Zhenzhen's AI Radar · MOSS Image · 2026-10-30 Evening Report
This issue features 37 new versions: Strongly Recommended 0 · Recommended 2 · Worth Watching 16 · Listed Only 19
Editor's Note This issue reviewed 37 new MOSS versions, with no S-tier entries: the 7 items automatically rated S were downgraded due to incomplete descriptions, mere quantization format changes, or standard character models. Top pick: MOSS_LUC V2, an anime-illustration-oriented merged base model. The author provided full details on the sampler, steps, CFG, Shift scheduler, and DMD acceleration LoRA strength, but the merge sources are unclear and only INT8 files are available, resulting in a downgrade from S to A. Another A-tier entry is a Honor of Kings official CG-style 3D character rendering style LoRA, which includes trigger words and accompanying descriptions but lacks strength parameters, and material copyright requires attention. B-tier entries include workflows for running 2.5D anime base models on 8GB VRAM, local prompt expansion nodes, car interior surveillance and interrogation room scene concepts, bald head correction and lighting LoRAs; numerous standard character LoRAs, concept LoRAs with only one-line descriptions, as well as voyeuristic perspective and realistic character LoRAs are rated C.
Recommended (2)
[A] MOSS_LUC [ moss-image-checkpoint ] · sft-int8-v2
Base Model / Checkpoint · New Model · Author ukr8b3g201 · Published 10-30 11:57 · Downloads 101
An anime-illustration-oriented MOSS merged base model (the author's Lucy series merges quantized to INT8 via convrot): The author claims V2 is more stable, has more detail, and follows prompts better than V1, at the cost of a slightly less distinct art style compared to V1, in exchange for broader applicability.
Downgraded from S to A: Parameters, accompanying LoRAs, schedulers, and prompts are fully documented, and the license is permissive; however, the merge sources are unclear, only INT8 files are available, and it is a standard merged base model rather than a new capability.
Images
Reasons for Recommendation
- New model: An anime-oriented checkpoint that directly replaces the MOSS base model, serving text-to-image users who want anime and illustration styles without stacking style LoRAs; the author explicitly states that realism is not its strength.
- The author provided full parameters: DPM++ SDE or DPM++ 2M SDE, 8–10 steps (starting from 8), CFG 2–5 (starting from 3), MOSS scheduler set to Shift with value 5, and DMD acceleration LoRA strength 0.7–0.8; recommended positive and negative prompts are also provided.
- Includes 9 visible examples and a workflow JSON; free, with a permissive license: allows image generation, renting, selling, and selling merges, with no attribution required.
- Approximately 9 hours after release, it has 101 downloads and 5 likes (indicating attention only).
How to Use
- Download mossLUCMossImage_sftInt8V2.safetensors (INT8 full weights) to replace the MOSS base model; the accompanying workflow JSON can be dragged directly into ComfyUI.
- Sampler: DPM++ SDE / DPM++ 2M SDE, starting from 8 steps, CFG starting from 3; the author strongly recommends installing their MOSS Shift scheduler extension (available for both ComfyUI and Forge Neo), with Shift value 5.
- Additionally load lodestones' DMD acceleration LoRA (lora_latest.safetensors) with strength 0.7–0.8; use 'masterpiece, best quality, amazing quality' for positive prompts and copy the author's list for negative prompts.
Risks and Limitations
- The author does not specify which models were merged or their ratios, only stating they come from their own Lucy series; V1 is currently not displayed, so comparison is impossible.
- Only the INT8 convrot quantized version was released, with no bf16 / fp8 files; recommended parameters depend on the external DMD LoRA and the author's custom scheduler extension.
- The author did not provide version update notes (changes from V1 to V2 are only mentioned in one sentence in the model card).
- 11 sensitive-level examples are hidden.
[A] (MOSS image) Honor of Kings 3D CG Style · v1.0
LoRA · New Model · Author mikezhang258 · Published 10-30 12:31 · Downloads 51
A 3D rendering LoRA in the official CG style of Honor of Kings: focusing on the 3D texture of skin and faces (smooth volumetric lighting, glossy skin with subsurface scattering), the author claims it works better with SFT models.
Maintained at A: The art style direction is clear, trigger words and accompanying descriptions are complete, with 10 examples; what is missing are strength parameters, and material copyright requires attention.
Images
Reasons for Recommendation
- New model: A semi-realistic 3D CG character rendering style LoRA for MOSS, serving text-to-image users who want game promotional CG-style character portraits and refined 3D portraits.
- The author provided the trigger word wz-cg-style and a suggested style description to be included simultaneously (e.g., Ultra-polished semi-realistic 3D CG character render).
- 10 visible examples; free, with a license allowing image generation, renting, in-site renting, and merging, with no attribution required.
- Approximately 9 hours after this version's release, it has 51 downloads and 12 likes (indicating attention only).
How to Use
- Load wz-cg-style.safetensors in a MOSS SFT workflow (the author has not tested it with the Base model).
- Include 'wz-cg-style' in the prompt, followed by 'Ultra-polished semi-realistic 3D CG character render, smooth volumetric shading, glossy subsurface-scattering skin, soft ambient occlusion'.
- The author did not provide strength, steps, or CFG parameters.
Risks and Limitations
- Training materials are official Honor of Kings CGs, with copyright belonging to the game company; the author did not specify authorization.
- No strength or sampling parameters provided; the author has not tested whether it works with the Base model.
- The trigger word is not registered in the site's trigger word section, only mentioned in the description.
- 10 sensitive-level examples are hidden.
Worth Watching (16)
[B] V67 Pantone - Colour of the Year · V1 MOSS
LoRA · Ported · Author angelomaiota · Published 10-30 19:26 · Downloads 16
The MOSS version of the 'Pantone Color of the Year' series LoRAs: The author created a color theme set from 27 Pantone Colors of the Year spanning from 2000 to 2026, used to generate minimalist color imagery featuring a single annual color as the protagonist.
Images
[B] AG: Asmoday - Soft Emo Niji Boys · MOSS-Image
LoRA · Ported · Author Azriaelis · Published 10-30 11:00 · Downloads 27
The MOSS version of the 'Asmoday' style LoRA from the Ars Goetia series: A soft, emo-tinged Niji-style semi-realistic male character art style; the author claims the style becomes significantly stronger when trigger words are added.
Images
[B] Realism By Stable Yogi Moss · v1.5 fp8
Base Model / Checkpoint · Version Update · Author Stable_Yogi · Published 10-30 20:30 · Downloads 43
The V1.5 fp8-scaled file of the photorealistic MOSS base model 'Realism by Stable Yogi': The author claims V1.5 offers more stable human anatomy in crowded scenes and complex poses, cooler and more neutral tones, and smoother skin; this version is merely another quantization format of the same V1.5.
Images
[B] MOSS 2.5D Anime on 8GB VRAM — 1024x1536 in 62s (Apache 2.0) · v1.0
Workflow · New Model · Author tinytensor · Published 10-30 19:26 · Downloads 9
A ComfyUI text-to-image workflow for running MOSS 2.5D anime base models on laptops with 8GB VRAM: The author provides the hardware environment recorded when generating 1024x1536 images in approximately 62 seconds, along with a five-layer prompt structure and pitfall avoidance guide; no new weights are included.
Images
[B] Urushihara Satoshi, 'Langrisser', 'Front Innocent' - Artist Style · MOSS_v1.0
LoRA · Ported · Author flyx3 · Published 10-30 18:10 · Downloads 68
The MOSS version of the art style LoRA for illustrator Urushihara Satoshi ('Langrisser', 'Front Innocent'): Uses two trigger words to switch between later and earlier art styles; the author claims training images exceed 1000, all taken from the artist's original art books.
Images
[B] Moss Image Text Enhancement & Image to Text Node (Comfy Local) · v2.0
Other · Version Update · Author Felldude · Published 10-30 18:26 · Downloads 35
A locally running ComfyUI node: Uses MOSS's text encoder for prompt expansion, and can reverse-engineer descriptions or describe images in a specific style when connected to an image; v2.0 adds three settings: temperature, Top P, and repetition penalty.
Images
[B] Car Cabin CCTV — Everything the Backseat Camera Saw | MOSS · v1.0
LoRA · New Model · Author 858654128784 · Published 10-30 11:16 · Downloads 130
A realistic concept LoRA for car interior surveillance perspective: Simulates a wide-angle camera mounted at the top center of the windshield facing inward, with the driver prominently in the right foreground, the passenger seat empty, rear passengers seated on the left, the entire rear row visible, streets, parking lots, or highways outside the windows, slight wide-angle distortion at the edges, and horizontal composition.
Images
【B】Genshin mitya | Genshin Impact Mitya · v1.0
LoRA · New Model · Author Suzuka_naoko · Published 10-30 10:51 · Downloads 18
A MOSS character LoRA experimental version for the Genshin Impact character Mitya (a boy with fluffy silver-white short hair and large purple eyes): clothing, brooch, short boots, and cyan elements are separated into individual tags, allowing for outfit changes and realistic cosplay photos. The author provides two sets of parameters: one for DMD acceleration and one for standard image generation.
Images
【B】Interrogation Room — One Night Behind the One-Way Glass | MOSS · v1.0
LoRA · New Model · Author 858654128784 · Published 10-30 18:30 · Downloads 67
A realistic police interrogation room scene concept LoRA: features a rectangular table, black mesh chairs, a dark one-way observation window covering an entire wall, a high small window, a corner surveillance camera, a gray door, and carpet/concrete flooring. The author states it can produce a clean interrogation room with even lighting, or create an oppressive interrogation atmosphere by using only a desk lamp or a single pendant light. Landscape composition.
Images
【B】Hanako Arasaka Cyberpunk2077 - Moss · v1.0
LoRA · New Model · Author gkhan_ · Published 10-30 19:07 · Downloads 14
A MOSS character LoRA for Hanako Arasaka from Cyberpunk 2077: the simplest trigger syntax locks in the black bun hairstyle and iconic white ring eyes. Lipstick, golden fingers, forearm golden spiral patterns, and three outfits are separated into optional tags.
Images
【B】Realistic Background Enchanted | Realistic Background Enchanted | Moss image · v2.0
LoRA · Version Update · Author Akai_azusa · Published 10-30 18:08 · Downloads 41
Version 2 of a style LoRA that naturally integrates anime characters into realistic scenes (environments, lighting, and atmosphere with realistic textures), with no trigger word. This version enhances the effect of "flat characters standing in the real world" and fine-tunes the lighting, but the author personally prefers v1 overall.
Images
【B】Abandoned America-MOSS-Image · v1
LoRA · New Model · Author illuminatipyramid · Published 10-30 11:07 · Downloads 56
A concept LoRA for generating abandoned landscapes, with the name pointing to American-style desolate scenes. The model card contains only one sentence: "Use this LoRA to create your own abandoned landscapes."
Images
【B】SteveChopz style | [MOSS | Illustrious] · moss_1.0
LoRA · Port · Author NoiseWriggler · Published 10-30 15:29 · Downloads 40
The MOSS version of a LoRA imitating the art style of illustrator SteveChopz: the original was designed for Illustrious, and the author states it is suitable for character and furry illustrations. The MOSS version is trained on Base, with examples generated on SFT.
Images
【B】Radiance Chrome Voluptuous · MOSS-Image-base v1.0
LoRA · Port · Author AIImageStudio · Published 10-30 19:42 · Downloads 1 · Paid
The MOSS version of a slide film (positive/slide film) texture style LoRA: features strong contrast and saturated colors. The author claims the lighting, especially sunlight representation, is stronger. The MOSS version is trained on Base and can be paired with a 8 step DMD accelerated version.
Images
【B】Baldness helper · baldness_helper_MOSS
LoRA · New Model · Author R_6283 · Published 10-30 20:06 · Downloads 15
A baldness correction concept LoRA: the author states that base models often leave a layer of stubble fuzz when drawing bald characters. This LoRA makes the scalp clean and smooth.
Images
【B】luminous - Trained on images selected for tension and excellent lighting · v1.0 for sft
LoRA · Version Update · Author NekoArc17 · Published 10-30 09:31 · Downloads 27
The SFT version of a lighting style LoRA: the author trained on images selected for their tension and excellent lighting, triggered using MOSS's <|visual|> tag syntax. This version is trained on MOSS-SFT.
Images
Included Only (19)
- Si | Arknights: Endfield | Moss-Image · v1.0 — LoRA: A MOSS character LoRA for the character "Si" from Arknights: Endfield (mint green twin low buns, dragon horns, scaled tail); the author states this is an experimental version used to test MOSS's detail reproduction, hence the inclusion of a very long full-body appearance prompt.
- Coria Style Anima&Moss · Moss — LoRA: The MOSS version of the "Coria" anime girl style LoRA (previously available as an Anima version TLora); the model card does not specify what this style looks like or its origin.
- Miko Character LoRA for MOSS Image and Krea 2 Turbo · 1.0 (MOSS) — LoRA: A character LoRA for the original realistic adult female character Miko (East Asian features, dark brown high messy bun), trained for both Krea 2 Turbo and MOSS Image SFT, with the trigger word mikoas.
- ✦ Z-Cast: Bastu R. [OC] · Moss — LoRA: A MOSS version character LoRA for the author's original male character Bastu (B. A. Reynald, bald, bearded, muscular Black male), with the trigger word Bastu.
- YUNSANG Style [MOSS // Anima] · moss v1.0 — LoRA: The MOSS version of the YUNSANG style LoRA, with the trigger word @yunsangstyle; the author provided MOSS-specific parameters, stating that the Base model reproduces the style better than SFT, but admits this version is still underfitted.
- AuthentiQ - HannaRS22 [Krea2/MOSS] · MOSS_Image_SFT_v1.0 — LoRA: The MOSS SFT version of the realistic blonde female character LoRA "HannaRS22", with the trigger word HannaRS22; the author states this is a test of Civitai's on-site training service, which has very limited configurable options.
- Secret Peek — Screen & inside Curtains | MOSS · v1.0 — LoRA: A realistic concept LoRA for the "peeking at someone's screen" perspective: the camera captures a man from behind the shoulder, through a door crack, the gap in bunk bed curtains, or from above/behind, looking at a phone/computer/tablet screen without realizing he is being watched; the author states the screen is the brightest part of the image and the chat text on it is readable.
- ✦ Z-Cast: Evelyn T. [OC] · Moss — LoRA: A MOSS version character LoRA for the original character Evelyn T., with the trigger word Evelyn; in the author's story, she is the alias used by Sylvia H. after the apocalypse, and the Krea 2 version includes both mechanical arm and severed arm forms.
- Pandora - MOSS · Pandora - MOSS — LoRA: A MOSS character LoRA for Pandora from Re:Zero - Starting Life in Another World: it provides two trigger phrases for the base appearance (gray super-long hair, blue headband, blue eyes) and a white off-shoulder dress, along with animation screenshot style tags.
- Natural Pattern · v1 — LoRA: A concept LoRA named Natural Pattern, with the trigger word pattern; the model card only contains the title sentence, and while the name suggests a relation to natural patterns/designs, the author does not specify its exact function.
- Meili - MOSS · Meili — LoRA: A MOSS character LoRA for Meili (Meili Portroute) from Re:Zero - Starting Life in Another World, featuring one set of casual attire and one set of cloaked attire.
- Liliana Masquerade - MOSS · Liliana Masquerade — LoRA: A MOSS character LoRA for Liliana (Liliana Masquerade) from Re:Zero - Starting Life in Another World, with the author providing complete appearance and clothing tags.
- MOSS Character Ajitani Hifumi · v1.0 — LoRA: A MOSS character LoRA for Ajitani Hifumi from Blue Archive, including three outfits: default uniform, racing suit, and swimsuit.
- Antonia NordhavenLove · MossImage — LoRA: A MOSS character LoRA for the fictional female character Antonia NordhavenLove; the author declares she is not a real person, and the model card contains a backstory describing her as a model from the west coast of Sweden.
- Tsaritsa - Genshin Impact (MOSS, ANIMA) · MOSS v1.0 — LoRA: A MOSS character LoRA for the Tsaritsa (Ice Queen) from Genshin Impact, with the author stating this is their first attempt at creating a character LoRA.
- Checkpik - Style - MOSS Image / Illustrious XL / Anima · MOSS Image — LoRA: The MOSS version of a 3D CG style LoRA imitating Checkpik's style (tagged with 3d, cgi, blender); the author states the MOSS version uses an updated dataset and requires no trigger word.
- kurumi fukuga,FX Warrior Kurumi · v1.0 — LoRA: A MOSS character LoRA for the character Kurumi Fukuga (kurumi fukuga, Chinese name listed as "FX Warrior Kurumi"), with the description consisting of only a single line of character tags.
- A MIXED STYLE · v1.0 — LoRA: A mixed style LoRA: the author first stacked multiple LoRAs to create a specific style, then used images generated with that style as the dataset to train this LoRA.
- Dark Star Comics Moss Turbo · 1.0 — Base Model / Checkpoint: A fully fine-tuned MOSS base model in digital comic style (labeled Turbo): the author states it features clean lines, soft bright highlights, neon accents, and strong contrast, targeting character illustrations for fantasy, sci-fi, and urban themes.
Data: Civitai Public API (base model MOSS Image), collection window 10-30 08:31 → 10-30 21:01 (Beijing Time). Recommendations have been reviewed by editors; download counts only reflect popularity.