comfyui-mcp
Version:
Local-first, agent-native control plane for ComfyUI — MCP server + autonomous sidebar agent that drives your live graph in natural language on ANY LLM: Claude/ChatGPT/Gemini on your subscription (no API key), free local models via Ollama (fully offline),
509 lines (485 loc) • 18.7 kB
JSON
{
"$schema": "model-settings.schema.json",
"_version": "0.2.0",
"_description": "Community-maintained best settings for ComfyUI models. Merge order: defaults (this file) → user overrides (model-settings.user.json). PRs welcome!",
"_updated": "2026-02-16",
"_merge_strategy": "deep_merge_user_wins",
"model_families": {
"qwen_image": {
"display_name": "Qwen Image (Base Architecture)",
"architecture": "qwen_image",
"clip_type": "qwen_image",
"vae": "qwen_image_vae.safetensors",
"prompt_style": "natural_language",
"prompt_notes": "Use natural language, 1-3 sentences. Put text to render in quotes. 'photograph' > 'photorealistic'. Negative prompts: use NLP-style descriptions, not keyword spam.",
"resolutions": {
"_note": "Qwen Image operates at ~1.6 megapixels natively",
"square": [1328, 1328],
"portrait_3_4": [1104, 1472],
"portrait_9_16": [928, 1664],
"portrait_2_3": [1056, 1584],
"landscape_4_3": [1472, 1104],
"landscape_16_9": [1664, 928],
"landscape_3_2": [1584, 1056],
"ultrawide_portrait": [1536, 2048]
},
"presets": {
"lightning_4step": {
"description": "Fast 4-step with LightX2V lightning LoRA. Good general-purpose.",
"steps": 4,
"cfg": 1.0,
"sampler": "euler",
"scheduler": "simple",
"denoise": 1.0,
"auraflow_shift": 3.1,
"lora": {
"name": "Qwen-Image-Lightning-4steps-V1.0.safetensors",
"strength": 1.0
},
"notes": "For 2511 edit models, use 2511-specific lightning LoRA."
},
"lightning_4step_2511_edit": {
"description": "Fast 4-step specifically for 2511 edit model.",
"steps": 4,
"cfg": 1.0,
"sampler": "euler",
"scheduler": "simple",
"denoise": 1.0,
"lora": {
"name": "Qwen-Image-Edit-2511-Lightning-4steps-V1.0-bf16.safetensors",
"strength": 1.0
},
"notes": "Cannot combine with distilled model variant."
},
"lightning_8step": {
"description": "Higher quality 8-step lightning. Better detail than 4-step.",
"steps": 8,
"cfg": 1.0,
"sampler": "euler",
"scheduler": "simple",
"denoise": 1.0,
"auraflow_shift": 3.1,
"lora": {
"name": "Qwen-Image-Lightning-8steps-V1.0.safetensors",
"strength": 1.0
}
},
"lightning_8step_character": {
"description": "8-step lightning tuned for detailed character/portrait work.",
"steps": 8,
"cfg": 2.5,
"sampler": "euler",
"scheduler": "simple",
"denoise": 1.0,
"auraflow_shift": 3.1,
"lora": {
"name": "Qwen-Image-Lightning-8steps-V1.0.safetensors",
"strength": 1.0
},
"notes": "CFG 2.5 gives better prompt adherence for character details (Imageized author examples)."
},
"standard": {
"description": "Official full-quality settings from ComfyUI template.",
"steps": 50,
"cfg": 4.0,
"sampler": "euler",
"scheduler": "simple",
"denoise": 1.0,
"auraflow_shift": 3.1,
"notes": "Official ComfyUI workflow template values. 30-50 steps work well."
},
"standard_golden": {
"description": "Community-tested 'golden configuration' for best quality.",
"steps": 50,
"cfg": 4.5,
"sampler": "euler",
"scheduler": "simple",
"denoise": 1.0,
"auraflow_shift": 3.1,
"notes": "Community consensus: CFG 4.5 + 50 steps for peak quality."
},
"standard_character_composition": {
"description": "Optimized for multi-character scenes and complex compositions.",
"steps": 30,
"cfg": 4.0,
"sampler": "euler_ancestral",
"scheduler": "beta",
"denoise": 1.0,
"auraflow_shift": 3.1,
"notes": "euler_ancestral + beta helps with character variety and scene coherence."
}
}
},
"qwen_image_2512": {
"display_name": "Qwen Image 2512 (December 2025)",
"inherits": "qwen_image",
"unet": "qwen_image_2512_fp8_e4m3fn.safetensors",
"clip": "qwen_2.5_vl_7b_fp8_scaled.safetensors",
"vae": "qwen_image_vae.safetensors",
"notes": "Latest official Qwen Image model. fp8 for 24GB VRAM. Inherits all qwen_image presets."
},
"qwen_image_2511_edit": {
"display_name": "Qwen Image Edit 2511 (November 2025)",
"inherits": "qwen_image",
"unet": "qwen_image_edit_2511_bf16.safetensors",
"clip": "qwen_2.5_vl_7b_fp8_scaled.safetensors",
"vae": "qwen_image_vae.safetensors",
"notes": "Optimized for image editing. bf16 precision. Use 2511-specific lightning LoRA.",
"presets": {
"edit_standard": {
"description": "Full quality img2img editing.",
"steps": 40,
"cfg": 4.0,
"sampler": "euler",
"scheduler": "simple",
"denoise": 0.75,
"notes": "Denoise 0.5-0.8: lower = closer to input. Negative conditioning should be zeroed."
}
}
},
"qwen_finetuned": {
"display_name": "Qwen Image Fine-tuned Models",
"architecture": "qwen_image",
"models": {
"qwenUltimateRealism_v11": {
"display_name": "Qwen Ultimate Realism v1.1",
"author": "Imageized",
"source": "https://civitai.com/models/2027494",
"unet_path": "Qwen/imageized/qwenUltimateRealism_v11.safetensors",
"clip": "qwen_2.5_vl_7b_fp8_scaled.safetensors",
"vae": "qwen_image_vae.safetensors",
"focus": "Product photography, hyper-realistic, e-commerce",
"presets": {
"author_quality": {
"description": "Author's best quality settings.",
"steps": 30,
"cfg": 7.5,
"sampler": "euler",
"scheduler": "simple",
"denoise": 1.0,
"notes": "Steps 20-40, CFG 5-10 (sweet spot 7-8)."
},
"author_fast": {
"steps": 4,
"cfg": 1.0,
"sampler": "euler",
"scheduler": "simple",
"lora": {
"name": "Qwen-Image-Lightning-4steps-V1.0.safetensors",
"strength": 1.01
},
"notes": "Author uses strength 1.01 specifically."
}
}
},
"copaxTimeless_qwenUltraRealistic": {
"display_name": "CopaxTimeless Qwen Ultra Realistic",
"author": "Copax",
"source": "https://civitai.com/models/copaxTimeless",
"unet_path": "Qwen/realistic/copaxTimeless_qwenUltraRealistic.safetensors",
"clip": "qwen_2.5_vl_7b_fp8_scaled.safetensors",
"vae": "qwen_image_vae.safetensors",
"focus": "Ultra-realistic portraits and scenes",
"presets": {
"author_recommended": {
"steps": 30,
"cfg": 4.0,
"sampler": "res_multistep",
"scheduler": "sgm_uniform",
"denoise": 1.0,
"auraflow_shift": 3.1,
"notes": "Consistently uses res_multistep + sgm_uniform. Steps 20/30/50 all work."
}
},
"resolutions": [
[1152, 1536], [992, 1776], [928, 1984], [1536, 2048]
]
},
"redcraftRedzimage_DX1": {
"display_name": "RedCraft Redzimage DX1",
"source": "https://civitai.com/models/958009",
"unet_path": "redcraftRedzimageUpdatedJAN30_redzibDX1.safetensors",
"loader": "CheckpointLoaderSimple",
"focus": "Realistic generation (combined checkpoint)",
"presets": {
"distilled_fast": {
"steps": 10,
"cfg": 1.0,
"sampler": "euler",
"scheduler": "simple"
},
"standard": {
"steps": 30,
"cfg": 4.0,
"sampler": "euler",
"scheduler": "simple"
}
},
"notes": "Combined checkpoint (UNET+CLIP+VAE). Use CheckpointLoaderSimple, not separate loaders."
}
}
},
"z_image": {
"display_name": "Z-Image Family",
"architecture": "z_image",
"notes": "Must use Qwen tokenizer, not standard CLIP nodes.",
"models": {
"z_image_turbo": {
"display_name": "Z-Image Turbo",
"unet": "z_image_turbo_bf16.safetensors",
"clip": "qwen_3_4b.safetensors",
"vae": "z-image-ae.safetensors",
"clip_type": "qwen_image",
"presets": {
"author_recommended": {
"steps": 14,
"cfg": 1.0,
"sampler": "res_2s",
"scheduler": "simple",
"notes": "From CopaxTimeless author. Distilled CFG 1."
},
"beauty_fashion": {
"steps": 10,
"cfg": 1.0,
"sampler": "euler_ancestral",
"scheduler": "beta",
"notes": "Smooth skin, ideal for beauty/fashion (560-image test)."
},
"sharpest": {
"steps": 10,
"cfg": 1.0,
"sampler": "dpmpp_sde",
"scheduler": "beta",
"notes": "Sharpest, most natural results from community testing."
}
},
"negative_prompt_template": "over-smooth skin, plastic skin, doll face, anime, CGI, waxy texture, blurry face, fake pores, exaggerated makeup, over-sharpening, unrealistic symmetry, flat lighting, low detail skin, extra fingers, distorted anatomy"
}
}
},
"sdxl": {
"display_name": "SDXL 1.0",
"architecture": "sdxl",
"clip_type": "sdxl",
"prompt_style": "tags_and_natural",
"presets": {
"realistic_photo": {
"description": "Gold standard for SDXL photorealism.",
"steps": 25,
"cfg": 5.0,
"sampler": "dpmpp_2m",
"scheduler": "karras",
"notes": "DPM++ 2M Karras. CFG 5-7. AVOID Euler for SDXL (causes washed-out images)."
},
"realistic_creative": {
"description": "More diverse realistic results.",
"steps": 30,
"cfg": 6.0,
"sampler": "dpmpp_2m_sde",
"scheduler": "karras",
"notes": "DPM++ 2M SDE Karras for more creative variation."
},
"realistic_best": {
"description": "Highest quality realistic (slow).",
"steps": 35,
"cfg": 6.0,
"sampler": "heun",
"scheduler": "normal",
"notes": "Heun is 2x compute per step but highest quality. Use for final renders."
},
"anime_pony": {
"description": "Pony Diffusion V6 XL author settings.",
"steps": 25,
"cfg": 7.0,
"sampler": "euler_ancestral",
"scheduler": "normal",
"clip_skip": 2,
"notes": "CLIP SKIP 2 IS CRITICAL for Pony — model outputs blobs without it. No neg prompt needed. Use score_9, score_8_up tags."
},
"anime_sharp": {
"description": "Sharp anime with SDE sampler.",
"steps": 25,
"cfg": 7.5,
"sampler": "dpmpp_sde",
"scheduler": "karras",
"notes": "DPM++ SDE Karras for sharper anime. Slower than 2M."
}
},
"resolutions": {
"square": [1024, 1024],
"portrait_3_4": [896, 1152],
"portrait_9_16": [768, 1344],
"landscape_4_3": [1152, 896],
"landscape_16_9": [1344, 768]
}
},
"illustrious": {
"display_name": "Illustrious XL",
"architecture": "sdxl",
"clip_type": "sdxl",
"prompt_style": "booru_tags",
"prompt_notes": "Use danbooru tags: 1girl, solo, long_hair, blue_eyes. Quality: masterpiece, best quality, absurdres. CFG > 7.5 causes oversaturation, < 3 loses color.",
"presets": {
"default": {
"description": "Community standard Illustrious settings.",
"steps": 25,
"cfg": 4.5,
"sampler": "euler_ancestral",
"scheduler": "simple",
"notes": "CFG 4.5-5.0 is the sweet spot. euler_ancestral is community favorite."
},
"detailed": {
"steps": 30,
"cfg": 5.0,
"sampler": "dpmpp_2m",
"scheduler": "karras",
"notes": "More consistent/detailed. For complex compositions, increase steps to 40."
},
"sharp_shading": {
"steps": 15,
"cfg": 5.0,
"sampler": "dpmpp_2s_ancestral",
"scheduler": "karras",
"notes": "Enhanced shading detail. Watch for noise in flat color areas."
}
},
"resolutions": {
"optimal_range": "1024-1536px per side"
}
},
"flux": {
"display_name": "Flux.1",
"architecture": "flux",
"clip_type": "flux",
"prompt_style": "natural_language",
"prompt_notes": "Natural language descriptions. No quality tags needed. Detailed, descriptive prompts work best.",
"notes": "Flux uses FluxGuidance node (guidance=3.5), NOT KSampler CFG. Set KSampler CFG=1.0 always.",
"presets": {
"dev": {
"description": "Flux.1-dev standard.",
"steps": 20,
"cfg": 1.0,
"sampler": "euler",
"scheduler": "simple",
"guidance": 3.5,
"notes": "Guidance 3.5 via FluxGuidance node. Steps 20-28. Short prompts: guidance 3.5-4.0, long prompts: 1.0-1.5."
},
"schnell": {
"description": "Flux.1-schnell (1-4 steps, no LoRA needed).",
"steps": 4,
"cfg": 1.0,
"sampler": "euler",
"scheduler": "simple"
},
"flux2_turbo": {
"description": "Flux 2 with turbo LoRA.",
"steps": 4,
"cfg": 1.0,
"sampler": "euler",
"scheduler": "simple",
"lora": {
"name": "flux2-turbo-lora.safetensors",
"strength": 1.0
}
}
},
"resolutions": {
"square": [1024, 1024],
"portrait_3_4": [896, 1152],
"landscape_4_3": [1152, 896]
}
},
"sd35": {
"display_name": "Stable Diffusion 3.5",
"architecture": "sd3",
"presets": {
"large": {
"steps": 25,
"cfg": 4.5,
"sampler": "dpmpp_2m",
"scheduler": "normal"
},
"large_turbo": {
"steps": 5,
"cfg": 1.0,
"sampler": "euler",
"scheduler": "normal"
},
"medium": {
"steps": 28,
"cfg": 4.5,
"sampler": "euler",
"scheduler": "normal"
}
}
}
},
"lightning_loras": {
"_description": "Index of available lightning/turbo LoRAs.",
"qwen_image_4step": {
"file": "Qwen-Image-Lightning-4steps-V1.0.safetensors",
"source": "https://huggingface.co/lightx2v/Qwen-Image-2512-Lightning",
"steps": 4,
"compatible_with": ["qwen_image", "qwen_image_2512", "qwen_finetuned"],
"strength": 1.0,
"cfg": 1.0,
"speedup": "12-25x"
},
"qwen_image_8step": {
"file": "Qwen-Image-Lightning-8steps-V1.0.safetensors",
"source": "https://huggingface.co/lightx2v/Qwen-Image-2512-Lightning",
"steps": 8,
"compatible_with": ["qwen_image", "qwen_image_2512", "qwen_finetuned"],
"strength": 1.0,
"cfg": 1.0,
"notes": "Better quality than 4-step. Use CFG 2.5 for character detail."
},
"qwen_edit_2511_4step": {
"file": "Qwen-Image-Edit-2511-Lightning-4steps-V1.0-bf16.safetensors",
"source": "https://huggingface.co/lightx2v/Qwen-Image-Edit-2511-Lightning",
"steps": 4,
"compatible_with": ["qwen_image_2511_edit"],
"strength": 1.0,
"cfg": 1.0,
"speedup": "~10x",
"notes": "Cannot combine with distilled model."
},
"wuli_turbo_v3": {
"file": "NOT_INSTALLED",
"source": "https://huggingface.co/Wuli-art/Qwen-Image-2512-Turbo-LoRA",
"steps": 4,
"compatible_with": ["qwen_image_2512"],
"strength": 1.0,
"cfg": 1.0,
"speedup": "20x",
"notes": "Wuli-art V3.0 (Jan 2026). CFG-distilled. Also has 2-step and 8-step variants. May need download."
},
"flux2_turbo": {
"file": "flux2-turbo-lora.safetensors",
"steps": 4,
"compatible_with": ["flux"],
"strength": 1.0,
"cfg": 1.0
}
},
"sampler_guide": {
"_description": "Quick reference for sampler/scheduler by use case.",
"general_fast": { "sampler": "euler", "scheduler": "simple", "notes": "Works everywhere. Default." },
"sdxl_realistic": { "sampler": "dpmpp_2m", "scheduler": "karras", "notes": "SDXL gold standard. AVOID Euler on SDXL." },
"sdxl_anime": { "sampler": "euler_ancestral", "scheduler": "normal", "notes": "Pony/anime SDXL. Clip skip 2 for Pony." },
"illustrious": { "sampler": "euler_ancestral", "scheduler": "simple", "notes": "CFG 4.5-5.0 sweet spot." },
"character_composition":{ "sampler": "euler_ancestral", "scheduler": "beta", "notes": "Multi-character scenes, varied compositions." },
"qwen_realistic_hq": { "sampler": "res_multistep", "scheduler": "sgm_uniform", "notes": "CopaxTimeless pick for Qwen." },
"z_image_sharp": { "sampler": "dpmpp_sde", "scheduler": "beta", "notes": "Sharpest Z-Image results (560-image test)." },
"z_image_beauty": { "sampler": "euler_ancestral", "scheduler": "beta", "notes": "Smooth skin, fashion/beauty." },
"z_image_turbo": { "sampler": "res_2s", "scheduler": "simple", "notes": "Z-Image Turbo distilled." }
},
"_contributing": {
"how_to_contribute": "Fork the repo, add your model/preset, submit a PR.",
"required_fields": ["Model name + author", "Source URL", "At least one preset (steps/cfg/sampler/scheduler)", "Resolution if non-standard"],
"optional_fields": ["Lightning LoRA compatibility", "Prompt style tips", "Known quirks", "img2img denoise range"],
"contributors": [
{ "name": "artokun", "role": "maintainer", "date": "2026-02-16" }
]
}
}