MOSS Image Morning Report 2026-11-01 | Top Pick: North American Passenger Trains | MOSS-Image LoRA

This issue reviews 53 new MOSS versions. Top pick: North American Passenger Trains LoRA: enables MOSS to recognize specific locomotive models, operator liveries, and passenger cars. Training data comes from publicly available, freely licensed photos. Prompting methods and sampling parameters are fully documented, and the author proactively lists limitations such as easily confused models. Other highlights: 90s Japanese anime style, a reverse slider for removing plastic-looking skin, a text-to-image workflow with bounding box positioning and Hires upscale, Da Vinci manuscript sketch style, and surreal fantasy photography style. Among the automatic S/A ratings, several ported versions lack MOSS-specific parameters (the captured parameters belong to other base models). Standard character and person LoRAs are downgraded to C; another version of Cassidy has the same weight hash as 3442296 and is excluded as a duplicate upload.

Zhenzhen's AI Radar · MOSS Image · 2026-11-01 Morning Report

This issue features 45 new versions: Strongly Recommended 1 · Recommended 6 · Worth Watching 22 · Included Only 16

Editor's Note This issue reviews 53 new MOSS versions. Top pick: North American Passenger Trains LoRA: enables MOSS to recognize specific locomotive models, operator liveries, and passenger cars. Training data comes from publicly available, freely licensed photos. Prompting methods and sampling parameters are fully documented, and the author proactively lists limitations such as easily confused models. Other highlights: 90s Japanese anime style, a reverse slider for removing plastic-looking skin, a text-to-image workflow with bounding box positioning and Hires upscale, Da Vinci manuscript sketch style, and surreal fantasy photography style. Among the automatic S/A ratings, several ported versions lack MOSS-specific parameters (the captured parameters belong to other base models). Standard character and person LoRAs are downgraded to C; another version of Cassidy has the same weight hash as 3442296 and is excluded as a duplicate upload.

Strongly Recommended (1)

[S] North American Passenger Trains | MOSS-Image LoRA · v1.0

LoRA · New Model · Author colinu · Published 11-01 01:46 · Downloads 23

North American Passenger Trains LoRA: enables MOSS to recognize specific locomotive models, operator liveries, and passenger cars, producing realistic photos with railway enthusiast photography compositions. The author states that locomotive road numbers can be rendered relatively stably.

Maintained at S: Clear new knowledge (specific models and liveries), transparent and compliant training data sources, complete parameters, and the author proactively documents limitations. This is the most actionable item in this group.

Images

Example 1216x832

Reasons for Recommendation

  • New model: The author states that the base MOSS can only draw "a train." This LoRA is trained on 2,667 freely licensed photos from Wikimedia Commons across 40 categories, covering models such as EMD F40PH, GE Genesis, Siemens Charger, ACS-64, and Acela, as well as liveries for Amtrak, VIA Rail, and Metra.
  • Serves railway-themed illustrations, transportation scenes, and railway photography-style image generation; the author has a separate freight train LoRA, while this one is dedicated to passenger trains.
  • Complete documentation and actionable: Prompting methods (start with operator + manufacturer + model, then describe livery, road number, camera angle, and environment), strength, steps, CFG, sampler, shift, and resolution are all provided. The author also proactively documents known limitations.
  • 17 example images; free, license allows merging and image generation; training data sources are public and freely licensed; approximately 7 hours after release, it has 23 downloads (indicating attention only).

How to Use

  • Load na_passenger_trains_moss_v1.0.safetensors (573 MB · FP32), no trigger word; start the prompt with the operator, manufacturer, and model, e.g., "Amtrak EMD F40PH diesel locomotive, Phase III paint," then describe the camera angle and scene like an image caption; if you want the locomotive number, specify the road number, and do not add quality tags like "masterpiece."
  • Use on MOSS-Image Base: Strength 1.0, 30–50 steps, CFG 3, euler + sgm_uniform, shift 5 (ModelSamplingSD3), approximately 1 MP landscape images such as 1216×832.

Risks and Limitations

  • Limitations listed by the author: Acela Express and Avelia Liberty may be confused; Coaster F40PH, F59PHI, SC-44, and LRC have very few training images, so liveries are more accurate than body details; control cars (cabbage cars) and real F40PH may be confused; long text and logos are often garbled, only short road numbers are relatively reliable.
  • Covers only North American passenger vehicles; freight, steam, and Eurasian vehicles are out of scope.
  • The file is FP32, resulting in a large size; example images do not include prompts.
  • Sampling parameters are in the full model card (the data.json description is truncated); this item has not been tested.

Recommended (6)

[A] Simple Workflow for MOSS Image (T2I BBOX, I2I) · T2I & BBOX V3.0 Hires

Workflow · Version Update · Author HissingCat · Published 11-01 06:22 · Downloads 25

Version 4 of the MOSS text-to-image workflow: allows using bounding boxes (BBox) to specify the position of characters in the image. This version adds a second Hires refinement upscale stage and character_boxes input.

Maintained at A: The update content is clearly written, has few dependencies, and a low barrier to entry; however, it is just a workflow with no new capabilities.

Images

Example 1456x2176

Reasons for Recommendation

  • Version update: Compared to the previous T2I & BBOX V2, it adds a second Hires fix stage (the author states it supplements details and incidentally upscales), character_boxes input, a lightweight custom node to fix BBox aspect ratios, and the second stage defaults to using the first stage's prompt; the recommended VAE is changed to flux2-vae_small, with links to 4-step and 2-step Turbo LoRAs included.
  • Serves users who want to control character placement in MOSS without installing a bunch of node packs; the author states they primarily use native nodes.
  • 6 example images; free, license is fully open; this version has 25 downloads, with the series totaling 3628 downloads (indicating attention only).

How to Use

  • Download simpleWorkflowForMOSSImage_t2iBBOXV30Hires.zip (13 KB) and import into ComfyUI; first update ComfyUI to the latest version and enable Node 2.0 (required for BBox functionality).
  • Write the content for each position in the respective boxes; if not using BBox, treat it as standard text-to-image; if the second stage prompt is left empty, it uses the first stage's prompt.
  • To add LoRAs: Click "Enter subgraph" on the MOSS-Image node and enable the LoRA node inside; MOSS model download links are in the workflow notes.

Risks and Limitations

  • It is just a workflow, not new weights; the author claims to use native nodes, but this version adds a custom node to fix BBox ratios, so compatibility needs to be verified.
  • Depth control for image-to-image is a separate I2I Depth version and is not included in this package.
  • Depends on the latest version of ComfyUI; this item has not been tested.

[A] 90sAnime Retro (MOSS) | styles · v1.0

LoRA · New Model · Author lerosay · Published 10-31 22:47 · Downloads 64

A MOSS style LoRA that recreates the 90s Japanese anime art style, with a trigger word. The author provides steps, CFG, and strength.

Downgraded from S to A: Parameters are mostly complete and examples are sufficient, but the description is only one sentence, and the style boundaries and training base model are not specified.

Images

Example 1056x1536

Reasons for Recommendation

  • New model: Serves users who want to generate 90s anime-style illustrations in MOSS; the author only describes the style as "90s anime style" in one sentence, so specific lines and colors must be judged from the examples.
  • Parameters are generally actionable: Trigger word, 50 steps, CFG 3, and strength 0.8-1.2 are provided; the author states it can be used with acceleration models but will affect results, and steps should remain above 8 when accelerating.
  • 7 visible example images; free, license allows merging and renting, attribution optional; approximately 10 hours after release, it has 64 downloads (indicating attention only).

How to Use

  • Load moss_90sAnime.safetensors (71 MB), write "90sAnime" in the prompt; 50 steps, CFG 3, strength 0.8-1.2.
  • When stacking acceleration LoRAs, steps should not be less than 8; the author states the image will be affected; sampler, scheduler, and resolution are not provided by the author.

Risks and Limitations

  • The style description is only one sentence; training data and applicable subjects are not specified, nor is it stated whether it was trained on Base or SFT.
  • Sampler and scheduler are not provided by the author.
  • 1 sensitive-level examples are hidden; this item has not been tested.

Trigger Word: 90sAnime

[A] Furansumoe · MOSS Image

LoRA · Port · Author pope_phred · Published 10-31 22:59 · Downloads 27

MOSS version of a style LoRA blending 18th-century French portrait painting (Ducreux style) and anime: pastel blending, thick black ink outlines, warm-toned paper with paper texture, paired with large, bright anime eyes.

Maintained at A: Style definition is clear, prompts are directly copyable, and the license is permissive; sampling parameters are missing, and training data consists of AI-generated images.

Images

Example 832x1248

Reasons for Recommendation

  • Port: A Krea 2 version previously existed; the MOSS version is retrained using images generated by the author's own Ducreux LoRA, Anime Flat Style, and MOSS Semi-Realistic Anime 2.5d.
  • Serves illustrations that want a classical portrait painting texture in MOSS while retaining anime facial features and hairstyles; the author has published the full style prompt used to generate the training images, which can be used directly as a prompt prefix.
  • The author states it works without a trigger word, but adding the trigger word makes anime features more prominent; for full effect, it is recommended to add a sentence describing pastel strokes before the subject description.
  • 8 visible example images; free, site license is fully open; approximately 10 hours after release, it has 27 downloads (indicating attention only).

How to Use

  • Load furansumoe-MOSS.safetensors (138 MB), write "Furansumoe" in the prompt, and add "Worked entirely in rough, visible pastel strokes on toothy toned paper, with clear dark ink outlines" before the subject.
  • Strength starts to take effect around 0.8, is stable between 0.8 and 1.2, and issues arise if higher; steps, CFG, and sampler are not provided by the author.

Risks and Limitations

  • Training images are all AI images generated by other LoRAs, resulting in a secondary distillation of the style.
  • Generation parameters are not provided by the author.
  • 12 sensitive-level examples are hidden, exceeding the number of visible examples; this item has not been tested.

Trigger Word: Furansumoe

[A] MOSS De-Render // Photoreal Negative Slider · v1.0

LoRA · New Model · Author TestingAiImages · Published 11-01 00:25 · Downloads 76

Reverse slider LoRA: Uses negative weights to remove the plastic-looking skin and game-like rendering light from MOSS after stacking 4-step/8-step Turbo LoRAs. The same version includes a positive weight sharpness compensation LoRA.

Downgraded from S to A: Addresses specific pain points in MOSS accelerated image generation, with detailed usage instructions; however, there are only 2 example images, making it impossible to compare effects, and the two sliders must be manually balanced.

Images

Example 832x1248

Reasons for Recommendation

  • New model: Addresses the issue where distillation-accelerated image generation with MOSS results in poreless skin and CGI-doll-like portraits; the author states it is a reverse patch trained against real photos while running Turbo LoRA.
  • Two-piece set: Set De-Render to negative values (-0.8 to -1.2) to remove the rendered look, and SharpDetail to positive values (+0.6 to +1.0) to restore edge sharpness; the author provides starting points for light, medium, and heavy settings, along with LoRA stacking order.
  • Turbo can also be disabled: The author states that using lighter negative values on standard MOSS acts as a "de-beautify" tool.
  • 2 visible examples; free, license allows merging; released about 9 hours ago with 76 downloads (indicating attention only).

How to Use

  • The version contains two files: MOSS De-Render (16 MB) with strength -0.8 to -1.2, and MOSS SharpDetail (16 MB) with strength +0.6 to +1.0; the author recommends loading both via LoraLoaderModelOnly, modifying only the DiT and leaving text encoding unchanged.
  • Paired with Turbo LoRA: 6 to 8 steps, CFG 1.0, euler + simple, EmptyFlux2LatentImage, flux2-vae.safetensors.
  • Optional prompts: real photograph, candid shot, authentic skin texture, natural ambient lighting, raw unedited photo.

Risks and Limitations

  • The author states that if the ratio between the two sliders is unbalanced, image quality will degrade, and the balance point shifts when stacking other LoRAs, requiring repeated adjustment; the author describes this as experimental.
  • Only 2 visible examples are provided; another 2 sensitive-level examples are hidden, so before-and-after comparisons are not visible.
  • Depends on a Turbo SFT LoRA, but the author does not specify the exact file; strength and step counts for non-Turbo usage are only documented in the full model card.
  • Trigger words are not registered in the version information; this entry has not been run.

[A] Da Vinci Drawing style - CE · V01 - MossImage Base/SFT

LoRA · Port · Author CreativeEdge · Published 11-01 02:24 · Downloads 52

A MOSS version of the Da Vinci manuscript sketch style LoRA: The author trained it on MOSS-Image Base, stating that both Base and SFT are usable, though all showcase images were generated using SFT.

Maintaining A rating: Purpose is immediately clear, strength range is actionable, and examples are sufficient; lacks sampling parameters and trigger word documentation, preventing an S rating.

Images

Example 768x1024

Reasons for Recommendation

  • Port: The same series previously had versions for SDXL, Flux.1 D, OmniGen, Qwen, Z-Image Base/Turbo, Flux.2 Klein 9B, and Krea 2, with the entire model accumulating 6669 downloads (indicating attention only); the difference for the MOSS version is that the training base is Base, image generation is recommended using SFT, and no trigger words are registered (versions like Krea 2 / Flux.1 D have trigger words such as dvnciCE_style).
  • Serves users who want to render characters and scenes in a Renaissance manuscript sketch illustration style within MOSS, such as for concept sketches or retro illustrations.
  • Strength is clearly defined: 0.9 is suitable for most images, adjustable between 0.5 and 1.2 based on desired style intensity; 7 visible examples, free.

How to Use

  • Load DaVinciDrawing01_CE_MossImg_AIT3k.safetensors (138 MB · BF16); the author recommends using MOSS-Image SFT for generation; no trigger words are registered, so describe the scene using the example prompts.
  • Start with strength 0.9, range 0.5 - 1.2; steps, CFG, and sampler are not provided by the author.

Risks and Limitations

  • The MOSS version description is only two or three sentences, without stating whether trigger words are required; generation parameters are not provided by the author.
  • The license does not allow merging, only image generation and on-site rental.
  • 2 sensitive-level examples are hidden; this entry has not been run.

[A] Oddling Reverie · MOSS Image

LoRA · Port · Author Wackiii · Published 11-01 00:10 · Downloads 50

A surreal fantasy photography style LoRA: Renders subjects as cinematic images featuring quirky small creatures, large eyes, and fluffy textures; the author describes the color tone as warm amber highlights paired with deep teal shadows and shallow depth of field.

Maintaining A rating: The style definition is the most specific in this group, examples are complete, and the license is permissive; lacks strength and sampling parameters.

Images

Example 1024x1024

Reasons for Recommendation

  • Released on the same day as the Krea 2 version: In the API, the Krea 2 version has a newer version number than the MOSS version, indicating the MOSS version was not ported from an older model; both versions share the same description.
  • Serves users creating cute monsters, fantasy creatures, surreal portraits, dreamscapes, and character design sheets; the author breaks down the style very specifically: exaggerated expressions on strange creatures, fine textures like hair fibers, fabric, and moss, cinematic lighting, and surreal skies and scenes.
  • 7 visible examples, no hidden examples; free, license allows image generation, rental, and sale (attribution required); released 9 hours ago with 50 downloads (indicating attention only).

How to Use

  • Load oddlingReverie_v1.safetensors (138 MB); no trigger words are registered, so use natural language to describe creatures, scenes, and lighting.
  • Strength, steps, CFG, and sampler are not provided by the author.

Risks and Limitations

  • The description consists entirely of style details, with no generation parameters provided.
  • The license does not allow merging, and usage requires attribution.
  • This entry has not been run; image quality is based solely on the author's description.

Worth Watching (22)

[B] Portrait Engine FLUX / FLUX 2 KLEIN / Z-Image, Detailed Skin - [LORA] · Moss_V1

LoRA · Port · Author The_Midnight_Lab · Published 10-31 22:07 · Downloads 169

The MOSS version of the realistic portrait skin detail LoRA "Portrait Engine": The author claims it produces pores, skin texture, eye reflections, and cinematic lighting, removing plastic or wax-like skin appearances.

Images

Video 1024x1920

[B] Niji Ultraviolet [Anima/Krea2/MOSS] · MOSS v1.0

LoRA · Port · Author Imagine__That · Published 10-31 22:35 · Downloads 73

The MOSS version of the "Niji Ultraviolet" art style LoRA: The author describes it as a thick-paint Niji style with a glowing effect, excelling at glossy objects and deep shadows.

Images

Example 1024x1536

[B] Kuro CAT: Black CAT Styles | 黒猫 [MOSS & IL] · Shadow Magic MOSS

LoRA · Port · Author CitronLegacy · Published 11-01 07:19 · Downloads 8

The MOSS version of the "Shadow Magic" style from the "Kuro CAT" black cat art style series; the author did not describe the style in text, providing only trigger words and example images.

Images

Example 832x1216

[B] FREE!! 免费！！MOSS Pro Grade | Professional Workflow with Image Edit 🖼️, 🥅 Controlnet, and all the other stuff · V6.0 Weighted Prompts

Workflow · Version Update · Author lonecatone23 · Published 11-01 05:02 · Downloads 38

An advanced MOSS ComfyUI workflow heavily reliant on custom nodes (including image editing, ControlNet, inpainting, skin refinement, etc.); this version adds prompt weighting.

Images

Example 2236x1402

[B] Cyberpunk 50s · MOSS Image

LoRA · Port · Author Granddyser · Published 10-31 21:56 · Downloads 79

The MOSS version of the 1950s retro-futurism (Atomic Age Cyberpunk) art style LoRA: featuring chrome prosthetics, neon diners, tail-finned hovercars, and Kodachrome film tones.

Images

Example 1088x1936

[B] iGEN One · Nova.3_MS

Workflow · Version Update · Author Rvage · Published 11-01 06:16 · Downloads 22

The MOSS-annotated version of the iGEN One modular ComfyUI workflow: Chains prompt assembly, generation, upscaling, Qwen/Krea2 editing, face swapping, tiled upscaling, two refiners, and watermarking into a single pipeline; this version is the M variant streamlined into a two-row layout.

Images

Example 2912x3904

[B] Amelicart [storybook] style · v3 - M0SS

LoRA · Port · Author Fatbuns · Published 11-01 05:31 · Downloads 32

The MOSS version of the Amelicart storybook illustration style LoRA; the model card only says "Hello," so the style can only be judged by the examples.

Images

Example 1680x2520

[B] Women's IFBB Wellness/Fitness & Superheroine Muscle LoRA/LoKR · v2.0 MOSS

LoRA · Version Update · Author TestingAiImages · Published 11-01 01:22 · Downloads 61

The 2th MOSS version of the female bodybuilding (IFBB Wellness division) muscle physique LoRA: Enhances lower body muscle, striations, and vascular details, while keeping the upper body athletic; also suitable for female superhero settings.

Images

Diagram 1254x1254

[B] Realism By Stable Yogi Moss · v1.5 INT8

Base Model / Checkpoint · Version Update · Author Stable_Yogi · Published 10-31 21:30 · Downloads 661

The INT8-ConvRot quantized file for the photorealistic MOSS fine-tuned base model V1.5: Focuses on casual phone snapshots, selfies, full-body portraits, and crowd scenes; the author states this version offers more stable poses in complex scenes, cooler and more neutral tones, and smoother skin.

Images

Diagram 1254x1254

[B] sky05 2.5D Fantasy Style [Anima/MOSS] · v1.0 Test [MOSS]

LoRA · Port · Author skyddw · Published 11-01 00:41 · Downloads 58

The MOSS test version of the sky05 series 2.5D fantasy illustration style LoRA; the author only provided strength and a non-commercial use statement.

Images

Example 1280x1920

【B】kok (Takenawa) Style | Pastel Anime Illustration [Anima / MOSS Image] · MOSS Image

LoRA · Port · Author marty · Published 11-01 07:03 · Download 39

MOSS version of the LoRA for artist kok (Takenawa)'s style: anime illustration style with thin, light lines, pastel color palettes, soft shadows, and glossy highlights.

Image

Illustration 1672x941

【B】Storyflow Character Card · MOSS Reference v1.0 - Single Pass

LoRA · Version Update · Author Coffmeteor · Published 11-01 06:21 · Download 40

Reference-image-driven four-panel character sheet LoRA: select one face and one outfit to generate a character card with portrait, front, back, and other views in a single sampling pass, then replace individual parts in the workflow for regeneration.

Image

Illustration 1448x1086

【B】Manhwa style (Solo Leveling) · MOSS Image

LoRA · Port · Author shnack · Published 11-01 05:46 · Download 25

MOSS version of the LoRA for the manhwa Solo Leveling style, used solely for style transfer without character tags.

Image

Example 1024x1536

【B】Moss/Anima-Stylr-Blender3D Rendering (nai5 Eight) · Moss-base-V2.0

LoRA · Version Update · Author hardworkingsaltedfish · Published 10-31 21:17 · Download 151

MOSS V2: Blender 3D Rendering Style LoRA: The author claims it can consistently render anime characters with the smooth, reflective skin texture characteristic of Blender-modeled and rendered 3D visuals, using training images generated by the author with NovelAI 5.

Image

Example 1024x1536

【B】Nagasaki_Soyo_MOSS_lora · v1.1

LoRA · New Model · Author yueyinganshuang · Published 10-31 23:49 · Download 28

Character LoRA for Nagasaki Soyo from BanG Dream! MyGO!!!!!, including copyable prompts for four outfits: summer uniform, winter uniform, stage outfit, and casual wear.

Image

Example 1024x1344

【B】Friendly Ghost (Super Mario) [ANI & IL & MOSS] · MOSS

LoRA · Port · Author CitronLegacy · Published 11-01 07:10 · Download 12

MOSS version of the Halloween series concept LoRA "Friendly Ghost": based on the tags, it is a concept for upper-body selfies featuring characters alongside Boo, the ghost from Mario.

Image

Example 832x1216

【B】[Moss] Wsman style · v1.0

LoRA · New Model · Author aki_11679 · Published 11-01 02:56 · Download 57

MOSS style LoRA replicating artist Wsman's style, triggered by "wsman style"; the author provided complete sampling parameters and specifically warned against using the er_sde sampler.

Image

Illustration 1086x1448

【B】MOSS Toonforge · v1.0

LoRA · New Model · Author _Envy_ · Published 11-01 04:48 · Download 25

Exaggerated, manic style cartoon LoRA on MOSS, pushing the image towards an American toon direction with highly exaggerated expressions and actions.

Image

Example 1200x1600

【B】Blue hour · v1.0

LoRA · New Model · Author blacktone · Published 11-01 03:38 · Download 33

Based on the title and filename, this is a "Blue Hour" atmosphere LoRA: adds a twilight atmosphere with blue skies after sunset; the author only specified the strength range.

Image

Example 832x1248

【B】moss lumbar lordosis posture ass out · v1

LoRA · New Model · Author hamstornado · Published 11-01 03:19 · Download 6 · Paid

Character pose concept LoRA: places the character in a lumbar lordosis standing pose—with hips pushed back, torso leaning forward, shoulders back, and head forward.

Image

Example 768x1024

【B】getsuki style [MOSS Image] · v1.0

LoRA · New Model · Author marty · Published 11-01 07:12 · Download 29

LoRA for artist getsuki (ge.tsuki)'s style, tagged as thick paint/painterly, trained on MOSS Image Base using 17 curated works, requiring no trigger word.

Image

Example 1440x2160

【B】moss upright-kneeling · v1

LoRA · New Model · Author hamstornado · Published 11-01 02:24 · Download 1 · Paid

Character pose concept LoRA: places the character in an upright kneeling pose (knees on the ground, thighs and torso straight), combinable with hands on head, arched back, and knees apart.

Image

Example 768x1024

Included Only (16)

Data: Civitai public API (base model MOSS Image), collection window 10-31 20:31 → 11-01 09:01 (Beijing Time). Recommendation reasons have been reviewed by editors; download counts only reflect attention levels.

Civitai

Explore the latest AI Radar updates