquid · v0.1

Base Model / Checkpoint · New Model · Author giko · Published 11-03 00:56 · Downloads 120

MOSSquid: The author’s personal MOSS fusion base model, combining 80% MOSS SFT + 20% MOEMOSS-Base, with tags leaning towards anime.

Images

Example 1024x1344

【B】Iris Von Everec MOSS · v2.0

LoRA · Version Update · Author gkhan_ · Published 11-03 00:18 · Downloads 36

Second version of the LoRA for Iris von Everec from The Witcher 3: Hearts of Stone: Freckles, yellow-green eyes, smudged eyeliner, dark lipstick, low bun, plus the original high-collared black patterned mourning dress.

Images

Example 1676x1216

【B】80s Fantasy Movie · MOSS Base

LoRA · Ported · Author ArsMachina · Published 11-02 21:17 · Downloads 49

MOSS Base version of the “80s Fantasy Movie” style LoRA, which mimics fantasy movie stills from the 80s. It shares the same style as the MOSS (SFT) version released on the same day but consists of two separate weights trained on different base models.

Images

Example 1024x1024

【B】Ferrari F40 - 1987 · v1.0 MOSS

LoRA · Ported · Author mastermerlin · Published 11-03 08:30 · Downloads 5

MOSS version of the LoRA for the 1987 Ferrari F40, used to accurately reproduce the car’s exterior (low front end, integrated large rear wing, cooling vents).

Images

Example 1680x944

【B】Sam_MOSS · Turbo v2.6 semi-realistic

Base Model / Checkpoint · Version Update · Author toya_san · Published 11-03 00:11 · Downloads 144

Turbo v2.6 semi-realistic version of the third-party MOSS accelerated base model “Sam_MOSS”: Generates images in 8 steps or more. The author claims it has slightly less realism than the previous version but is cleaner and sharper.

Images

Example 1440x1088

【B】Genshin Sandrone | 原神 桑多涅 · v1.0

LoRA · New Model · Author Suzuka_naoko · Published 11-02 21:29 · Downloads 40

Character LoRA for Sandrone from Genshin Impact: Two outfits, maid attire and white nightgown. Headpieces, hairbands, gear-shaped pupils, clockwork parts, etc., have separate Chinese and English tags, allowing outfit changes and realistic cosplay photos.

Images

Example 1024x1536

【B】Moss-image - ammyruki | Style · v1.0

LoRA · New Model · Author GreatWhiteWolf · Published 11-02 23:02 · Downloads 25

Artist ammyruki style LoRA: The author provides no textual description of the style. Based on tags and examples, it appears to be bright cel-shaded anime illustrations featuring primarily young male characters.

Images

Example 832x1216

【B】Moss Image Compact T2I workflow · v1.0

Workflow · New Model · Author NeuralVandal · Published 11-02 22:02 · Downloads 49

A streamlined ComfyUI workflow for MOSS text-to-image generation, featuring the author’s custom Civitai Crybaby node: Scans prompts against Civitai’s official word list, highlights potentially violating words with adjustable warning levels.

Images

Example 2323x1131

【B】MOSS Kawaii Impressionism · v1.0

LoRA · New Model · Author _Envy_ · Published 11-02 21:28 · Downloads 45

“Kawaii Impressionism” style LoRA: Short, dotted brushstrokes and pastel colors create an impressionist oil painting feel, applied to scenes like town squares, mall atriums, amusement parks, shrines, and castles, along with some anime characters.

Images

Example 1200x1600

【B】Medea (Hades) Moss · v1.0

LoRA · New Model · Author gkhan_ · Published 11-02 22:06 · Downloads 32

LoRA for Medea, a character from 2 Hades II: long straight black hair covering one eye, the other eye glowing teal, deep red lips, wearing a sleeveless robe with glowing green rib patterns and spiked bracers.

Images

Example 832x1216

【B】sky02 Light and shadow Style [Anima/MOSS] · v2.0 [MOSS]

LoRA · Version Update · Author skyddw · Published 11-02 22:03 · Downloads 111

Second MOSS version of the sky02 light and shadow style LoRA: Enhances anime illustrations with backlighting, rim lighting, highlights, and glowing particles to make images brighter and more luminous.

Images

Example 1280x1920

【B】MOSS Liminality · v1.0

LoRA · New Model · Author _Envy_ · Published 11-02 21:07 · Downloads 52

“Liminal Space” surreal digital painting style LoRA: Empty corridors, theaters, platforms, bus stops, water towers, and observatories with thick paint textures and dramatic lighting, creating quiet yet slightly unsettling images.

Images

Example 1200x1600

【B】Pigment Paradox · MOSS Image

LoRA · Ported · Author Wackiii · Published 11-03 01:38 · Downloads 44

“Black and white photo + rainbow pigment explosion” mixed media style LoRA: The main subject is high-contrast black and white photography, partially disrupted by bursts of pink, purple, yellow, and cyan colored smoke and watercolor clouds, featuring aged textures and ink splatters.

Images

Example 1024x1024

【B】Cyclope · 1 - MOSS

LoRA · Port · Author moutonrebelle · Published 11-03 00:48 · Downloads 30

The MOSS version of the original one-eyed alien female character "Cyclope": featuring a giant eye occupying most of the face, blue-gray skin, hair with a gradient from teal to coral, slender limbs, and a small mouth.

Image

Example 1472x1024

【B】Vellus Hair Peach Fuzz-GMR · Vellus Hair Peach FuzzGMR

LoRA · Port · Author Artikuz_Ai · Published 11-03 04:42 · Downloads 8 · Paid

The MOSS version of a LoRA that enhances the texture of vellus hair (peach fuzz, the fine hairs on cheeks and skin surfaces) in realistic portraits.

Image

Example 832x1216

【B】MOSS Image Base T2I (Prompt Enhance) · v1.0

Workflow · New Model · Author yosxpeee · Published 11-03 00:27 · Downloads 29

A workflow that integrates a local LLM for automatic prompt expansion into the official ComfyUI MOSS Image Base text-to-image template, including a MOSS-specific system prompt for the LLM.

Image

Example 1024x1024

【B】Fictional Foundation Moss Turbo · 1.0

Base Model / Checkpoint · New Model · Author alexds9 · Published 11-03 04:17 · Downloads 0

The Turbo version of the MOSS fine-tuned base model "Fictional Foundation": described by the author as a stylized 3D digital art style, featuring smooth shading, refined facial features, high-contrast lighting, glossy/metallic surfaces, and glowing effects.

Image

Video 1024x1536

【B】Yunhui · Eastern Fantasy | Yunhui Style · v1.0

LoRA · New Model · Author oreolove · Published 11-02 21:10 · Downloads 0 · Paid

An Eastern fantasy / Chinese-style Xianxia art style LoRA: featuring a semi-realistic anime texture, flowing long hair and garments, exquisite jewelry and metal accessories, dragon patterns, smoke, and soft lighting.

Image

Example 1088x1920

Included Only (25)

Data: Civitai public API (base model MOSS Image), collection window 11-02 20:31 → 11-03 09:01 (Beijing Time). Recommendations are editorially reviewed; download counts only reflect attention.

Civitai

Explore the latest AI Radar updates