Zhenzhen's AI Radar · MOSS Image · 2026-11-09 Morning Report
This issue features 40 new versions: Strongly Recommended 1 · Recommended 10 · Worth Watching 10 · Included Only 19
Editor's Note This issue reviews 46 new MOSS versions: 2 S, 10 A, 11 B, and 23 C. Top pick: Juggernaut Moss, a full fine-tuned base model for MOSS-Image Base, focusing on realistic photography and cinematic feel, with Euler 30 steps, CFG 3, Shift 5, and an int8 version available. Another S-tier pick is the backlit, translucent xiaoyue anime style LoRA; A-tier includes the 'Kill the Flat Look' workflow and accompanying ray anime style, along with two lens effect LoRAs for rainbow prism bokeh and swirly bokeh, plus several artist/retro animation styles. The general quality prompt pack V67 was downgraded from S to B, indicating that character and prop LoRAs with vague descriptions or fully hidden examples are classified as C.
Strongly Recommended (1)
[S] Art style: xiaoyue style (Personal Style) · v1.0
LoRA · New Model · Author Suzuka_naoko · Published 11-08 21:22 · Downloads 49
A backlit, translucent anime style LoRA: fine lines, clean skin coloring, subtle highlights in the hair, and small patches of light on the face.
Images
Reasons for Recommendation
- New model; serves anime illustrations featuring backlighting, shimmering sea surfaces, elaborate costumes, and close-up two-person compositions
- Complete parameters: Trigger word 'xiaoyue' (write 'xiaoyue style' in Chinese prompts) placed at the beginning, strength 1.0, euler / simple, 50 steps, CFG 3, shift 5, recommended to add SkipLayerGuidanceDiTSimple (single 22, 0–50%), test mostly on landscape images 1536×1024, negative prompts and example prompts are provided
- Trained on four writing styles: English tags, English sentences, Chinese tags, and Chinese sentences; the author detailed training specifics (58 images, rank 32 / alpha 128) and known limitations; 7 visible examples; permissive license
How to Use
- Start prompts with 'xiaoyue' (write 'xiaoyue style' in Chinese), strength 1.0; euler / simple, 50 steps, CFG 3, shift 5; add SkipLayerGuidanceDiTSimple single 22, 0–50%; negative prompt: worst quality, bad anatomy, watermark, jpeg artifact, blurry
Risks and Limitations
- Contradictory base model recommendations: Recommended settings list MOSS-Image-heal, but known limitations state this tier appears gray and washed out on heal, suggesting SFT instead; both need to be tested
- Limitations listed by the author: Dark scenes still carry the warm cinematic lighting of SFT, large bright areas tend to blow out, irises are simpler than the original art, and faces tend to be rounder
- The trigger word is not registered in the page's training words; rely on the main text; version is v1.0 but the filename is xiaoyue_v2_heal_s3400; 21 images in the training set are supplementary images selected after model self-generation, and the source of 37 original images is unexplained
- 1 examples are hidden; this entry has not been run
Recommended (10)
[A] Wada Arco - Artist Style · MOSS-Image
LoRA · Port · Author Azriaelis · Published 11-08 22:31 · Downloads 86
A MOSS version of the Wada Arco (ワダアルコ) style LoRA, by the illustrator of the Fate/EXTRA series, featuring an anime digital illustration style.
Downgraded from automatic S to A: MOSS parameters are complete, but it is merely a single artist style imitation and relies on a separately sourced acceleration LoRA.
Images
Reasons for Recommendation
- Port: The same page already has Krea 2 and Anima versions, with a cumulative 789 downloads on the model page
- The author provided specific MOSS parameters: Base with 8 step turbo LoRA, euler / simple, strength 0.8–1.2, 8–16 steps, CFG 1–3; the author claims the style is significantly stronger when the trigger word is used
- 10 visible examples
How to Use
- Write '@wada_arco' in the prompt; MOSS-Image Base + 8 step turbo LoRA, euler / simple, 8–16 steps, CFG 1–3, LoRA strength 0.8–1.2
Risks and Limitations
- Imitates the style of a living artist; the model card does not describe style characteristics, only parameters
- Requires a separate 8 step turbo LoRA, which the author does not specify
- 10 examples are hidden due to rating; the license does not allow publishing derivative models
- This entry has not been run
Trigger Words: @wada_arco
[A] Neo-Fantasy · MOSS
LoRA · Port · Author NeuralVandal · Published 11-09 08:01 · Downloads 14
A MOSS version of the Neo-Fantasy fantasy illustration style LoRA: subjects have distinct outlines, neon-like rim lighting and volumetric light, detailed backgrounds, and can handle themes from medieval fantasy to cyberpunk.
Images
Reasons for Recommendation
- Port: The same page already has SDXL, Krea 2, and Anima versions, with a cumulative 1999 downloads on the model page
- Clear trigger words; 11 visible examples
How to Use
- Write 'Illustration artwork' and '@NeoFantasy' in the prompt; the author did not provide strength or sampling parameters
Risks and Limitations
- The model card consists entirely of descriptions for the SDXL and Krea2 V2 versions (the author states the SDXL example images used the SDXL base model), with no specific instructions for the MOSS version
- The original style was trained on generated images from multiple Flux fantasy style models
- 9 examples are hidden due to rating; the license does not allow commercial use of generated images
- This entry has not been run
Trigger Words: Illustration artwork, @NeoFantasy
[A] 'SAFARI EYES' · MOSS
LoRA · Port · Author John_KSampler · Published 11-08 21:32 · Downloads 63
A MOSS version of the 80s animation style LoRA recreating the opening sequence of the 1986 film 'Dirty Pair: Project Eden' (Safari Eyes by Koji Morimoto): bold solid colors, flat shading, smoke/fluid effects, and silhouettes.
Images
Reasons for Recommendation
- Port: The same page already has ten versions including MiniMax H3, Krea 2, Anima, Illustrious, and Flux, with a cumulative 3601 downloads on the model page
- Training images are 30 screenshots of each major shot from the opening sequence; the author clearly states strengths (smoke/clouds/fluids, silhouettes, bold color blocks, patterns, close-ups) and known issues
- 8 visible examples
How to Use
- Write 'safarieyes' in the prompt, then add at least one of '1980s (style)' or 'flat color' (little change if only the trigger word is used); most noticeable when the image is abstract; the author did not provide strength
Risks and Limitations
- Tends to add black borders (the author attributes this to dataset flaws), which need to be cropped
- Training material consists of screenshots from commercial animation; the MOSS version lacks parameter instructions
- 2 examples are hidden; this entry has not been run
Trigger Words: safarieyes
[A] AG: Marbas - Soft Crayon Style · MOSS-Image
LoRA · Port · Author Azriaelis · Published 11-09 09:10 · Downloads 5
A MOSS version of Marbas from the Ars Goetia series: a soft crayon texture, grainy hand-drawn cute style LoRA.
Downgraded from automatic S to A: Parameters are complete and there are many examples, but the author admits quality is unstable and training image sources are mixed.
Images
Reasons for Recommendation
- Port: The same page already has Krea 2 and Anima versions, with a cumulative 658 downloads on the model page
- The author provided MOSS parameters: Base with 8 step turbo LoRA, euler / simple, strength 0.8–1.2, 8–16 steps, CFG 1–3; the author claims the trigger word makes the effect significantly stronger and it is highly prompt-dependent
- 12 visible examples
How to Use
- Main trigger word M4RB4SFX, then add 'hand-drawn', 'soft', 'grainy' as needed (the author suggests trying different combinations); Base + 8 step turbo LoRA, euler / simple, 8–16 steps, CFG 1–3, strength 0.8–1.2
Risks and Limitations
- The author describes it as a casual small experiment with inconsistent quality within the series; training images mix sources like Midjourney, Tumblr, and NijiJourney
- Requires a separate 8 step turbo LoRA; 5 examples are hidden; published about 1 hours ago with very little feedback
- The license does not allow publishing derivative models
- This entry has not been run
Trigger Words: M4RB4SFX, hand-drawn, soft, grainy
[A] MOSS LoRA Chibi Character · v1.0
LoRA · New Model · Author ccooccoo · Published 11-09 03:01 · Downloads 49
A cocoart-style chibi LoRA: big-head full-body chibi characters or chibi animals, with simple backgrounds and thick black lines.
Images
Reasons for Recommendation
- New model; serves chibi images for avatars, stickers, and memes
- The author provided two complete trigger sentences (one for characters, one for animals); 5 visible examples
How to Use
- Write 'chibi, chibi of big head, full body with simple background and black bold line style.' in the prompt; for animals, change to 'chibi, chibi of animal with simple background and black bold line style.'; the author did not provide strength or sampling parameters
Risks and Limitations
- Description is only two lines, with no parameters
- The author states that reposting to other sites requires payment; the license does not include commercial use of generated images
- This entry has not been run
Trigger Words: chibi, chibi of big head, full body with simple background and black bold line style., chibi, chibi of animal with simple background and black bold line style.
[A] Eruu (780599) Style [MOSS] · v1.0
LoRA · New Model · Author Imagine__That · Published 11-08 23:06 · Downloads 55
A LoRA with a manhua art style featuring slightly sketchy lines, imitating the artist Eruu.
Images
Reasons to Recommend
- New model; trigger words and strength are provided
- 4 visible examples
How to Use
- Write "eruu style" in the prompt; when used alone, strength is 0.8–1.00; sampling parameters were not provided by the author
Risks and Limitations
- Description is only three lines; imitates an artist's style
- 4 examples are hidden; license does not allow commercial use of generated images
- This entry has not been tested
Trigger Words: eruu style
[A] Prism Rainbow Lens — Soft Rainbow Flare at the Edge of the Frame | MOSS · v1.0
LoRA · New Model · Author 858654128784 · Published 11-08 21:29 · Downloads 46
A LoRA for lens flare effects with a prism held in front of the lens: a transparent rainbow flare floats at the edge of the frame, containing slight displaced scene ghosts.
Images
Reasons to Recommend
- New model; serves portrait, street photography, and other photographic images by adding lens flare effects
- The author clearly states usage points: most visible under clear sunlight; this is a lens effect, so it should be placed at the edge of the frame; placing it on the subject will result in sticker-like color bars
- 4 visible examples; permissive license
How to Use
- Start the prompt with "prismrainbowlens," and specify the flare position as "at the edge of the frame"; strength and sampling parameters were not provided by the author
Risks and Limitations
- Trigger words are not registered in the page's training words; rely on the main text; description is only two sentences
- 2 examples are hidden; this entry has not been tested
[A] Swirly Bokeh Lens — Cat-Eye Highlights Circling the Subject | MOSS · v1.0
LoRA · New Model · Author 858654128784 · Published 11-08 21:25 · Downloads 62
A swirly bokeh LoRA for vintage portrait lenses: background highlights become cat-eye shaped flares swirling around a sharp subject, with darkened corners.
Images
Reasons to Recommend
- New model; serves portrait photography images requiring a vintage lens texture
- The author clearly states usage points: the effect is subtle; the background needs dense, translucent foliage; do not write "whirlpool," or it will actually draw a vortex
- 4 visible examples; permissive license
How to Use
- Start the prompt with "swirlybokehlens," and specify the background as dense, translucent foliage; strength and sampling parameters were not provided by the author
Risks and Limitations
- The author states the effect is weak and scene-dependent
- Trigger words are not registered in the page's training words; 3 examples are hidden; this entry has not been tested
[A] MOSS-Image: Kill the Flat Look-More features, solving flat coloring v1 · v1.0
Workflow · New Model · Author NekoArc17 · Published 11-08 21:24 · Downloads 19
A ComfyUI workflow targeting MOSS's tendency to "revert to flat coloring": increases latent space contrast before VAE decoding, or bypasses the FLUX.2 VAE and uses NVIDIA PiD for direct 4× decoding.
Upgraded from automatic B to A: The problem, two solutions, and parameters are clearly written and actionable; the barrier is the number of dependency nodes.
Images
Reasons to Recommend
- New workflow (JSON, not new weights); the author claims that increasing steps, adjusting shift, and adjusting rank during style LoRA training did not eliminate the flat coloring regression, finally pinpointing the FLUX.2 VAE decoding stage
- Two decoding paths are switched via the master control panel: LatentMultiply coefficient 1.03–1.08 (default 1.05) plus 1.1× latent space pre-upscaling; or use PiD, with the distilled version taking 4 steps and the original taking 12–15 steps
- Includes seed locking, 1× / 4× upscaling model selection, optional SeedVR2 super-resolution, and grouped toggle panels
- Follows the same approach as the same author's current "ray anime v1.0", with consistent parameter recommendations
How to Use
- Import "Mossimage-More features.json", select "Latent Enhancement" or the PiD path in the master control panel; style LoRA strength 1.1–1.2
- Write "flat color, 2d, cel shading, bad lighting, dull skin, matte, muted colors, low contrast, simple shading" in the negative prompt; do not stack Turbo acceleration LoRAs; if it still looks flat, first adjust the shift in the sampling group
Risks and Limitations
- Depends on custom nodes and models like NVIDIA PiD, SeedVR2, and rgthree, which must be installed separately
- Only 1 visible examples, no before-and-after comparison images; "the problem lies in the VAE" is the author's judgment
- This entry has not been tested
[A] ray anime-Enhanced ray tracing and lines, removing flat look v0.1 · v1.0
LoRA · Version Update · Author NekoArc17 · Published 11-08 21:10 · Downloads 40
An anime style LoRA aiming to enhance lighting (ray-tracing-like illumination) and lines, and remove the flat look from MOSS outputs; this release is v1.0.
Upgraded from automatic B to A: Parameters and pitfalls are clearly written; recommended to view alongside the same author's "Kill the Flat Look" workflow.
Images
Reasons to Recommend
- Version update: The previous version on the same page was v0.1; this version includes two checkpoints, ray_anime2000steps3 and ray_anime1750steps3, for comparison
- Usage is fully documented: trigger word "ray_anime", strength 1.1–1.2 (author says 1.0 is insufficient), accompanying negative prompt, and a reminder not to stack Turbo LoRAs
- The author explains that the effect is also influenced by shift and the skip ratio of the 22 layer; the same author released the "Kill the Flat Look" workflow this period to solve flat coloring, which can be used in combination
- Released about 12 hours ago with 40 downloads (only indicates attention level)
How to Use
- Write "ray_anime" in the prompt, strength 1.1–1.2; write "flat color, 2d, cel shading, bad lighting, dull skin, matte, muted colors, low contrast, simple shading" in the negative prompt
- Do not stack Turbo acceleration LoRAs (the author claims it will clearly revert to flat coloring); steps and CFG were not provided by the author
Risks and Limitations
- The author did not write update notes; from API filenames, v0.1 appears to be checkpoints for 1250 / 1500 steps, while this version is for 1750 / 2000 steps, so the difference may mainly be longer training
- Trigger words are only written in the main text, not registered on the page
- 1 visible examples, another 2 hidden by rating; the author self-assesses "not yet perfect"
- This entry has not been tested
Worth Watching (10)
[B] V67 Quality Boost · V1 MOSS
LoRA · Port · Author angelomaiota · Published 11-09 00:31 · Downloads 61
The MOSS version of a general quality boost LoRA: uses a long string of quality keywords to push the image towards sharper, more detailed, realistic results resembling commercial retouching.
Images
[B] Kiko MOSS-Image · v6
Workflow · New Model · Author kiko9 · Published 11-09 07:43 · Downloads 9
Kiko's MOSS-Image text-to-image ComfyUI workflow: configured according to official templates (er_sde / simple, 25 steps, CFG 2.5, layer skip guidance for the first half at layer 22), works with both SFT and Base, with optional prompt expansion.
Images
[B] Gladas' MOSS Image Workflow · v2
Workflow · Version Update · Author Gladas · Published 11-09 07:13 · Downloads 19
Gladas' MOSS text-to-image workflow v2: multi-LoRA loading, wildcards, prompt rewriting, optional upscaling, and saving images with Civitai-recognizable metadata.
Images
[B] NIJI Phi - MOSS-Image · v1
LoRA · New Model · Author JamesKim97 · Published 11-08 22:10 · Downloads 41
A style LoRA trained on images generated by Niji Journey; the author does not describe the specific style.
Images
[B] Gladas' MOSS Image Workflow for Nodes 2.0 (no custom nodes) · v1
Workflow · New Model · Author Gladas · Published 11-08 21:31 · Downloads 32
A MOSS text-to-image workflow that does not rely on any custom nodes, designed specifically for the ComfyUI Nodes 2.0 interface: text-to-image, multi-LoRA loading, and SeedVR2 cleanup or upscaling.
Images
[B] Kiko MOSS-Image LoRA checkpoint compare + head-to-head · v7
Workflow · New Model · Author kiko9 · Published 11-09 07:43 · Downloads 12
Two ComfyUI workflows for LoRA trainers to select the best checkpoint: under the same seed, each checkpoint is combined with the base model into a comparison image, with the 0th cell containing the actual training image as a reference; then finalists are re-evaluated with multiple seeds.
Images
[B] soft cel shading / pastel color style for Anima · moss v1.0
LoRA · Port · Author azukiuy · Published 11-08 22:24 · Downloads 52
The MOSS version of an anime style LoRA featuring soft cel shading and pastel colors (originally made for Anima).
Images
[B] StainedGlass [MOSS / Krea2/ SD] · MOSS v1.0
LoRA · Port · Author TeraWatt · Published 11-09 01:46 · Downloads 0 · Paid
Stained glass (rose window) style LoRA for MOSS: Renders any subject—portraits, landscapes, pets, etc.—as a stained glass craft effect with metal framing.
Image
【B】ELEVATE BY STX · V1
LoRA · New Model · Author STXai · Published 11-09 04:27 · Downloads 0 · Paid
LoRA for scenes of people floating in water: Renders bubbles and light filtering through the water surface along with the subject.
Image
【B】BRAND BY STX · v1
LoRA · New Model · Author STXai · Published 11-09 02:21 · Downloads 2 · Paid
Fire-themed LoRA: Renders people or objects on fire; the author later discovered that it adds grit and detail to images even when fire is not specified.
Image
Archive Only (19)
- Kira — Photorealistic LoRA [MOSS / Krea,Qwen 2.1] · v2.0 MOSS — LoRA: v2 of Kira, a photorealistic Slavic female character LoRA. The author claims it was an accidental result of a failed Cosplay training run; the face remains unchanged, but the styling is softer, warmer, and more casual.
- Witcheress: Kira / Yennefer · v1.0 MOSS — LoRA: A character LoRA transforming the author's realistic character Kira into a Yennefer Cosplay from The Witcher: purple eyes, dark curly hair, black velvet clothing, star-shaped necklace, and spell-casting poses.
- [BSS] - Styles for MOSS Image · Amber — LoRA: The Amber style LoRA from the [BSS] style collection; the author did not specify what the Amber style looks like.
- Homer Simpson · Homer Simpson — LoRA: Character LoRA for Homer Simpson from The Simpsons.
- Makaria Silverstova (This person does not exist) · MOSS — LoRA: MOSS version of the character LoRA for the fictional female character Makaria Silverstova (the author states the character was generated by GPT and does not exist in reality).
- Tarutoru | Style · MOSS(test) — LoRA: MOSS test version of the male-oriented (bara) anime art style LoRA by artist Tarutoru.
- moss base highwaist-boyshorts · v1 — LoRA: High-waist boyshorts clothing LoRA, one of a set of four by the same author (micro / midi / maxi / highwaist).
- moss base midi-boyshorts · v1 — LoRA: Mid-length (midi) boyshorts clothing LoRA, one of a set of four by the same author (micro / midi / maxi / highwaist).
- Riding_Crop · v1 — LoRA: Riding crop prop LoRA; the author claims MOSS originally struggled to render this prop correctly.
- Lucie Belwyn · MossImage — LoRA: MOSS version of the character LoRA for the fictional character Lucie Belwyn (owner of an old bookstore in Bruges), part of MisterMR's Belwyn family series.
- Kaia · v1.0 — LoRA: MOSS character LoRA for the fictional female character Kaia, the author's first MOSS character LoRA.
- Alsares style · v1.2 MOSS-image — LoRA: Furry art style LoRA imitating Instagram artist alsares_art.
- Liä Kalimullina (This person does not exist) · MOSS — LoRA: MOSS version of the character LoRA for the fictional female Liä Kalimullina; the author states the image was generated using GPT Image and does not exist in reality.
- Thessaly · v1 — LoRA: MOSS version of the character LoRA for Thessaly, created by the author using screenshots of a virtual avatar from the Replika app.
- MOSS cocoball style · v1.0 — LoRA: "cocoball" anime style LoRA; the author provided no description.
- MOSS image Workflow for ComfyUI with Multy_propmt and LoRA stack loader · v1.0 — Workflow: A MOSS text-to-image ComfyUI workflow featuring multi-prompt input and LoRA stack loading.
- ALTERNATE MOSS BY STX · V1 — LoRA: MOSS version of STX's general "artistic + detail" LoRA; the author did not specify what changes were made.
- DARE BY STX · V1 — LoRA: A LoRA that helps render scenes of "women holding guns"; the author claims they consistently struggled to generate such images.
- COST BY STX · V1 — LoRA: A LoRA for themes featuring female warriors wielding swords and blades (samurai, ninjas).
Data: Civitai Public API (Base Model MOSS Image), inclusion window 11-08 20:31 → 11-09 09:18 (Beijing Time). Recommendations have been reviewed by editors; download counts reflect popularity only.