MOSS Image Morning Report 2026-11-19 | Top Pick: Sint Albedo · Tileable Game Textures - Krea/MOSS

This issue reviews 44 new versions. Top pick: Sint Albedo 1.1: input a material description to generate top-down, unlit game diffuse maps; after retraining, some materials can tile seamlessly on all four sides. Other highlights include the AlchemicMoss workflow with built-in prompt expansion, a region-specific workflow that restricts LoRA effects to individual boxes, and UltraReal V3 which removes the plastic look. Standard character LoRAs and art styles with all examples hidden are downgraded to C; the 8-step acceleration LoRA is a third-party repack, and the AuthentiQ 3D Environ model page has been 404, both listed as C.

Zhenzhen's AI Radar · MOSS Image · 2026-11-19 Morning Report

This issue features 41 new versions: Strongly Recommended 1 · Recommended 12 · Worth Watching 10 · Included Only 18

Editor's Note This issue reviews 44 new versions. Top pick: Sint Albedo 1.1: input a material description to generate top-down, unlit game diffuse maps; after retraining, some materials can tile seamlessly on all four sides. Other highlights include the AlchemicMoss workflow with built-in prompt expansion, a region-specific workflow that restricts LoRA effects to individual boxes, and UltraReal V3 which removes the plastic look. Standard character LoRAs and art styles with all examples hidden are downgraded to C; the 8-step acceleration LoRA is a third-party repack, and the AuthentiQ 3D Environ model page has been 404, both listed as C.

Strongly Recommended (1)

[S] Sint Albedo · Tileable Game Textures - Krea/MOSS · MOSS 1.1

LoRA · Version Update · Author AriaMaiven · Published 11-19 03:43 · Downloads 36

Turn MOSS into a game texture machine: input a material description, and it outputs top-down, unlit diffuse (albedo) maps; in version 1.1, after retraining, the edges of some materials can connect seamlessly.

Images

Example 2048x2048

Reasons for Recommendation

  • The version update is very specific: 1.0 can produce tileable unlit maps but the edges don't connect; 1.1 was retrained by placing seams in the center of half the training images, teaching the model to extend textures across boundaries
  • The author checked full-size tiling at 2x2 and reported by category: lava and cracked clay connected in both test seeds (only a thin line remains); stone, crystal, and chainmail connected in one of two seeds; herringbone flooring and hand-painted grass still have breaks
  • Serves game art and 3D material creation: no baked highlights, shadows, or reflections; camera is vertically downward with no perspective, ready to use as a diffuse map in engines; parameters are complete, free

How to Use

  • Start the prompt with the trigger word sintalbedo, then describe the material and variations, e.g., sintalbedo, cooled lava rock, black porous basalt with glowing orange cracks
  • Strength 1.0 (0.8 retains more dimensionality, 1.2 flattens it more); MOSS Image SFT, 30 steps, CFG 3, euler / simple, 1024 x 1024, use squares only
  • Scale defaults to small, so include scale words like large stones or close-up; for regular patterns (floors, sci-fi panels), run through a tiling node after generation

Risks and Limitations

  • Outputs only one albedo map; normal, roughness, and height maps must be generated with other tools
  • Seamless results depend on material and seed; the author only checked with two seeds; wide aspect ratios break tiling
  • 5 visible examples (1 sensitive-level examples hidden); the author requires permission for redistribution or using outputs to train models

Trigger Word: sintalbedo

Recommended (12)

[A] Takeuchi Takashi - Artist Style · MOSS-Image

LoRA · Port · Author Azriaelis · Published 11-18 23:40 · Downloads 69

A LoRA mimicking the art style of Takeuchi Takashi (original artist for Fate / Type-Moon series); the MOSS version requires pairing with a 8-step acceleration LoRA on the Base model.

Images

Example 1024x1536

Reasons for Recommendation

  • Port: The same page already has Anima and Krea 2 versions, with 1159 total downloads for the model page; the MOSS version lacks version notes, and the author hasn't specified performance differences compared to other base models
  • Serves anime character illustrations in the Type-Moon style; the author provided parameters for each base model, with the MOSS column specified separately, so you can follow them
  • 14 visible examples; free download

How to Use

  • MOSS usage (provided by author): Base model + 8-step turbo LoRA, euler / simple, LoRA strength 0.8-1, 8-16 steps, CFG 1-3
  • Trigger word takeuchi_takashi; the author claims the style becomes significantly stronger when added

Risks and Limitations

  • Mimics the style of a living artist; the license does not allow derivatives
  • Depends on an additional 8-step acceleration LoRA, but the author didn't specify which one; 6 sensitive-level examples are hidden

Trigger Word: takeuchi_takashi

[A] UltraReal - Moss, Krea2, Klein9b · MOSS_V3

LoRA · Version Update · Author vizsumit · Published 11-18 21:44 · Downloads 291

Third version of the realistic enhancement LoRA that removes the AI plastic look: adds skin texture and clothing details to portraits, reducing the soft and slippery tendency of MOSS outputs, most noticeable in close-ups and medium shots.

Images

Example 2368x1760

Reasons for Recommendation

  • Version update: The author states in the model card version comparison that MOSS_V3 was trained with 3 times more SFW data than the previous version (primarily people), resulting in better composition and color, and fixing minor anatomical flaws in MOSS SFT without bias toward specific ethnicities
  • Serves realistic portraits, fashion, and lifestyle photos; no trigger word needed, active upon loading, strength range provided by author
  • Most permissive license tier (allows selling models and merges); 14 visible examples; 11 hours since release with 291 downloads, 55765 total downloads for the model page (indicates attention level)

How to Use

  • Load the LoRA; the MOSS version requires no trigger word; strength 0.5-0.8
  • Steps, CFG, and sampler not provided by author; the author has published their ComfyUI workflow on GitHub for reference

Risks and Limitations

  • The version update section is empty; changes are only described in the model card's version comparison section, all author claims without same-seed comparison images
  • The author stated MOSS_V2 is not good at artistic drawings and complex text; whether V3 improved this is not specified; 6 sensitive-level examples are hidden

[A] ⚗ AlchemicMoss — MOSS-Image workflow (prompt by its own Qwen3.5 encoder, framing presets, calibrated sampler) · v1.0.1 · AlchemicMoss_T2I

Workflow · New Model · Author Nessuno17 · Published 11-18 22:13 · Downloads 31

MOSS text-to-image ComfyUI workflow + custom node package: uses MOSS's built-in Qwen3.5-9B text encoder to expand a single idea into positive and negative prompts, includes framing/camera/lighting presets and a sampler calibrated to the base model.

Images

Example 1024x1536

Reasons for Recommendation

  • New tool (workflow and nodes, not new weights): generate images from a single idea in any language; prompt expansion is handled by MOSS's loaded Qwen3.5-9B, requiring no Ollama or external services
  • Addresses MOSS's non-distilled nature and need for true CFG: negative prompts use CFG 2.5; when rendering text, use double quotes and automatically remove text/letters negative prompts to avoid conflicts
  • Includes a calibration process: measures how your base model generates images and writes out sampling configuration; built-in configuration for moss_image_sft_int8_convrot (tested with 12 prompts); 10 examples all visible, free

How to Use

  • Requires ComfyUI 0.42+ and custom node ComfyUI-AlchemicMoss 1.0.1 (install via Manager or GitHub); models use moss_image_sft_int8_convrot (or bf16) from Comfy-Org/MOSS-Image, qwen3.5_9b_w4a8_gptq text encoder, and flux2-vae
  • Open AlchemicMoss_T2I, write your idea in the Prompt node, and Queue; the author claims approx. 21 seconds for 1.5 MP, 25 steps on an RTX 5070 Ti 16 GB, with longer times for prompt writing on first run
  • When switching to bf16, fine-tuned base models, or adding LoRAs, run AlchemicMoss_CALIBRATION first to generate new configuration; the author suggests CFG 2.5-3, 25-50 steps

Risks and Limitations

  • Depends on third-party custom nodes (GPL-3.0), not run by this column; license does not allow derivatives
  • Author tip: Do not add --use-sage-attention when launching ComfyUI, otherwise MOSS outputs will look gray; SFT base models have vertical stripes in skies and walls, requiring a separate DeBand LoRA

[A] Medieval Worlds : The Crypt - MOSS · V1.5

LoRA · New Model · Author Z1os · Published 11-19 00:10 · Downloads 40

Medieval dark underground scene LoRA: Gothic churches, corridors, crypts, ruins, plus a necromancer character.

Images

Example 896x1152

Reasons for Recommendation

  • New model: serves scene concept art and backgrounds for dark fantasy games/tabletop games, clear direction
  • 14 visible examples; the author suggests writing prompts based on the example images
  • Trigger word, strength, and steps provided (tested by author with turbo acceleration)

How to Use

  • Trigger word z1os; strength 1, 10 steps (tested by author with turbo)
  • CFG, sampler, and which acceleration LoRA to use not provided by author

Risks and Limitations

  • Strict license: no commercial use, merging, or retraining without author consent; generated images cannot be sold
  • Author claims it's a model made "for fun, to see if MOSS can handle it"; 1 sensitive-level examples are hidden

Trigger Word: z1os

[A] Dizzle's Punk Sticker · v1

LoRA · New Model · Author Maridizzle · Published 11-19 03:56 · Downloads 26

Punk sticker style LoRA: black ink line art + large white space + only one neon accent color, making any subject look like a hard-edged sticker suitable for a skateboard.

Images

Example 1024x1024

Reasons for Recommendation

  • New model; effect rules are clear: one fluorescent accent color, specify the color by writing color words like pink or cyan in the prompt
  • Serves graphic design for stickers, T-shirts, skateboard patterns, etc.; author's examples range from animals to people
  • Trigger word and strength provided; 5 visible examples; free

How to Use

  • Start prompt with dz_punk, then describe subject and scene; strength 1.0
  • To change the accent color, add color words (pink, cyan, etc.); steps and CFG not provided by author

Risks and Limitations

  • Description is only one paragraph, no parameter details; 4 sensitive-level examples are hidden

Trigger Word: dz_punk

[A] Ragnarok Online Sprite - Pixel Art - Style MOSS LORA · v1.0

LoRA · New Model · Author Konan · Published 11-19 00:33 · Downloads 78

A pixel art LoRA in the style of character sprites from the game Ragnarok Online, producing retro pixel character portraits.

Image

Example 1280x1840

Reasons to Recommend

  • New model; serves game pixel characters, avatars, and nostalgic illustrations, with starting prompts and strength ranges provided.
  • Published 8 hours ago with 78 downloads and 14 likes (indicating attention only); free.

How to Use

  • Start the prompt with "Retro pixel art character sprite"; strength 0.8-1.3.
  • Steps, CFG, and resolution are not provided by the author; whether it outputs strictly aligned pixel grids is unspecified, so you may need to downsample and clean up for use as assets.

Risks and Limitations

  • Only 2 visible examples, with another 4 sensitive-level images hidden; mimics a specific game art style.

Trigger Words: Retro pixel art character sprite,

【A】AuthentiQ 3D: Thintoon Style [Krea2/MOSS] · MOSS_v1.0

LoRA · Ported · Author AuthentiQ · Published 11-18 21:12 · Downloads 46

A LoRA for the "Thintoons" 3D animation character style: extremely elongated and slender bodies, long necks, paired with large Disney-style heads and glowing big eyes.

Image

Example 1088x1936

Reasons to Recommend

  • Ported: Same page as the Krea 2 version (old version deprecated, new version just uploaded); the MOSS version filename indicates training on SFT, differences from the Krea version are not specified by the author.
  • Specific description of style features, serving stylized 3D animation character designs and posters; all 7 examples are visible.
  • Both trigger phrases are registered; free.

How to Use

  • Write "Premiun 3D Animation still" and "Thintoons character style" in the prompt (the registered spelling is indeed "Premiun").
  • Strength, steps, and CFG are not provided by the author.

Risks and Limitations

  • No parameters provided; license does not allow derivatives or selling the model.

Trigger Words: Premiun 3D Animation still, Thintoons character style

【A】Dizzle's 1950's Mens Magazine · v1

LoRA · New Model · Author Maridizzle · Published 11-19 02:37 · Downloads 58

A LoRA in the style of 1950s men's magazine illustrations: deep teal and cherry red backgrounds, flat large color blocks, and a vintage print gloss; adding one prompt line can create the fold crease of a magazine center spread.

Image

Example 1024x1024

Reasons to Recommend

  • New model; visual features are described specifically (background colors, flat painting, print feel), covering subjects like people, knights, dragons, lighthouses, animals, and still life.
  • There is a directly usable trick: add "a two page spread" to make a center fold crease appear, looking as if torn from a magazine.
  • Trigger words and strength are provided; free.

How to Use

  • Start the prompt with dz_1950, set strength to 1.0, and explicitly write the subject (data leans towards male work scenes, landscapes, and objects).
  • Add "a two page spread" for the spread effect; steps and CFG are not provided by the author.

Trigger Words: dz_1950

【A】Dizzle's Dark Color Splash · v1

LoRA · New Model · Author Maridizzle · Published 11-19 05:13 · Downloads 18

A dark color splash LoRA: turns the image into black and white ink lines, then lets one glowing color pop out from the darkness; you can specify this bright accent with "floating [color] [thing]".

Image

Example 1024x1024

Reasons to Recommend

  • New model; control method is clear: the trigger word works on its own, and adding "floating [color] [thing]" specifies the bright element.
  • The author lists floating objects the model recognizes: koi fish, jellyfish, moths, birds, flies, which can be used directly.
  • Serves posters, covers, and black-and-white illustrations to create visual focal points; free.

How to Use

  • Write the trigger word dz_dark in the prompt, then add phrases like "floating purple koi fish", "floating green jellyfish", or "floating orange birds".
  • Strength, steps, and CFG are not provided by the author (other LoRAs in the same series by this author specify full strength).

Risks and Limitations

  • Only 3 visible examples, with another 4 sensitive-level images hidden; published less than 4 hours ago.

Trigger Words: dz_dark

【A】BBox Studio · v1.4

Workflow · Version Update · Author plk · Published 11-19 04:37 · Downloads 24

A standalone desktop mini-program for drawing MOSS Image bounding box (bbox) structured prompts: drag boxes on the canvas, write descriptions for each box, and export JSON readable by MOSS; v1.4 adds configuration memory, a custom tag library, and batch conversion between JSON and PNGs with metadata.

Similar in concept to this issue's "LoRA inside boxes" Moss-image boxes workflow: one handles placing the boxes, the other handles applying different LoRAs to different boxes.

Image

Example 1322x892

Reasons to Recommend

  • Version update (version 5): The full version notes list new features in this version—customizable fonts and interface colors; using a config file to remember states like resolution and window size; storing custom tags in meta.txt, with the right panel showing currently enabled tags and allowing add/delete; two batch conversions added to the settings menu: converting a folder of JSONs into PNG previews with embedded JSON, or reverse-converting such PNGs back to JSON.
  • Serves multi-object, multi-character compositions: placing boxes on the canvas before generating is more intuitive than hand-writing bbox coordinates; after batch converting to PNG, you can browse your composition library by thumbnail in the file manager, and others' bbox JSONs can also be converted.
  • Pure Python single file, author claims zero external dependencies; code is public on GitHub; free, license fully open.
  • It is a tool, not weights: it does not generate images itself, requiring coordination with an image generation workflow that can read MOSS bbox JSON.

How to Use

  • Install Python 3, unzip BBox Studio.zip, then double-click the py file or run.bat to start.
  • Drag boxes on the canvas, write descriptions for each box, export JSON and paste it into a prompt node supporting MOSS bbox JSON; load existing JSONs via File > Import JSON.
  • The settings panel allows opening config and meta files directly with Notepad for manual editing.

Risks and Limitations

  • Only 1 visible examples are interface screenshots, with another 1 sensitive-level images hidden; image generation results depend on the MOSS workflow you use.
  • The API changelog field only excerpts one sentence, full changes require reading the version notes; this tool has not been run.

【A】Dizzle's Smoke Bomb · v1

LoRA · New Model · Author Maridizzle · Published 11-19 03:33 · Downloads 33

A colored smoke bomb effect LoRA: adds large, rolling clouds of high-saturation colored smoke to the image, like a smoke bomb exploding in the lens; applicable to people, creatures, objects, and landscapes.

Image

Example 1024x1024

Reasons to Recommend

  • New model: The author positions it as the first in the "material series"—the smoke itself is the style, not changing the subject but adding colored smoke to the scene.
  • Serves text-to-image generation for posters, fashion editorials, and atmospheric illustrations requiring strong color atmosphere; colors are controllable: write "blue colored smoke", "green colored smoke", etc. in the prompt to specify smoke color.
  • Trigger words, weights, and prompt order are specified; free; published about 5.5 hours ago with 33 downloads (indicating attention only).

How to Use

  • Start the prompt with dz_smoke, then write the subject and scene; weight 1.0.
  • Add "<color> colored smoke" for specific colors; steps, CFG, and sampler are not provided by the author.

Risks and Limitations

  • 3 visible examples, with another 5 sensitive-level images hidden; example image metadata shows er_sde, 25 steps, CFG 2.5, for reference only.
  • License does not allow selling generated images (only generation and rental are checked).

Trigger Words: dz_smoke

【A】Moss-image boxes with loras inside · v1.0

Workflow · New Model · Author saehara151 · Published 11-18 22:01 · Downloads 35

A MOSS Image text-to-image workflow: attaches a LoRA to each bbox box individually, so the LoRA only takes effect within its own box without polluting other boxes or the background, used for multi-character compositions.

The same author's Albert Wesker character LoRA uses this method of attaching LoRAs inside boxes to create two-person images in the examples.

Image

Example 1280x1024

Reasons to Recommend

  • New workflow: Solves the issue of cross-contamination when stacking multiple character LoRAs—each character's LoRA is limited to its respective box.
  • Implementation relies on the third-party node comfyui-moss-image-fancy-bbox, with the GitHub address provided by the author; the workflow filename also indicates wildcards.
  • Serves multi-character scenes, custom OCs, and copyrighted characters in the same frame; free, published about 11 hours ago with 35 downloads (indicating attention only).
  • The workflow JSON is not new weights.

How to Use

  • First install the comfyui-moss-image-fancy-bbox node, then import eto_Moss_T2I_BBOX_wildcards_lorainbox.json.
  • Write the character description in each box and attach the corresponding LoRA; steps, CFG, and required model files are not listed in the instructions, check the workflow itself.

Risks and Limitations

  • Only 1 visible examples, with another 2 sensitive-level images hidden; instructions are brief and do not explain how to set up boxes.
  • Example image metadata shows use of the third-party base model MOSS-FET_v05, Euler, 30 steps, CFG field 38.4 (suspected internal node value), verify yourself before reproducing.
  • License does not include commercial use of generated images; relies on third-party nodes, this tool has not been run.

Worth Watching (10)

【B】Rui Komatsuzaki Style V2+ [Danganronpa Style] · Moss-Image v2.0

LoRA · Ported · Author emerjk · Published 11-19 06:35 · Downloads 21

A MOSS version of the Rui Komatsuzaki (Danganronpa original artist) style LoRA, allowing switching between sub-styles like Danganronpa portraits, anime screenshots, Rain Code, Akudama Drive, and Hundred Line using branch prompts.

Image

Example 896x1152

【B】Dizzle's Space Dust · v1

LoRA · New Model · Author Maridizzle · Published 11-19 03:42 · Downloads 27

A space dust effect LoRA: sprinkles twinkling stars and flowing nebula colors on the subject, as if dusted with a layer of galaxy powder.

Image

Example 1024x1024

【B】Dizzle's 1940's Pin-Up Photography · v1

LoRA · New Model · Author Maridizzle · Published 11-19 01:13 · Downloads 47

A 1940s American pin-up photography style LoRA: turns the image into a mid-century poster/calendar photo texture, with the author claiming best results for animals, landscapes, and objects.

Image

Example 1024x1024

【B】ComfyUI inpaint workflow | Updated for MOSSimage with guide · Moss V1.1

Workflow · Version Update · Author PotatCat · Published 11-19 02:38 · Downloads 41

PotatCat's established ComfyUI inpainting workflow, MOSS Second Edition: adds generation by selected region and spectrum acceleration.

Image

Example 636x818

【B】Cat Matroskin & Cheburashka [MOSS / Krea 2 · v2.0 MOSS

LoRA · Version Update · Author bazya022020810 · Published 11-19 01:31 · Downloads 19

A LoRA for plush toy/live-action movie-style characters of Matroskin the Cat (grey-blue striped fur, red-green striped knitted scarf) from the Soviet animation "Prostokvashino"; the v2.0 MOSS version adds Cheburashka (cheburashka_classic), allowing both characters to appear in the same frame.

Image

Example 832x1248

【B】Bruce Timm - Naughty and Nice: The Good Girl Art of Bruce Timm - Style MOSS LORA · v1.0

LoRA · New Model · Author Konan · Published 11-19 00:13 · Downloads 46

A style LoRA mimicking the American "good girl art" pin-up illustration style from Bruce Timm's art book "Naughty and Nice".

Image

Illustration 1448x1086

【B】Eoneum/hh48288196 Style - MOSS Image v1 · v2.0

LoRA · Version Update · Author inpaint · Published 11-19 08:59 · Downloads 3

The second version of the LoRA for the style of Korean artist Eoneum (hh48288196): the author trained using a mix of original artwork and AI-composited images, aiming to reduce the biases present in the first version.

Image

Example 832x1216

【B】Dizzle's Art Deco Peirrot Era · v1

LoRA · New Model · Author Maridizzle · Published 11-19 02:01 · Downloads 25

An Art Deco print style LoRA: flat ink lines, color blocks, low-saturation color schemes, and cream-colored borders make any subject look like an old-fashioned Art Deco print.

Image

Example 1024x1024

【B】[MOSS] King of Dragon Pass Style · v1.0

LoRA · New Model · Author Selenusuka · Published 11-18 22:58 · Downloads 27

A LoRA in the hand-drawn illustration style of the old game "King of Dragon Pass": the author trained on 195 original game artworks, trying to see if MOSS could learn non-anime art styles.

Image

Example 1024x1024

【B】Darkness Slider · Moss-image

LoRA · Port · Author Maruhashi · Published 11-18 21:03 · Downloads 0 · Paid

A slider LoRA for adjusting image darkness, ported from the Anima / Illustrious version to MOSS; it requires no trigger words and is controlled by weight.

Image

Example 2496x2796

Included Only (18)


Data: Civitai Public API (Base Model MOSS Image), inclusion window 11-18 20:31 → 11-19 09:01 (Beijing Time). Recommendations are reviewed by editors; download counts only reflect attention.

Civitai

Explore the latest AI Radar updates