[MOSS-5B] Modern TV Anime Screencap / Sakuga Broadcast (in sakuga_screencap style)

作者：MaisonLatentia

[MOSS-5B] Modern TV Anime Screencap / Sakuga Broadcast (in sakuga_screencap style)

Mainstream Showcase Edition (civitai.com) | Curated by Maison Latentia

Base Model: MOSS-Image Base 5B | Category: Style LoRA | Trigger: in sakuga_screencap style

Quickstart Settings (Read Before Generating)

• Trigger Phrase: in sakuga_screencap style

• Standalone Weight: 0.80 (0.65 - 1.00)

• Character LoRA Stack Weight: 0.65 - 0.70 (with OC LoRA @ 0.85 - 0.95)

• Sampler / Steps / CFG: Euler / Simple / 36 Steps / CFG 3.0 / Shift 5.0 / SLG Layer 22

• Model AutoV2: 2a6ee5fcd8 (sakuga_screencap_moss5b_comfy.safetensors)

1-Click Copy-Paste Prompts

• Fantasy Sakuga Action (1344x768, Weight 0.80): in sakuga_screencap style, Inside a shattered stone gothic cathedral lit by stained glass moonlight, a silver-armored knight plants both boots and parries a colossal blazing greatsword. Bright orange fire embers and sharp white spark streaks burst violently across the cracked marble floor.

• Cyberpunk Street / Character (896x1152, Weight 0.80): in sakuga_screencap style, A young female courier wearing a high-collar bright yellow techwear jacket and dark cargo pants sits astride a sleek matte-black electric motorcycle under a concrete highway overpass. Sharp cyan and amber city spotlights glint along the bike fairing and wet road behind her.

• Recommended Negative: photorealistic, live action, 3d render, cgi, realistic skin pores, photographic skin, blurry, washed out, bad anatomy, extra fingers, deformed eyes, watermark, signature, misspelled text, gibberish text, letterbox, black bars, sakuga_screencap text

Drag & Drop Workflow Included: Every gallery image has embedded Civitai + ComfyUI generation metadata (AutoV2: 2a6ee5fcd8), and sakuga_screencap_moss5b_workflow.json is attached directly.

About This Style LoRA

Trained on MOSS-Image Base 5B (bf16) using Maison Latentia's curated 60-plate modern TV anime screencap dataset (1344x768, 896x1152, 1024x1024), sakuga_screencap converts MOSS-Image's native semi-painterly output into crisp broadcast anime key-animation linework, high-contrast cel shading, luminous FX streaks, and clean background art.

• Training Spec: 2040 steps / 34 epochs | Rank 16 / Alpha 16 | MUON optimizer | bf16 | 60 curated multi-bucket plates

• Character LoRA Cross-Compatibility: Stacks cleanly with Maison Latentia MOSS-Image 5B character LoRAs (Clementine zclclementine, Nola znolanola, Cassidy zcscassidy) at 0.65 - 0.70 style weight + 0.85 - 0.95 character weight.

• Tip for Dense Scenes: If a close-up character prompt shows minor hand/limb tension at 0.80, bumping style weight to 1.00 (as in the commuter train dusk plate) or lowering to 0.65 with an OC LoRA resolves anatomy cleanly.

ComfyUI Installation

• Place sakuga_screencap_moss5b_comfy.safetensors into ComfyUI/models/loras/ (AutoV2: 2a6ee5fcd8)

• Drag and drop sakuga_screencap_moss5b_workflow.json directly onto your ComfyUI canvas (MOSS-5B + Shift 5.0 + SLG 22 pre-configured)

• Lead Cover Plate: sc_04_fantasy_cathedral_duel_w0.8.png