🎨 SeeMoss-Genesis

My anime look on a new base: MOSS-Image — natural language, tags and JSON-layout prompting

🚀 Quick start: put the file you chose in ComfyUI/models/diffusion_models/ (the text encoder and VAE are separate files). Standard (Preview): 25 steps, CFG 2.5, Shift 5, euler / simple. Turbo (Preview-Turbo): 8 steps, CFG 1, Shift 5. Trigger words: seesee_elf, seesee_kitsune.

✨ At a glance

Property	Details
🧩 Type	Anime fine-tune of MOSS-Image-SFT (diffusion model files, no AIO)
🏗️ Base model	OpenMOSS-Team/MOSS-Image-SFT (~5B DiT, Apache-2.0)
📦 Versions	🟢 Preview (standard) · 🚀 Preview-Turbo (distilled)
📦 Formats	BF16 · fp8_scaled · int8_convrot · mxfp8 · nvfp4 · int4_convrot
🎛️ Settings	Standard: 25 steps · CFG 2.5 · Shift 5 — Turbo: 8 steps · CFG 1 · Shift 5
🔑 Triggers	seesee_elf · seesee_kitsune · pixel art and Game Boy look
🧰 Workflows	In workflows/ (Civitai page)

🟢 Standard model: Preview

The non-Turbo version. Civitai: Preview · Files: Preview/

• ⚡ Base (default): 25 steps · CFG 2.5 · Shift 5 · euler / simple
• 💎 Best quality: 50 steps · CFG 3 · Shift 5 · euler / simple
• Skip Layer Guidance: on, layer 22 in the negative branch (MOSS-Image default)
• Files: moss_anime_sft_{BF16, fp8_scaled, int8_convrot, mxfp8, nvfp4, int4_convrot}.safetensors → models/diffusion_models/

Version note (Civitai): First Preview release of SeeMoss-Genesis on MOSS-Image-SFT. This version contains six files: BF16, fp8_scaled, int8_convrot, mxfp8, nvfp4 and int4_convrot. Put the file you choose into ComfyUI/models/diffusion_models/. Triggers: seesee_elf, seesee_kitsune. Base: 25 steps · CFG 2.5 · Shift 5. Best quality: 50 steps · CFG 3 · Shift 5 · euler / simple. Needs the Text Encoder and VAE versions from this page. A Turbo version is available as Preview-Turbo.

🖼️ Standard (Preview) gallery

All images below were posted to the Preview version on Civitai.

---

🚀 Turbo model: Preview-Turbo

The Turbo distilled version for fast, low-step generation. Civitai: Preview-Turbo · Files: Preview-Turbo/

• 🚀 Settings: CFG 1 · 4–8 steps (8 recommended) · Shift 5 · euler / simple · Skip Layer Guidance off
• 🔬 Optional Detail Refiner (group 07 in my workflows): latent upscale 1.5× · 8 steps · CFG 1 · denoise 0.5
• Files: moss_anime_turbo_sft_{BF16, fp8_scaled, int8_convrot, mxfp8, nvfp4, int4_convrot}.safetensors → models/diffusion_models/

Version note (Civitai): Turbo distilled Preview of SeeMoss-Genesis. All six files: BF16, fp8_scaled, int8_convrot, mxfp8, nvfp4, int4_convrot. Settings: CFG 1, 4-8 steps (8 recommended), euler / simple, Shift 5. Optional Detail Refiner: Latent Upscale 1.5x, 8 steps, CFG 1, denoise 0.5. Needs the same Text Encoder and VAE versions as Preview. Files go in models/diffusion_models/.

🖼️ Turbo (Preview-Turbo) gallery

All images below were posted to the Preview-Turbo version on Civitai.

---

✨ What is SeeMoss-Genesis?

SeeMoss-Genesis is my personal anime model built on MOSS-Image-SFT by OpenMOSS. It is trained on the same dataset as my See-Krea2-Genesis, so the same direction carries over to a new base: detailed anime women, rich backgrounds, my own pixel-art look and Game Boy-inspired imagery.

🧪 This is a Preview. It is an early release so you can try the look on MOSS-Image now. Expect updates. 🚀 Turbo is here: the Preview-Turbo version is available for fast, low-step generation (CFG 1 · 4–8 steps).

---

🆕 What does it bring?

• Anime as the natural direction. Detailed anime characters and backgrounds, with attention to both the character and the scene around them.
• Two trained triggers: seesee_elf and seesee_kitsune.
• My own pixel-art look. It can be requested with different phrases, including a simple pixel art cue.
• Game Boy-inspired imagery. My own Game Boy captures were included in the training data.
• MOSS-Image prompting. Long natural-language descriptions work best. Tag lists and JSON layouts are supported by the base model. JSON layout control is still in testing: my first tests with the Bounding Box Editor workflow follow the layout well, but boxes are a strong hint, not pixel-perfect.

📝 About English, Japanese and manga text: MOSS-Image is strong at rendering text, and my dataset includes English and Japanese text and manga material. I have not confirmed how well readable text holds up in this Preview. Please treat spelling, lettering and longer text as something to test, rather than a promised feature.

---

🧝 Trained character triggers

These two trigger words were included during training. Use the exact spelling at the start of your positive prompt, then describe the scene, pose, clothing and lighting you want.

🧝 seesee_elf

Trigger: seesee_elf

EXAMPLE 01 · SEESEE_ELF

Example prompt

seesee_elf, A single female character stands leaning against a glowing pink vending machine in a neon-lit urban alleyway. The image is rendered in a high-quality, semi-realistic digital anime style with sharp focus and vibrant color grading. The framing is a full-body vertical shot, capturing the subject from head to toe with a slight low-angle perspective. She leans her left hip against the machine, her left hand tucked into her skirt pocket, while her right hand holds a pink beverage can at chest level. Her gaze is directed forward, looking slightly past the camera. She has long, straight platinum blonde hair that falls past her waist, pointed elf ears, and large amber eyes. Her face is pale with a neutral, calm expression. Her body is slender with long legs covered in sheer black pantyhose. She wears a cropped pink blazer over a black bandeau top, a matching high-waisted pink mini skirt, and black platform high-heeled shoes with ankle straps. A black lace choker and a silver necklace with a cross pendant adorn her neck, and long black earrings dangle from her ears. The background features a wet, reflective cobblestone street and blurred neon signs, including a vertical pink sign reading "OPEN" and a purple sign reading "BAR". The vending machine displays Japanese text "ドリンク" at the top and contains rows of colorful bottles. The lighting is dominated by the intense pink glow from the vending machine, which casts a strong magenta hue over the scene and creates sharp reflections on the wet ground.

50 steps · CFG 3 · Shift 5 · euler / simple · seed 5021 · 768×1360

🦊 seesee_kitsune

Trigger: seesee_kitsune

EXAMPLE 02 · SEESEE_KITSUNE — the original example prompt and image are not reproduced on this page.

50 steps · CFG 3 · Shift 5 · euler / simple · seed 7004 · 832×1248

👾 Pixel art & Game Boy look

Add pixel art (or similar phrases like pixel-art style, retro handheld game screen) to your description to pull the image toward my pixel look and the Game Boy-inspired direction.

EXAMPLE 03 · PIXEL ART COCKPIT

A solitary figure sits in a command chair facing a vast spaceship viewport, rendered in pixel art with a visible grid of square pixels, a limited palette of neon blues, magentas, and reds, and hard-edged blocky shapes. The camera angle is a symmetrical, eye-level rear view from behind the pilot, looking forward into the cockpit. The figure is a dark silhouette with a rounded head, seated centrally and facing away from the viewer toward the controls. The cockpit interior is dense with angular consoles on the left and right, featuring small glowing blue screens and clusters of red, orange, and green square buttons. The large, multi-paned window reveals outer space, dominated by a large blue planet with bright rings on the left and a swirling nebula of pink, purple, and orange on the right. A small, cyan wireframe sphere floats in the nebula. The floor is dark blue with red rectangular lights, and thick, segmented pipes run along the bottom corners of the frame.

50 steps · CFG 3 · Shift 5 · euler / simple · seed 5023 · 1248×832

EXAMPLE 04 · PIXEL ART CATGIRL — the original example prompt and image are not reproduced on this page.

50 steps · CFG 3 · Shift 5 · euler / simple · seed 4024 · 832×1248

EXAMPLE 05 · GAME BOY LOOK

A blue cat character stands on a grassy ledge in a retro 8-bit video game level, rendered in Game Boy Color era pixel art with a visible coarse pixel grid, small resolution, and limited color palette. The framing is a horizontal side-scrolling view showing a light blue sky with white clouds, a distant green mountain range, and a tan sandy ground area. The blue cat faces forward with a simple closed-eye expression, positioned near small orange flowers and green bushes on the right. Floating above are two yellow, round creatures with green accents, one resting on a small floating platform and another hovering in the upper right corner. A brown wooden crate sits on the far left edge, and a blue water hazard spans the bottom left. The bottom of the screen displays a black HUD bar with white pixelated text reading "WORLD 1", a counter showing "x4", and a timer reading "T:79".

50 steps · CFG 3 · Shift 5 · euler / simple · seed 5025 · 1152×864

🖼️ NEW: Image Edit version

SeeMoss-Genesis can now edit existing pictures with a short instruction, using the MOSS-Image Edit LoRA. Change outfits, hair color, season or accessories while face, pose and composition stay the same. Works with Preview (25 steps, CFG 4) and Preview-Turbo (8 steps, CFG 1.5).

👉 Workflow: SeeMoss-Genesis-Workflows → version Image Edit v1.0 (here: workflows/SeeMoss-Genesis-Image-Edit-v1.0.json) · Edit LoRA: moss_image_edit_lora_rank_64_bf16

🗺️ NEW: ControlNet version

SeeMoss-Genesis can now follow the pose and composition of a reference picture with depth control (MOSS-Image color/depth adapter by darask0). The prompt decides outfit, scene and style; an optional color map switch also keeps the room layout of the reference.

👉 Workflow: SeeMoss-Genesis-Workflows → version ControlNet v1.0 (here: workflows/SeeMoss-Genesis-ControlNet-v1.0.json) · Adapter: moss_image_color_depth_control

---

⚙️ Recommended settings

• ⚡ Base (default): 25 steps · CFG 2.5 · Shift 5 · euler / simple
• 💎 Best quality: 50 steps · CFG 3 · Shift 5 · euler / simple
• 🚀 Turbo (Preview-Turbo): CFG 1 · 4–8 steps (8 recommended) · Shift 5 · euler / simple · Skip Layer Guidance off · optional Detail Refiner: latent upscale 1.5× · 8 steps · CFG 1 · denoise 0.5 (group 07 in my workflows)
• Skip Layer Guidance (Base / Best quality): on, layer 22 in the negative branch (MOSS-Image default)
• Resolution: around 1–2 megapixels, for example 1024×1024 or 832×1472 portrait. Width and height must be multiples of 16.
• Text encoder: qwen3.5_9b_w4a8_gptq (CLIP type moss_image) · VAE: flux2-vae_small. Both are included in this repository (Text-Encoder/, VAE/).

Details per model: see 🟢 Standard and 🚀 Turbo above; workflows: see 🧩 Workflows below.

---

📦 Which file should I use?

• BF16 (9.5 GB): full quality
• fp8_scaled / int8_convrot / mxfp8 (about 5 GB): good balance. mxfp8 and fp8 are best on newer NVIDIA GPUs
• nvfp4 (3.1 GB): for RTX 50-series (Blackwell)
• int4_convrot (2.8 GB): smallest, for low VRAM
• Preview-Turbo: the same six formats (BF16 · fp8_scaled · int8_convrot · mxfp8 · nvfp4 · int4_convrot) in the Turbo version
• Text Encoder and VAE: separate versions, needed for all model files (Preview and Preview-Turbo)

I have not compared the quantized files side by side yet.

Files in this repository

Folder	Civitai version	Files	ComfyUI folder
Preview/	Preview	moss_anime_sft_{BF16, fp8_scaled, int8_convrot, mxfp8, nvfp4, int4_convrot}.safetensors	models/diffusion_models/
Preview-Turbo/	Preview-Turbo	moss_anime_turbo_sft_{BF16, fp8_scaled, int8_convrot, mxfp8, nvfp4, int4_convrot}.safetensors	models/diffusion_models/
Text-Encoder/	Text Encoder	qwen3.5_9b_w4a8_gptq.safetensors (CLIP type moss_image)	models/text_encoders/
VAE/	VAE	flux2-vae_small.safetensors	models/vae/

Version notes (Civitai)

• Preview: First Preview release of SeeMoss-Genesis on MOSS-Image-SFT. This version contains six files: BF16, fp8_scaled, int8_convrot, mxfp8, nvfp4 and int4_convrot. Put the file you choose into ComfyUI/models/diffusion_models/. Triggers: seesee_elf, seesee_kitsune. Base: 25 steps · CFG 2.5 · Shift 5. Best quality: 50 steps · CFG 3 · Shift 5 · euler / simple. Needs the Text Encoder and VAE versions from this page.
• Preview-Turbo: Turbo distilled Preview of SeeMoss-Genesis. All six files: BF16, fp8_scaled, int8_convrot, mxfp8, nvfp4, int4_convrot. Settings: CFG 1, 4-8 steps (8 recommended), euler / simple, Shift 5. Optional Detail Refiner: Latent Upscale 1.5x, 8 steps, CFG 1, denoise 0.5. Needs the same Text Encoder and VAE versions as Preview. Files go in models/diffusion_models/.
• Text Encoder: Text encoder for SeeMoss-Genesis / MOSS-Image: qwen3.5_9b_w4a8_gptq.safetensors. Put it into ComfyUI/models/text_encoders/ and load it with CLIP type moss_image. Required for every SeeMoss-Genesis file.
• VAE: VAE for SeeMoss-Genesis / MOSS-Image: flux2-vae_small.safetensors. Put it into ComfyUI/models/vae/. Required for every SeeMoss-Genesis file.

---

🧩 Workflows

Four ComfyUI workflows from SeeMoss-Genesis-Workflows are included in workflows/. Each block below shows the real workflow overview, what it does, its default settings, the required models (with download links and ComfyUI folders) and the custom nodes. Result images are folded below each workflow. Images in these blocks come from the respective workflow version on Civitai.

Hugging Face copies: the content mode switch defaults to SFW and the optional NSFW prompt-enhancer system prompt text was removed; everything else is unchanged.

Ultimate v1.0

Download: workflows/seemossGenesis_ultimateV10.json · Civitai: Ultimate v1.0

What it does: Text-to-image: write an idea and get a detailed anime illustration, with optional Prompt Enhancer, content mode switch (default SFW on this page), optional LoRA loader, 🔬 Detail Refiner (latent upscale 1.5× + refine) and 🔼 Tiled Upscale. Every stage saves its own PNG with Civitai metadata.

⚙️ Settings: Default model: Preview (Standard). Base 25 steps · CFG 2.5 · Shift 5 · euler / simple · best quality 50 steps · CFG 3 · Shift 5 · Skip Layer Guidance optional (layer 22). With Preview-Turbo: CFG 1 · 4–8 steps.

📥 Required models

File	Required?	Download	ComfyUI folder
Diffusion model	required	moss_anime_sft_int8_convrot.safetensors (Preview / Standard, workflow default; other Preview formats in Preview/)	models/diffusion_models/moss_anime/ (the workflow loads moss_anime\moss_anime_sft_int8_convrot.safetensors)
Text encoder (CLIP type moss_image)	required	qwen3.5_9b_w4a8_gptq.safetensors	models/text_encoders/
VAE	required	flux2-vae_small.safetensors	models/vae/
Upscale model (🔬 Detail Refiner / 🔼 Tiled Upscale)	optional	2x-AnimeSharpV4_RCAN.safetensors by Kim2091	models/upscale_models/

🧩 Required custom nodes

Custom node pack (external)	Nodes used in this workflow	Used for
SeeSee Civitai Metadata	SeeSeeCivitaiSaveImage	save nodes with Civitai-readable metadata (my own node pack)
ComfyUI-Lora-Manager	Lora Loader (LoraManager), TriggerWord Toggle (LoraManager)	LoRA loader + trigger words
rgthree-comfy	Fast Groups Bypasser, Image Comparer	group switches, image comparers
ComfyUI_UltimateSDUpscale	UltimateSDUpscale	🔼 Tiled Upscale
comfy-image-saver	Seed Generator	seed

Install with ComfyUI-Manager (search the name) or git clone into custom_nodes/, then restart ComfyUI.

Everything else is ComfyUI core. Needs ComfyUI 0.42.0+ (native MOSS-Image support, subgraphs).

🖼️ Ultimate v1.0 · result images (3)

📄 Ultimate v1.0 · full Civitai version description

First release of SeeMoss-Genesis Ultimate for MOSS-Image: Base 25 steps · CFG 2.5, best quality 50 steps · CFG 3 · Shift 5 (Turbo: CFG 1 · 4–8 steps, see Preview-Turbo). Optional Prompt Enhancer, SFW / NSFW switch, LoRA loader, Detail Refiner and Tiled Upscale. Saves with Civitai metadata.

### Bounding Box Editor v1.0

Download: workflows/SeeMoss-Genesis-Bounding-Boxes-v1.0.json · Civitai: Bounding Box Editor v1.0

What it does: Layout-first text-to-image: draw boxes for character, object and text, caption each box, describe the background in the summary caption, and MOSS-Image builds the scene to your layout, including headlines and typography.

⚙️ Settings: Default model: Preview-Turbo int8 · CFG 1 · 4–8 steps · Shift 5 · euler / simple · 🔬 Detail Refiner (group 07): latent upscale 1.5× + 8 steps · CFG 1 · denoise 0.5. With the Standard Preview model: 25 steps · CFG 2.5 (best quality 50 · CFG 3).

📥 Required models

File	Required?	Download	ComfyUI folder
Diffusion model	required	moss_anime_turbo_sft_int8_convrot.safetensors (Preview-Turbo, workflow default; other Turbo formats in Preview-Turbo/)	models/diffusion_models/moss_anime_turbo/ (the workflow loads moss_anime_turbo\moss_anime_turbo_sft_int8_convrot.safetensors)
Text encoder (CLIP type moss_image)	required	qwen3.5_9b_w4a8_gptq.safetensors	models/text_encoders/
VAE	required	flux2-vae_small.safetensors	models/vae/
Upscale model (🔬 Detail Refiner / 🔼 Tiled Upscale)	optional	2x-AnimeSharpV4_RCAN.safetensors by Kim2091	models/upscale_models/

🧩 Required custom nodes

Custom node pack (external)	Nodes used in this workflow	Used for
SeeSee Civitai Metadata	SeeSeeCivitaiSaveImage	save nodes with Civitai-readable metadata (my own node pack)
ComfyUI-Lora-Manager	Lora Loader (LoraManager), TriggerWord Toggle (LoraManager)	LoRA loader + trigger words
rgthree-comfy	Fast Groups Bypasser, Image Comparer	group switches, image comparers
ComfyUI_UltimateSDUpscale	UltimateSDUpscale	🔼 Tiled Upscale
comfy-image-saver	Seed Generator	seed

Install with ComfyUI-Manager (search the name) or git clone into custom_nodes/, then restart ComfyUI.

The Bounding Box Editor and MOSS Layout Prompt (MOSSImageLayoutPrompt) nodes are ComfyUI core; a recent ComfyUI is needed.

🖼️ Bounding Box Editor v1.0 · result images (3)

📄 Bounding Box Editor v1.0 · full Civitai version description

📐 SeeMoss-Genesis Bounding Box Editor v1.0

A layout-first version of the SeeMoss-Genesis workflow: instead of describing where things go in words, you draw boxes on the canvas and MOSS-Image places character, objects and text inside them. Same models, same save nodes with Civitai metadata, same Detail Refiner — plus a visual 📐 Bounding Box Editor.

🧭 How to use it

1. Open 📐 03 · Bounding Box Layout + Prompt and draw a box for every element in the 📐 Bounding Box Editor.
2. Give each box a label: character (a person, pose and outfit), object (props like a crystal, staff, table, candles) or text (typography).
3. Write a caption per box. For text boxes also fill in the exact words — headline, title, quote. Size and position of the box decide the size and position of the lettering.
4. The background has no box: describe the whole scene, setting and lighting in the summary caption.
5. Character boxes are passed on as character_boxes, so the pose and figure follow your layout. Hit Run.

The default layout is a small example (magical girl with staff, background and a title) — just move the boxes and rewrite the captions.

🧾 MOSS layout JSON

The layout is stored in MOSS-Image's native format: a summary.caption for the whole scene plus a list of items, each with label, caption, optional text and a bbox in yxyx order with coordinates relative to the image center (bbox_coord_space: abs_centralized). You can paste such a JSON directly, or drag one of the gallery PNGs into ComfyUI — every image here has its workflow and box layout embedded.

⚙️ Default settings

• 🚀 Turbo (default): SeeMoss-Genesis Preview-Turbo int8 · CFG 1 · 4–8 steps · Shift 5 · euler / simple
• 🔬 Detail Refiner (group 07): latent upscale 1.5× + 8 steps · CFG 1 · denoise 0.5
• Using the non-Turbo Preview model instead: 25 steps · CFG 2.5 (best quality 50 · CFG 3).
• NSFW mode, Prompt Enhancer and SLG are off by default; Tiled Upscale (06) is optional and off.

📥 Models

All from SeeMoss-Genesis: Preview-Turbo (default) or Preview → models/diffusion_models/ · Text Encoder qwen3.5_9b_w4a8_gptq (CLIP type moss_image) → models/text_encoders/ · VAE flux2-vae_small → models/vae/.

🧩 Required Custom Nodes

• SeeSee Civitai Metadata — save nodes with Civitai-readable metadata
• ComfyUI-Lora-Manager — LoRA loader + trigger words
• rgthree-comfy — group switch, image comparers
• ComfyUI_UltimateSDUpscale — Tiled Upscale
• comfy-image-saver — Seed Generator

The Bounding Box Editor and MOSS Layout Prompt nodes are ComfyUI core — no extra pack, but you need a recent ComfyUI. Update ComfyUI if the nodes show up red.

📝 Honest notes

• Boxes are a strong hint, not pixel-perfect: small objects (e.g. a quill) can come out a bit smaller, and hands don't always land exactly where you drew them.
• Long text renders best in large boxes; tiny text boxes may get garbled lettering.
• The gallery shows the same scenes as the Ultimate version, rebuilt with box layouts.

### Image Edit v1.0

Download: workflows/SeeMoss-Genesis-Image-Edit-v1.0.json · Civitai: Image Edit v1.0

What it does: Edit an existing picture with a short instruction (MOSS-Image Edit LoRA): outfit, hair color, season, accessories, eye color, while face, pose and composition stay the same.

⚙️ Settings: Default model: Preview (Standard) · 25 steps · CFG 4 · Shift 5 · euler / simple. Preview-Turbo: 8 steps · CFG 1.5 (about 3× faster).

📥 Required models

File	Required?	Download	ComfyUI folder
Diffusion model	required	moss_anime_sft_int8_convrot.safetensors (Preview / Standard, workflow default; other Preview formats in Preview/)	models/diffusion_models/moss_anime/ (the workflow loads moss_anime\moss_anime_sft_int8_convrot.safetensors)
Text encoder (CLIP type moss_image)	required	qwen3.5_9b_w4a8_gptq.safetensors	models/text_encoders/
VAE	required	flux2-vae_small.safetensors	models/vae/
MOSS-Image Edit LoRA (loaded in 🧬 LoRA Loader, strength 1)	required	moss_image_edit_lora_rank_64_bf16.safetensors (297 MB, Comfy-Org rehost of chinoll/Moss-Image-Edit)	models/loras/
Upscale model (🔬 Detail Refiner / 🔼 Tiled Upscale)	optional	2x-AnimeSharpV4_RCAN.safetensors by Kim2091	models/upscale_models/

🧩 Required custom nodes

Custom node pack (external)	Nodes used in this workflow	Used for
SeeSee Civitai Metadata	SeeSeeCivitaiSaveImage	save nodes with Civitai-readable metadata (my own node pack)
ComfyUI-Lora-Manager	Lora Loader (LoraManager), TriggerWord Toggle (LoraManager)	LoRA loader + trigger words
rgthree-comfy	Fast Groups Bypasser, Image Comparer	group switches, image comparers
ComfyUI_UltimateSDUpscale	UltimateSDUpscale	🔼 Tiled Upscale
comfy-image-saver	Seed Generator	seed

Install with ComfyUI-Manager (search the name) or git clone into custom_nodes/, then restart ComfyUI.

Everything else is ComfyUI core (TextEncodeMOSSImageEdit, subgraphs). Needs ComfyUI 0.43.0+.

🖼️ Image Edit v1.0 · result images (4)

📄 Image Edit v1.0 · full Civitai version description

🖼️ SeeMoss-Genesis · Image Edit v1.0

A ComfyUI workflow to edit an existing picture with a short instruction, using SeeMoss-Genesis together with the MOSS-Image Edit LoRA. The source image goes in as a reference, the model redraws the same frame and only applies the change you describe: outfit, hair color, season, accessories like glasses, eye color.

The gallery shows original (left) | edited (right), the same showcase images as the Ultimate and Bounding Box versions, each with a different edit. Every edited PNG has the workflow embedded.

▶️ How to use

1. Download the edit LoRA (297 MB) into models/loras/ and keep it active in 🧬 LoRA Loader (strength 1).
2. Load your picture in 🖼️ Source Image (scaled to 1 MP, keeps aspect).
3. Write the change in ✅ Edit Instruction and queue.

✍️ Prompt tips

• Best phrasing: Change only the X to Y. Keep the character, face, hair, pose, background and composition unchanged.
• One change per run works best; chain runs for more.
• Color, outfit, hair and accessory edits are very reliable. For a new season or scene be explicit (e.g. "heavy snow falling outside the windows").
• Up to 4 reference images: connect them to both encode nodes and say what each picture is for.

⚙️ Settings

• Preview (default): 25 steps · CFG 4 · Shift 5 · euler / simple
• Preview-Turbo: 8 steps · CFG 1.5 · about 3× faster
• CFG 4 matters on Preview: lower values can add stray accessories.
• Optional 🔼 Hi-Res (Tiled Upscale / Detail Refiner) like in Ultimate.

📥 Models

• SeeMoss-Genesis Preview · Preview-Turbo
• moss_image_edit_lora_rank_64_bf16.safetensors (Comfy-Org rehost of chinoll/Moss-Image-Edit)
• Text encoder qwen3.5_9b_w4a8_gptq (CLIP type moss_image) and VAE flux2-vae_small: version dropdown on the model page or Comfy-Org/MOSS-Image

🧩 Custom Nodes

• SeeSee Civitai Metadata · save with Civitai metadata
• ComfyUI-Lora-Manager · LoRA Loader
• rgthree-comfy · Hi-Res switch, comparers
• ComfyUI_UltimateSDUpscale · Tiled Upscale
• comfy-image-saver · Seed Generator

Everything else is ComfyUI core. Needs ComfyUI 0.43.0+.

⚠️ Notes

• An edit is slower than text-to-image because the reference is processed every step (~6 min Preview / ~2 min Turbo at 1 MP on an RTX 4060 Ti class GPU).
• Big scene changes can stay partial; small details (gloves, jewellery) may change along with the edit.

### ControlNet v1.0

Download: workflows/SeeMoss-Genesis-ControlNet-v1.0.json · Civitai: ControlNet v1.0

What it does: Keep the pose, framing and silhouette of a reference picture with depth control (MoGe 3 depth map → MOSS-Image color/depth control adapter); the prompt decides character, outfit, scene and style. Optional 🎨 Color Map (default OFF) also keeps the room layout of the reference.

⚙️ Settings: Default model: Preview (Standard) · Base 25 steps · CFG 2.5 · Shift 5 · euler / simple. Control node: strength 1.0 · end 1.0 (try 0.5 for synthetic maps).

📥 Required models

File	Required?	Download	ComfyUI folder
Diffusion model	required	moss_anime_sft_int8_convrot.safetensors (Preview / Standard, workflow default; other Preview formats in Preview/)	models/diffusion_models/moss_anime/ (the workflow loads moss_anime\moss_anime_sft_int8_convrot.safetensors)
Text encoder (CLIP type moss_image)	required	qwen3.5_9b_w4a8_gptq.safetensors	models/text_encoders/
VAE	required	flux2-vae_small.safetensors	models/vae/
Color/depth control adapter (LoraLoaderModelOnly)	required	moss_image_color_depth_control.safetensors (313 MB, Comfy-Org repack of the adapter by darask0)	models/loras/moss/ (the workflow loads moss\moss_image_color_depth_control.safetensors)
Depth estimator MoGe 3 (LoadMoGeModel)	required	moge_3_vitg_fp16.safetensors (2.5 GB)	models/geometry_estimation/
OneFormer-ADE20K (🎨 Color Map only)	optional	downloads automatically on first run of the AIO Aux Preprocessor	(handled by comfyui_controlnet_aux)
Upscale model (🔬 Detail Refiner / 🔼 Tiled Upscale)	optional	2x-AnimeSharpV4_RCAN.safetensors by Kim2091	models/upscale_models/

🧩 Required custom nodes

Custom node pack (external)	Nodes used in this workflow	Used for
SeeSee Civitai Metadata	SeeSeeCivitaiSaveImage	save nodes with Civitai-readable metadata (my own node pack)
ComfyUI-Lora-Manager	Lora Loader (LoraManager), TriggerWord Toggle (LoraManager)	LoRA loader + trigger words
rgthree-comfy	Fast Groups Bypasser, Image Comparer	group switches, image comparers
ComfyUI_UltimateSDUpscale	UltimateSDUpscale	🔼 Tiled Upscale
comfy-image-saver	Seed Generator	seed
comfyui_controlnet_aux	AIO_Preprocessor	🎨 Color Map only (optional, default OFF)

Install with ComfyUI-Manager (search the name) or git clone into custom_nodes/, then restart ComfyUI.

Depth estimation (MoGe) and MOSSImageControl are ComfyUI core. Civitai version page: needs ComfyUI 0.44+; the note inside the workflow says 0.43.0+.

🖼️ ControlNet v1.0 · result images (7)

📄 ControlNet v1.0 · full Civitai version description

🗺️ SeeMoss-Genesis ControlNet v1.0 — Depth Control (+ optional Color Map)

Keep the pose, framing and silhouette of any reference picture and let your prompt decide everything else: character, outfit, scene, light and style. The workflow builds a depth map from the reference automatically (MoGe 3, ComfyUI core) and feeds it into the MOSS-Image color/depth control adapter. Everything from Ultimate is kept: colored groups, notes with download links, SFW/NSFW switch (default SFW), short negative prompt, Prompt Enhancer and Hi-Res options.

✨ What's inside

• 🗺️ 08 · Depth Control — load a reference, the depth map is generated and previewed. Output size follows the reference (~1 MP).
• 🎨 09 · Color Region Map (optional) — one AIO Aux Preprocessor with OneFormer-ADE20K preselected turns the same reference into a flat-color segmentation map (person, wall, window, table…).
• 🔘 Color Map ON / OFF — one switch, default OFF.
  • OFF = only pose/shape from depth, background free from the prompt.
  • ON = also keeps the background/room layout of the reference (windows, walls, furniture positions).
• 🎛️ One MOSS-Image Depth + Color Control node · strength 1.0 · end 1.0 (recommended for maps from real images; try 0.5 for synthetic maps or more freedom).

🔍 Does the color map matter?

Honest answer from my A/B tests (same seed and prompt, see gallery): with depth on, the color map barely changes the character — pose, face and outfit stay almost the same. What it changes is the background layout: windows, arches and furniture land where they were in the reference, even if the prompt asks for an empty beach. So leave it OFF for a new scene, switch it ON to keep the room.

📥 Downloads

• Control adapter: moss_image_color_depth_control.safetensors (Comfy-Org repack) → models/loras/
• Depth estimator: moge_3_vitg_fp16.safetensors → models/geometry_estimation/
• Base model, text encoder and VAE: same as SeeMoss-Genesis Ultimate
• Custom node for the color map: comfyui_controlnet_aux (OneFormer model downloads automatically). Needs ComfyUI 0.44+.

💡 Tips

• The depth map fixes the silhouette — a long gown on a reference wearing shorts may come out shorter. Pick a reference whose outline fits.
• Describe the new outfit and scene fully; nothing is copied from the reference except shape (and layout with the color map ON).
• There is no pose (OpenPose) control for MOSS-Image yet — depth already keeps poses very well.

🙏 Credits: color/depth control adapter by darask0, ComfyUI files by Comfy-Org.

Gallery: reference | depth map | result, the workflow overview, then with vs without color map comparisons (reference | color map | depth only | depth + color | color only | difference).

📄 Full description of the Civitai workflow page (SeeMoss-Genesis-Workflows)

🎨 SeeMoss-Genesis Ultimate — Anime Text-to-Image Workflow for MOSS-Image (v1.0)

ComfyUI workflow for SeeMoss-Genesis, my anime model on MOSS-Image-SFT (OpenMOSS, ~5B DiT). Write an idea, get a detailed anime illustration — with an optional Qwen3.5 prompt enhancer, an SFW / NSFW switch, an optional LoRA loader and two hi-res finishing branches.

Every saved PNG carries Civitai-readable generation data: upload it and Civitai fills in prompt, seed, sampler and the model link by itself.

How it works: 🧠 load models → 🧬 optional LoRAs → ✍️ prompt → ✨ optional enhancer → 🎨 base image → 🔬 Detail Refiner (latent upscale 1.5× + refine) and/or 🔼 Tiled Upscale (1.5×, tile by tile) → 💾 every stage saves its own PNG with Civitai metadata. Everything is grouped, color-coded, switchable and documented with notes inside the workflow, including clickable download links.

---

📐 NEW: Bounding Box Editor version

Second version on this page: Bounding Box Editor v1.0. Same models and finishing as Ultimate, but you lay out the image by drawing boxes — label them character, object or text, caption each one, describe the background in the summary caption, and MOSS-Image builds the scene to your layout, including headlines and typography. Defaults to Preview-Turbo (CFG 1 · 4–8 steps) with the Detail Refiner (group 07). The box nodes are ComfyUI core (recent ComfyUI needed); the custom node list below is the same. Pick the version at the top of the page — the gallery shows the same scenes as Ultimate, rebuilt with box layouts.

---

🗺️ NEW: ControlNet version

Third version on this page: ControlNet v1.0. Same models and finishing as Ultimate, plus depth control: load any reference picture and the image keeps its pose, framing and silhouette while your prompt changes character, outfit, scene and style. The depth map is made automatically (MoGe 3, ComfyUI core) and fed into the MOSS-Image color/depth control adapter by darask0. An optional Color Map (one AIO Aux Preprocessor, OneFormer-ADE20K) can also keep the background/room layout of the reference — one ON / OFF switch, default OFF. Extra downloads: control adapter and MoGe 3; extra custom node: comfyui_controlnet_aux (color map only). Pick the version at the top of the page.

---

⚙️ Recommended settings

• 🟢 Base (default): 25 steps · CFG 2.5 · Shift 5 · euler / simple
• 💎 Best quality: 50 steps · CFG 3 · Shift 5 · euler / simple
• 🚀 Turbo: CFG 1 · 4–8 steps · Preview-Turbo
• Skip Layer Guidance: optional, layer 22 · Triggers: seesee_elf, seesee_kitsune · pixel look: pixel art

---

📥 Required Models

All from the model page: SeeMoss-Genesis on Civitai

• SeeMoss-Genesis (version Preview) → ComfyUI/models/diffusion_models/ · int8_convrot (workflow default), BF16, fp8_scaled, mxfp8, nvfp4 or int4_convrot
• Text encoder qwen3.5_9b_w4a8_gptq (version Text Encoder) → ComfyUI/models/text_encoders/ · CLIP type moss_image
• VAE flux2-vae_small (version VAE) → ComfyUI/models/vae/
• 2x-AnimeSharpV4_RCAN by Kim2091 (optional, hi-res) → ComfyUI/models/upscale_models/ · Download

Upstream: MOSS-Image-SFT · OpenMOSS/MOSS-Image · Comfy-Org repack · ComfyUI guide

⚠️ Civitai renames files on download. Open the workflow and pick your file in each loader.

---

🧩 Required Custom Nodes

Install via ComfyUI-Manager (search the name) or git clone into custom_nodes/:

• SeeSee Civitai Metadata — the save nodes: Civitai-readable metadata in every PNG
• ComfyUI-Lora-Manager — LoRA loader + trigger words
• rgthree-comfy — hi-res group switch, image comparers
• ComfyUI_UltimateSDUpscale — Tiled Upscale
• comfy-image-saver — Seed Generator

Everything else is ComfyUI core. Needs ComfyUI 0.42.0+ (native MOSS-Image support, subgraphs).

---

🛠️ Installation

1. Install the 5 node packs with ComfyUI-Manager → restart ComfyUI.
2. Download model, text encoder, VAE (and optionally AnimeSharp) into the folders above.
3. Open the workflow and select your files in the loaders.
4. Write your idea in ✍️ Positive Prompt, set 🔞 NSFW MODE, switch groups on or off, hit Run. Images land in output/ as SeeMoss-Genesis_…, SeeMoss-Genesis-Refined_… and SeeMoss-Genesis-Tiled-Upscale_….

---

📝 Notes

• ✨ Prompt Enhancer: check 👁️ Final Prompt Preview before important renders. MOSS-Image likes long, natural-language English descriptions.
• 🔞 NSFW MODE switches the enhancer system prompt and the negative prompt together.
• Adult content: consenting-adult content only, where platform and law allow.

Credits: OpenMOSS (MOSS-Image, Apache 2.0) · Comfy-Org · Kim2091 (AnimeSharpV4) · rgthree · willmiao (LoRA Manager) · ssitu (Ultimate SD Upscale) · giriss (comfy-image-saver) · model + workflow: SeeSeeLP

---

💖 Special Thanks

A big thank you to @Fischi for taking the time to run tests with the model.

---

📜 License & credits

Based on MOSS-Image-SFT by OpenMOSS, released under the Apache License 2.0. Thanks to the OpenMOSS team and Comfy-Org for the ComfyUI integration.

• Text encoder and VAE files are the MOSS-Image components as distributed for ComfyUI (Comfy-Org/MOSS-Image, Apache-2.0; Qwen3.5 base: Apache-2.0).
• This repository uses Apache-2.0, consistent with the base model.
• Author: SeeSee21 on Hugging Face · SeeSeeLP on Civitai

---

🧭 More from SeeSee — all models

🧩 Checkpoints & AIO

Model	Hugging Face	Civitai
AniSee	SeeSee21/AniSee	Model · Workflows
Z-Anime	SeeSee21/Z-Anime	Model
Z-Image-Turbo-AIO	SeeSee21/Z-Image-Turbo-AIO	Model · Workflow
ERNIE-AIO	SeeSee21/ERNIE-AIO	Model · Workflow
Chroma-Anime-AIO	SeeSee21/Chroma-Anime-AIO	Model · Workflow
FLUX.2-klein-4B-AIO	SeeSee21/FLUX.2-klein-4B-AIO	Model · Workflow
QWEN-Anime	SeeSee21/QWEN-Anime	Model · Workflow
SeeMoss-Genesis	SeeSee21/SeeMoss-Genesis	Model · Workflow
See-Krea2	SeeSee21/Krea-2-See-Krea2-Genesis	Model · Workflow

🎀 LoRAs

Model	Hugging Face	Civitai
Illustrious Anime Collection	SeeSee21/Illustrious-Anime-Collection	Model
Chroma Semi-Anime	SeeSee21/chroma-Semi-Anime	Model
Kitsune Furry Girl	SeeSee21/Kitsune-Furry-Girl-Z-Image	Model
Ayase Momo (Dandadan)	SeeSee21/Ayase-Momo-Z-Image	Model

🧩 Checkpoints & AIO  ·  🎀 LoRAs  ·  🌐 Civitai  ·  ▶️ YouTube

_{Made with ❤️ by SeeSee}