Zhenzhen's AI Radar · MOSS Image · 2026-11-14 Morning Report
This issue features 40 new versions: Strongly Recommended 2 · Recommended 9 · Worth Watching 14 · Included Only 15
Editor's Note This issue reviewed 43 new MOSS versions: 2 S-tier, 9 A-tier, 14 B-tier, 17 C-tier, with 1 excluded. Top pick: HaDeS MOSS Turbo v1.0: a full model starting from MOSS Base, featuring built-in 8-step acceleration and a refined LoRA; after loading, use CFG 1 with euler / simple for direct image generation. Another S-tier entry is the Ad Infinitum full model, which merges 1885–1948 year-old advertising layouts into SFT; SILBRIO’s woodcut relief, neon print, and bright anime ports, Dizzle’s sky clock and beach film, Heikala watercolor, Camio brush fantasy, cross-eye expressions, and MossYume large-scale fine-tuning are listed as A-tier. Character LoRAs in the automatic S-tier, V67 models lacking MOSS-specific parameters, and experimental Murata Range style downgrades were demoted; BBox Studio v1.1 was superseded by concurrent v1.2 and excluded after merging.
Strongly Recommended (2)
[S] HaDeS MOSS Turbo · v1.0 BF16
Base Model / Checkpoint · Version Update · Author _HaDeS_ · Published 11-14 07:07 · Downloads 29
HaDeS MOSS Turbo v1.0: A full model starting from MOSS-Image Base, featuring built-in 8-step Turbo LoRA and the author's proprietary QwenLook refinement LoRA. After loading, it generates images directly in 8 steps without needing to attach an acceleration LoRA.
Can directly replace the SFT + Turbo LoRA combination for quick image generation; when comparing with v0.9.4 on the same page, use the same prompt and seed.
Images

Reasons for Recommendation
- Version update (relative to v0.9.4): The author states that blind tests were conducted using one seed per prompt across 20 new prompts; in 16 out of 20 pairs, v1.0 was selected. Topics covered photos, blackboards with text and infographics, comics, pixel art, clay, ink wash, cyberpunk, and children's books.
- The author claims reduced appearance bias: For a "photo of a woman in a cafe" without specifying ethnicity, approximately 1 out of 16 seeds produced East Asian faces, compared to 9–11 for v0.9.4; QwenLook makes details sharper and outputs cleaner.
- Suitable for general text-to-image scenarios where users want finished products with fewer steps; 18 visible examples are provided, and the author states that sample images include generation parameters, along with a text-to-image workflow using only core nodes.
- Free; license allows derivatives and commercial use of outputs.
How to Use
- Place the full model in models/diffusion_models and load it using Load Diffusion Model; additionally configure MOSS's Qwen text encoder (select moss_image for Load CLIP type) and flux2-vae.
- KSampler with 8 steps, CFG 1, euler / simple; the author recommends 1344×1792; do not add Turbo LoRA as it is already built-in.
- The author suggests writing long and specific prompts: light source direction, camera lens, scene, fine imperfections; for a rough documentary feel, specify the camera and lighting in the prompt.
Risks and Limitations
- The file contains only the diffusion model (BF16, 10.16 GB), excluding text encoders and VAE, which must be downloaded separately.
- Blind tests were selected by the author personally with one seed per question; the author notes that other step counts, CFG greater than 1, other sizes, and int8 versions have not been tested.
- Women generated without character descriptions remain very similar (young, long brown curly hair, smiling); characters must be clearly specified in the prompt.
[S] MOSS Ad Infinitum Press · Vintage Ad Posters, 1885-1948 · MOSS 1.0 Classic
Base Model / Checkpoint · New Model · Author AriaMaiven · Published 11-14 02:13 · Downloads 20
Ad Infinitum Press · Classic: A full model merging the layouts of 1885–1948 year-old commercial advertisements (trade cards, posters, magazine ads) into MOSS-Image SFT—featuring stippled colors, ink outlines, cream-colored paper, and color blocks reserved for slogans.
The author also offers a Krea 2 / MOSS LoRA version of the same style (Ad Infinitum); use that if you want to layer it over other MOSS fine-tunes.
Images

Reasons for Recommendation
- New model: The style is directly merged into SFT weights, requiring no LoRA attachment and having no trigger words; the author states that writing period scenes (ladies in feather hats, men in striped vests, steam locomotives) will directly produce ad cards.
- Suitable for vintage advertising posters, product cards, and layouts with title text; the author provides typography instructions: write each line of text in quotes at the end of the prompt specifying position and font, or simply write "leave a blank cream-colored area at the top" to reserve space for later typesetting.
- Complete parameters: 30 steps, CFG 3, euler / simple, with SkipLayerGuidance enabled at layer 22 for the first half of generation, size 1024×1536; 6 visible examples are provided, and the author states each includes prompts and parameters.
How to Use
- Load using Load Diffusion Model (same as official MOSS-Image SFT), text encoder qwen3.5_9b (type moss_image), VAE flux2-vae.
- 30 steps, CFG 3, euler / simple, SkipLayerGuidance at layer 22 for the first half, size 1024×1536; no trigger words.
- Modern objects (ramen, sneakers, phones) must start with "a vintage lithograph advertising poster of", otherwise they will still appear as photos; describe actions as poses ("holding a lit cigarette between two fingers" instead of "smoking"); trains and landscapes require title and slogan panels, otherwise the output tends toward oil paintings.
Risks and Limitations
- Only one int8 convrot file is provided (approx. 5.4 GB, author states it runs on 16 GB VRAM), with no BF16 version.
- The author declares that permission is required for redistribution or using its outputs to train other models; 3 examples are hidden.
Recommended (9)
[A] SILBRIO VIVID ANIME · MOSS Image Base
LoRA · Port · Author Silbrio · Published 11-14 02:01 · Downloads 26
MOSS version of SILBRIO VIVID ANIME: A thick-line 2D animation style, featuring flat coloring with cel shading, bright colors, and slight glowing accents.
Images

Reasons for Recommendation
- Port: The same author first released a Krea 2 version; the MOSS version was retrained on MOSS Image Base BF16 using 96 curated images. The version notes provide a separate MOSS image generation recipe (strength 1.0, 25 steps, CFG 2.5, ER-SDE / Simple), which differs from the Krea 2 version's 8 steps and CFG 1.
- Suitable for animating character portraits, street scenes, and fantasy landscapes; the author suggests defining a main color with few bright accents; for stronger flatness, write "clean contours", "flat colours", "cel shading".
- 10 visible examples, none hidden; license allows derivatives and commercial use of outputs.
How to Use
- Start the prompt with silbriovividanime, then describe subject, action, composition, and color scheme in natural language; strength 1.0, 25 steps, CFG 2.5, ER-SDE / Simple.
Risks and Limitations
- The model card body follows the Krea 2 version, listing the trigger word vividcolorsanimeneon, which is inconsistent with silbriovividanime registered for the MOSS version; refer to the version notes as authoritative.
- The author states that complex multi-person interactions are less stable than single clear poses; hands, contact points, and mechanical details need checking.
Trigger Word: silbriovividanime
[A] Dizzle's Mysticlock · v1
LoRA · New Model · Author Maridizzle · Published 11-14 00:07 · Downloads 26
Dizzle's Mysticlock: A massive brass clockwork mechanism suspended in the sky—gears, rings, armillary spheres, and pendulums, combining steampunk with celestial magic.
Images

Reasons for Recommendation
- New model; suitable for fantasy posters, concept scenes, and covers; the author suggests pairing with dramatic skies (storm clouds, starry nights, dusk) and placing small-scale reference objects like lighthouses, figures, or rooms below.
- Complete parameters: Trigger word dz_clock followed by mystical sky clockwork, strength 0.9–1.0, 30–50 steps, CFG 3, Euler flowmatch; use full sentences for description rather than tags.
- 7 visible examples (3 hidden).
How to Use
- Start the prompt with dz_clock, mystical sky clockwork, then describe the scene in full sentences; strength 0.9–1.0, 30–50 steps, CFG 3, Euler flowmatch.
Risks and Limitations
- The effect is a fixed element (clockwork in the sky), making compositions prone to repetition across multiple images; the author states it looks best suspended in open skies, with no guidance provided for indoor or sky-less compositions.
- Page contains mixed content.
Trigger Word: dz_clock,
[A] MossYume · v0.1
Base Model / Checkpoint · New Model · Author duongve13112002 · Published 11-14 00:33 · Downloads 205
MossYume v0.1: A full model fine-tune continuing training on MossImage-base, mixing anime and realistic images at a 1:1 ratio. The author's goal is to correct the original base model's anatomical structure issues and unstable weight distribution.
Images

Reasons for Recommendation
- New model; the author states it was trained on 20 million human-created images (excluding AI-generated images, no distillation), with anime parts sourced from Danbooru, e621, Gelbooru, Pixiv, and Konachan.
- The author claims native support for image generation up to 2048×2048, with training resolution dynamically varying between 1024–2048.
- 12 visible examples (7 hidden); 205 downloads, making it the most downloaded new version in this issue's non-sensitive tier.
How to Use
- Contains only the diffusion model; requires separate MOSS text encoder and VAE (download links provided by author), loaded using Load Diffusion Model.
- The author only suggests writing complex, detailed prompts; step count, CFG, and sampler are not specified by the author.
Risks and Limitations
- v0.1 is the base version without SFT / RL; the author describes it as a community experiment
- Parameters are missing; the commercial license scope only covers on-site rental and rental, excluding image generation; the page contains mixed content
【A】Dizzle's Beach Light Photography · v1
LoRA · New Model · Author Maridizzle · Published 11-13 22:05 · Downloads 43
Dizzle's Beach Light Photography: Adjusts images to a retro summer beach photography style—golden coastal light, tanned skin tones, soft film grain, and warm faded tones.
Images

Reasons for Recommendation
- New model; suitable for realistic photos of beaches, sunlight, outdoor portraits, and lifestyle themes; the author suggests slightly overexposing highlights to follow the warm faded tone
- Parameters provided: trigger word dz_beach followed by bright sunny photograph, strength 0.9–1.0, 30–50 steps, CFG 3, Euler flowmatch
- 5 visible examples (5 hidden)
How to Use
- Start the prompt with dz_beach, bright sunny photograph, then describe the scene in full sentences; strength 0.9–1.0, 30–50 steps, CFG 3, Euler flowmatch
Risks and Limitations
- Color grading direction is fixed (warm, faded, slightly overexposed); the author states it works best with natural daylight and coastal scenes, while indoor and night scenes are not specified
- Half of the examples are hidden; the page contains mixed content
Trigger Words: dz_beach
【A】SILBRIO WOODCUT · MOSS Image Base
LoRA · Ported · Author Silbrio · Published 11-14 02:04 · Downloads 26
MOSS version of SILBRIO WOODCUT: Carves subjects into reliefs on natural wood boards, where grooves, ridges, and wood grain form a single continuous carved plate.
Images

Reasons for Recommendation
- Ported: Original Krea 2 version exists; the MOSS version was trained on MOSS Image Base BF16 using 60 images for 1,800 steps. The version notes provide a MOSS recipe (strength 1.0, 25 steps, CFG 2.5, ER-SDE / Simple)
- Suitable for portraits, flora and fauna, architecture, landscapes, and fantasy scenes with woodcut relief textures; the author suggests adding the description 'An engraving on a wooden board, with moderate relief and depth', and specifying unpainted matte wood, side lighting, and where to carve deep or fine
- 10 visible examples, none hidden
How to Use
- Add silbriowoodcut and relief carving in a natural wood panel (both registered as trigger words) to the prompt, then describe the subject and composition; strength 1.0 (author states it can go up to 1.1), 25 steps, CFG 2.5, ER-SDE / Simple
Risks and Limitations
- License does not allow derivatives
- The 8 steps, CFG 1, and upscaling to 3072×4096 mentioned in the model card body are for Krea 2 Turbo showcase images and cannot be directly applied to the MOSS version
Trigger Words: silbriowoodcut, relief carving in a natural wood panel
【A】Heikala Art Style · MOSS v1
LoRA · New Model · Author SoyKaf · Published 11-14 01:58 · Downloads 52
Heikala Art Style LoRA: Mimics the watercolor illustration style of illustrator Heikala.
Images

Reasons for Recommendation
- New model; suitable for character illustrations with watercolor textures; the author includes a tip: write 'white pupils'
- The author provides their complete settings: MOSS SFT, strength 1.2, CFG 3, 50–70 steps, Euler with Beta57 or sgm_uniform, and adding SkipLayerGuidanceDiTSimple at layer 22 (0.0–1.0); the author claims Base / Turbo should also work
- 4 visible examples, none hidden
How to Use
- Add 'In the style of heikala.' to the prompt; strength 1.2, MOSS SFT, CFG 3, 50–70 steps, Euler + Beta57 or sgm_uniform
- Set SkipLayerGuidanceDiTSimple to single layer 22, start percent 0.0, end percent 1.0
Risks and Limitations
- Mimics the personal art style of a living illustrator; the model card only lists the artist's IG, without stating if authorization was obtained; license does not allow derivatives
- Only 4 examples; generating images with 50–70 steps is slow, and the author has not verified the results under Base / Turbo
Trigger Words: In the style of heikala.
【A】yorime (cross-eyed lora) · MOSS Image
LoRA · Ported · Author yakitoritamago · Published 11-14 02:35 · Downloads 64
MOSS version of the yorime (cross-eyed) expression LoRA: Makes characters make a cross-eyed expression, which can be combined with sticking out the tongue, surprise, focus, and other expressions.
Images

Reasons for Recommendation
- Ported: Original Krea 2 and Z-Image Base versions exist; the MOSS version was trained using natural language annotations and the Civitai trainer, with original data being photos
- Suitable for quirky expressions and realistic portraits; the author states that close-up eye shots of the upper body left/right are most stable, but farther compositions also work
- 7 visible examples (1 hidden); cumulative downloads on the page: 487
How to Use
- Add yt_yorime and crossed eyes to the prompt, then write descriptions for cross-eyed looks and other actions like mouth movements
- Strength, steps, and CFG were not provided by the author
Risks and Limitations
- The author states that closed-eye effects are difficult to control
- License does not allow derivatives; commercial use is listed as rental only; the page contains mixed content
Trigger Words: yt_yorime,, crossed eyes
【A】NEONDORÉ · MOSS Image Base
LoRA · Ported · Author Silbrio · Published 11-14 02:03 · Downloads 19
MOSS version of NEONDORÉ: Gustave Doré-style fine engraving hatching combined with restrained neon colors—hatching along contours, deep shadows, and clear highlights.
Images

Reasons for Recommendation
- Ported: Original Krea 2 version exists; the MOSS version was trained on MOSS Image Base BF16 using 60 images. A separate MOSS recipe is provided (strength 1.0, 25 steps, CFG 2.5, ER-SDE / Simple)
- Suitable for dark fantasy, sci-fi cities, monsters, and portrait illustrations; the author suggests choosing only one or two neon color pairs, such as cyan with magenta, or sulfur yellow with purple
- 9 visible examples (1 hidden)
How to Use
- Add silbrioneondore to the prompt, and after describing the composition, name one or two neon colors; start strength at 1.0, author states it can be adjusted between 0.6–1.4; 25 steps, CFG 2.5, ER-SDE / Simple
Risks and Limitations
- The trigger word written in the model card body is silneondoré (Krea 2 version), which is inconsistent with silbrioneondore registered for the MOSS version; refer to the version notes as authoritative
- The page contains mixed content
Trigger Words: silbrioneondore
【A】AG: Camio - Wild Brush Fantasy · MOSS-Image
LoRA · Ported · Author Azriaelis · Published 11-14 09:06 · Downloads 3
MOSS version of AG: Camio - Wild Brush Fantasy: A fantasy illustration style with wild brushstrokes, leaning towards gouache and thick brush textures.
Images

Reasons for Recommendation
- Ported: Original Krea 2 Turbo and Anima versions exist; the model card separately provides MOSS settings—Base with 8 step Turbo LoRA, euler / simple, strength 0.8–1.2, 8–16 steps, CFG 1–3
- Suitable for fantasy characters and scene illustrations with distinct brushstrokes; the author states that adding the trigger word makes the style significantly stronger
- 12 visible examples (8 hidden)
How to Use
- Add @C4M10V01C to the prompt; MOSS Base with 8 step Turbo LoRA, strength 0.8–1.2, 8–16 steps, CFG 1–3, euler / simple
Risks and Limitations
- The author states that training images are mixed from sources like Midjourney, NijiJourney, and Tumblr, belonging to a casual experiment series with varying LoRA quality
- License does not allow derivatives; just released with few downloads; the page contains mixed content
Trigger Words: @C4M10V01C
Worth Watching (14)
【B】V67 Fashionable, modest, well-tailored · V1 MOSS
LoRA · Ported · Author angelomaiota · Published 11-14 03:18 · Downloads 30
MOSS version of the V67 Clothing LoRA: Makes characters dress fashionably, modestly, and with well-tailored fits—long sleeves, high collars, knee-length skirts, suit sets, and other non-revealing clothing.
Images

【B】Mavis Dracula - Hotel Transylvania - MOSS LORA · v1.0
LoRA · New Model · Author Konan · Published 11-13 23:19 · Downloads 99
Character LoRA for Mavis Dracula (Dracula's daughter) from Hotel Transylvania: Black short choppy hair, pale skin, blue eyes, dark lipstick, with signature outfit prompts included.
Images

【B】Range Murata Style | MOSS Image · v1.0
LoRA · New Model · Author deitychaser · Published 11-14 01:35 · Downloads 123
Range Murata Style LoRA: Clean lines and soft painted color palettes; the author focused on tuning facial styles with Range Murata-style eye reflections.
Images

【B】Eve - Multi Outfit - Stellar Blade - MOSS LoRA · v1.0
LoRA · New Model · Author Konan · Published 11-13 23:34 · Downloads 109
Multi-outfit character LoRA for Eve from Stellar Blade: One file contains the base character and 7 outfits, with separate outfit description sentences provided for each.
Images

【B】Dizzle's Neon Pop Photography · v1
LoRA · New Model · Author Maridizzle · Published 11-13 21:41 · Downloads 56
Dizzle's Neon Pop Photography: A hyper-saturated neon fashion photography style featuring candy colors, wet glossy highlights, and surreal stage lighting.
Images

【B】rori chuushin style · v1
LoRA · New Model · Author NiSorasu · Published 11-14 01:54 · Downloads 51
rori chuushin anime style LoRA, the author's first LoRA, trained using the on-site trainer.
Images

【B】Jnkku Style [Anima/Krea2/MOSS] · MOSS v1.0
LoRA · Port · Author Imagine__That · Published 11-13 21:49 · Downloads 78
MOSS version of the Jnkku artist style LoRA: vibrant colors and delicate lines in a Korean manhwa style.
Images

【B】Witches: Yennefer cosplay | MOSS / Krea 2 /Qwen 2.1 · v2.0 MOSS
LoRA · Version Update · Author bazya022020810 · Published 11-13 23:02 · Downloads 43
v2.0 of the Yennefer (The Witcher) realistic cosplay character LoRA: fixed face, black wavy hair, purple eyes, and star necklace, with toggleable signature outfit.
Images

【B】FREE!! 免费!!MOSS Pro Grade | 专业级工作流 with Image Edit 🖼️, 🥅 Controlnet, and all the other stuff · V8.0 Relay VSR Upscale
Workflow · Version Update · Author lonecatone23 · Published 11-14 08:08 · Downloads 15
Lonecat's MOSS Pro Grade ComfyUI workflow V8.0: A large workflow integrating image editing, ControlNet, BBox, inpainting, skin details, etc., with this version swapping the upscaling module.
Images

【B】Golden Axe Style (TV Series) - [MOSS] · v1
LoRA · New Model · Author postkist · Published 11-14 01:30 · Downloads 54
Golden Axe TV anime style LoRA: Western 2D cartoon style, capable of generating 5 characters from the show.
Images

【B】Black Eyes - Moss · v1.0
LoRA · New Model · Author EvilPeterPan · Published 11-14 07:09 · Downloads 22
MOSS version of the black sclera (all-black eyes) effect LoRA, for drawing fully black eyes on horror, possessed, or otherworldly characters.
Images

【B】Deepseek & Claude 娘 · v1.0
LoRA · New Model · Author Lost_Heart · Published 11-14 00:26 · Downloads 21
LoRA for two anthropomorphized characters, DeepSeek Girl and Claude Girl, using Chinese descriptive sentences as trigger words.
Images

【B】BBox Studio · v1.2
Workflow · Version Update · Author plk · Published 11-14 02:30 · Downloads 20
BBox Studio: A standalone desktop mini-app for dragging boxes on a canvas to edit structured bounding box prompts for MOSS Image and exporting JSON.
Images

【B】Glamour Realism – Soft Natural Look · v1
LoRA · New Model · Author TP_AI_63 · Published 11-13 23:31 · Downloads 0 · Paid
Glamour Realism – Soft Natural Look: A portrait style LoRA that makes MOSS photos softer and more natural—low-saturation film colors, soft lighting, natural skin, no trigger word.
Images

Included Only (15)
- Toph Beifong - Avatar: The Last Airbender - MOSS LORA · v1.0 — LoRA: Toph Beifong character LoRA from Avatar: The Last Airbender: blind eyes, black hair bun, green headband with pom-poms, includes signature outfit prompts.

- Gwen Stacy - Spider-Verse - MOSS LORA · v1.0 — LoRA: Gwen Stacy character LoRA from Spider-Man: Across the Spider-Verse: short blonde hair with pink tips, undercut, eyebrow piercing, includes prompts for both spider suit and casual wear.

- Annie Holloway - Natural Character LoRA · MOSS v1.0 — LoRA: Annie Holloway: A fictional realistic red-haired female character LoRA, focusing on a natural bare-face look and a distinctive nose shape.

- Chel - The Road to El Dorado - MOSS LORA · v1.0 — LoRA: Chel character LoRA from The Road to El Dorado (Konan version): long straight black hair, bangs, tan skin, includes signature outfit prompts.

- Lucy - Lucyna Kushinada - Cyberpunk Edgerunner - MOSS LORA · v1.0 — LoRA: Lucy (Lucyna Kushinada) character LoRA from Cyberpunk: Edgerunners: rainbow-colored short bob, red eye makeup, includes prompts for the full outfit and back cyberware.

- Frederica Baumann - Re:Zero | MOSS-Image Lora · v1 — LoRA: Frederica Baumann character LoRA from Re:Zero - Starting Life in Another World, adding this character which MOSS does not recognize.

- MOSS Image Turbo 4 Step · SFT INT8 — Base Model / Checkpoint: SFT INT8 version of MOSS Image Turbo 4 Step: Merges the 4-step Turbo LoRA into MOSS Image SFT and then converts to INT8, single file approx. 5 GB.

- plow_witch_parfait | Style · MOSS(test) — LoRA: MOSS test version of the plow_witch_parfait artist style LoRA, cute anime girl style.

- Thirty 8ght Style (MOSS-Image) · v1.0 — LoRA: MOSS version of the Thirty_8ght artist style LoRA, anime illustration style.

- MOSS-yul刘如烟OC001。 · MOSS-yul刘如烟OC001。 — LoRA: LoRA for the original ancient-style female character 'Liu Ruyan' (first in author yul's OC series): black off-shoulder long dress, jet-black long hair, thin metal chain with green gem accents at the waist.

- Arianna Costa · MossImage — LoRA: MOSS version of the Arianna Costa character LoRA: a fictional female character who is a receptionist in Madrid.

- Esther NordhavenLove · MossImage — LoRA: MOSS version of the Esther NordhavenLove character LoRA: a fictional female character who is a florist on the Swedish west coast.

- Kigurumi Onesie: Lion [Krea2/MOSS] · MOSS_SFT_v1.0 — LoRA: MOSS SFT version of the Lion Kigurumi Onesie clothing LoRA, dressing characters in a lion-hooded onesie.

- MOSS image Workflow for ComfyUI with Multy_propmt and LoRA stack loader · v1.1 — Workflow: ComfyUI text-to-image workflow for MOSS Image: multi-prompt input plus LoRA stack loader.

- Andrea (AI Model) [MOSS / Krea2 / SD] · MOSS v1.0 — LoRA: MOSS version of the Andrea character LoRA: a fictional female character with a red asymmetric bob, freckles, and hazel-green eyes.

Data: Civitai public API (base model MOSS Image), inclusion window 11-13 20:31 → 11-14 09:18 (Beijing Time). Recommendations reviewed by editors; download counts only reflect attention.