ImageGen / yaml /chain_features.yaml
RioShiina's picture
chore: refresh ComfyUI deps, fix stale model sources, gate V3 input normalization
08bcdf2 verified
Raw History Blame Contribute Delete
26.6 kB
# Complete Feature & Chain Definitions Configuration for MCP Tools
# Every chain injector in chain_injectors/ corresponds 1-to-1 with an entry here (21 injectors total).
lora:
chains: lora
display_name: "LoRA Fine-tuning Injector"
description: "Injects LoRA weights into UNet/DiT model and CLIP text encoder for custom style, character, or domain adaptation. Note: For Civitai, specify the Version ID (modelVersionId) instead of the main Model ID."
max_count: 5
usage_guideline: "Specify source ('Civitai', 'Hugging Face', or 'Upload File'). For Civitai, provide lora_value as the Version ID (modelVersionId). For Hugging Face, provide HF repo file path. For Upload File, provide uploaded filename. Scale (0.0~2.0) controls weight strength."
parameters_schema:
type: object
properties:
source:
type: string
enum: ["Civitai", "Hugging Face", "Upload File"]
description: "Download/upload source for the LoRA model. Use 'Civitai' to download by Version ID (modelVersionId), 'Hugging Face' to download by repo file path, or 'Upload File' for uploaded files."
lora_value:
type: string
description: "For Civitai: the Version ID (modelVersionId). For Hugging Face: repo_id/filename.extension or repo_id/folder_path/filename.extension. For Upload File: uploaded filename."
scale:
type: number
default: 1.0
minimum: 0.0
maximum: 2.0
description: "Unified strength applied to both the UNet/DiT model and CLIP text encoder (0.0 to 2.0)."
required:
- source
- lora_value
ipadapter:
chains: ipadapter
display_name: "IP-Adapter Image Prompt"
description: "Uses reference images to guide generation style, composition, structure, or face appearance without prompt text restrictions (SD1.5 & SDXL)."
max_count: 5
usage_guideline: "Supply global settings (preset, embeds_scaling, combine_method, final_weight) and up to 5 reference images with individual weights. Preset must match the target model architecture (SD1.5 or SDXL)."
parameters_schema:
type: object
properties:
image:
type: string
description: "Reference image encoded as Base64 Data URI (e.g., data:image/png;base64,...) or HTTP/HTTPS URL."
weight:
type: number
default: 1.0
minimum: 0.0
maximum: 2.0
description: "Influence weight of the individual image prompt (0.0 to 2.0)."
preset:
type: string
default: "STANDARD (medium strength)"
description: "IPAdapter preset model variant loaded from ipadapter.yaml. Must match model architecture (SD1.5 vs SDXL)."
embeds_scaling:
type: string
default: "V only"
enum:
- "V only"
- "K+V"
- "K+V w/ C penalty"
- "K+mean(V) w/ C penalty"
description: "Embedding scaling method for IPAdapter."
combine_method:
type: string
default: "concat"
enum:
- "concat"
- "add"
- "subtract"
- "average"
- "norm average"
- "max"
- "min"
description: "Combination method for multiple reference images."
final_weight:
type: number
default: 1.0
minimum: 0.0
maximum: 2.0
description: "Global weight multiplier for IPAdapter conditioning (0.0 to 2.0)."
lora_strength:
type: number
default: 0.6
description: "LoRA weight strength for FaceID adapter variants."
required:
- image
controlnet:
chains: controlnet
display_name: "ControlNet Spatial Guidance"
description: "Applies structural and spatial conditioning (depth, pose, lineart, tile, scribble, canny) to guide output composition."
max_count: 5
usage_guideline: "Specify ControlNet type and series (must match requested model architecture e.g., SD1.5, SDXL, SD3.5, FLUX.1, Qwen-Image), pre-processed control image (Base64 Data URI or HTTP/HTTPS URL; system does NOT pre-process raw RGB images), and guidance strength."
parameters_schema:
type: object
properties:
type:
type: string
description: "ControlNet conditioning type (must match model architecture)."
series:
type: string
description: "ControlNet model series (must match model architecture)."
image:
type: string
description: "Pre-processed control image (e.g., Depth, Canny, Pose, Lineart map) encoded as Base64 Data URI or HTTP/HTTPS URL. Note: System does NOT automatically pre-process raw RGB images."
strength:
type: number
default: 1.0
minimum: 0.0
maximum: 2.0
description: "Control influence strength (0.0 to 2.0)."
required:
- type
- series
- image
conditioning:
chains: conditioning
display_name: "Regional Conditioning / Area Prompt"
description: "Defines rectangular areas (X, Y, Width, Height) and assigns specific text prompts and conditioning strengths to them."
max_count: 10
usage_guideline: "Define rectangular spatial areas (X, Y, width, height) and assign specific prompts and strengths to them. Supports up to 10 area prompts."
parameters_schema:
type: object
properties:
prompt:
type: string
description: "Text prompt for this specific rectangular area."
x:
type: integer
default: 0
description: "Top-left X coordinate of the rectangular area."
y:
type: integer
default: 0
description: "Top-left Y coordinate of the rectangular area."
width:
type: integer
default: 512
description: "Width of the rectangular area."
height:
type: integer
default: 512
description: "Height of the rectangular area."
strength:
type: number
default: 1.0
minimum: 0.1
maximum: 2.0
description: "Conditioning strength for this area (0.1 to 2.0)."
required:
- prompt
vae:
chains: vae
display_name: "Custom VAE Loader"
description: "Overrides default VAE model used for latent space encoding and final image decoding."
max_count: 1
usage_guideline: "Specify source ('Civitai', 'Hugging Face', or 'Upload File'), then provide the vae_value (Civitai Version ID, HF file path, or uploaded filename)."
parameters_schema:
type: object
properties:
source:
type: string
enum: ["Civitai", "Hugging Face", "Upload File"]
description: "Download/upload source for the VAE model. Use 'Civitai' to download by Version ID, 'Hugging Face' to download by repo path, or 'Upload File' for uploaded file."
vae_value:
type: string
description: "For Civitai: the Version ID (e.g., '456' from civitai.com/models/123?modelVersionId=456). For Hugging Face: repo_id/filename.extension or repo_id/folder_path/filename.extension (e.g., 'madebyollin/sdxl-vae-fp16-fix/sdxl_vae.safetensors'). For Upload File: uploaded file name."
required:
- source
- vae_value
pid:
chains: pid
display_name: "PiD High-Resolution Refinement"
description: "Progressive Detail (PiD) upscale injector for fine detail enhancement and resolution upscaling."
max_count: 1
usage_guideline: "Enables PiD detail refinement pipeline using a boolean switch ('enabled': true/false)."
parameters_schema:
type: object
properties:
enabled:
type: boolean
default: true
description: "Enable or disable PiD High-Resolution Refinement."
required:
- enabled
flux1_style:
chains: style
display_name: "FLUX.1 Style Reference"
description: "Applies artistic style conditioning from reference images onto FLUX.1 model generated outputs."
max_count: 5
usage_guideline: "Supply style reference image(s) (up to 5) encoded as Base64 Data URI or HTTP/HTTPS URL and optional strength."
parameters_schema:
type: object
properties:
image:
type: string
description: "Style reference image encoded as Base64 Data URI or HTTP/HTTPS URL."
strength:
type: number
default: 1.0
minimum: 0.0
maximum: 2.0
description: "Style influence strength (0.0 to 2.0)."
required:
- image
reference_edit:
chains: reference_latent
display_name: "Reference Edit"
description: "For multimodal models, this feature enables powerful editing and combining capabilities. In txt2img mode, adding a single reference image performs an Image Edit, while adding multiple images performs an Image Combine."
max_count: 10
usage_guideline: "Supply reference image(s) encoded as Base64 Data URI or HTTP/HTTPS URL. Passing a single reference image performs an Image Edit, while passing multiple images (up to 10) performs an Image Combine."
parameters_schema:
type: object
properties:
image:
type: string
description: "Reference image encoded as Base64 Data URI or HTTP/HTTPS URL."
required:
- image
mage_flow_reference_edit:
chains: reference_image
display_name: "Reference Image Edit"
description: "For multimodal models, this feature enables reference image editing and combining capabilities. In txt2img mode, adding a single reference image performs an Image Edit, while adding multiple images performs an Image Combine."
max_count: 10
usage_guideline: "Supply reference image as Base64 Data URI or HTTP/HTTPS URL."
parameters_schema:
type: object
properties:
image:
type: string
description: "Reference image encoded as Base64 Data URI or HTTP/HTTPS URL."
required:
- image
fibo2_reference_edit:
chains: reference_image
display_name: "Fibo-2.0 Reference Edit"
description: "For Fibo 2.0 models, this feature enables reference image editing and combining capabilities. In txt2img mode, adding a single reference image performs an Image Edit, while adding multiple images performs an Image Combine."
max_count: 10
usage_guideline: "Supply reference image as Base64 Data URI or HTTP/HTTPS URL."
parameters_schema:
type: object
properties:
image:
type: string
description: "Reference image encoded as Base64 Data URI or HTTP/HTTPS URL."
required:
- image
minimax_h3_image_reference_edit:
chains: reference_image
display_name: "MiniMax-H3-Image Reference Edit"
description: "For MiniMax-H3-Image, this feature enables reference image editing and combining capabilities. In txt2img mode, adding a single reference image performs an Image Edit, while adding multiple images performs an Image Combine. The prompt refers to them as <Picture 1>, <Picture 2>, ..."
max_count: 8
usage_guideline: "Supply reference image(s) encoded as Base64 Data URI or HTTP/HTTPS URL (up to 8). Refer to them in the prompt as <Picture 1>, <Picture 2>, ...; <Picture 1> is the picture to edit."
parameters_schema:
type: object
properties:
image:
type: string
description: "Reference image encoded as Base64 Data URI or HTTP/HTTPS URL."
required:
- image
p_h3_image_reference_edit:
chains: reference_image
display_name: "P-H3-Image Reference Edit"
description: "For P-H3-Image, this feature enables reference image editing and combining capabilities. In txt2img mode, adding a single reference image performs an Image Edit, while adding multiple images performs an Image Combine. The prompt refers to them as <Picture 1>, <Picture 2>, ..."
max_count: 8
usage_guideline: "Supply reference image(s) encoded as Base64 Data URI or HTTP/HTTPS URL (up to 8). Refer to them in the prompt as <Picture 1>, <Picture 2>, ...; <Picture 1> is the picture to edit. Note: steps and cfg are fixed by the model (8-step ManualSigmas grid, no CFG)."
parameters_schema:
type: object
properties:
image:
type: string
description: "Reference image encoded as Base64 Data URI or HTTP/HTTPS URL."
required:
- image
krea2_identity_edit:
chains: krea2_identity_edit
display_name: "KREA2 Identity Edit"
description: "Processed using the lbouaraba/comfyui-krea2edit node. (Krea-2-Turbo recommended, Krea-2-Raw need set ZeroGPU Duration (s) to 120 ) In txt2img mode, adding a single reference image performs an Image Edit, while adding multiple images performs an Image Combine."
max_count: 2
usage_guideline: "Supply reference image as Base64 Data URI or HTTP/HTTPS URL."
parameters_schema:
type: object
properties:
image:
type: string
description: "Reference image encoded as Base64 Data URI or HTTP/HTTPS URL."
required:
- image
krea2_style_reference:
chains: krea2_style_reference
display_name: "KREA2 Style Reference"
description: "(Krea-2-Turbo recommended) Add style reference images to perform style reference editing."
max_count: 3
usage_guideline: "Supply style reference image as Base64 Data URI or HTTP/HTTPS URL."
parameters_schema:
type: object
properties:
image:
type: string
description: "Style reference image encoded as Base64 Data URI or HTTP/HTTPS URL."
required:
- image
joyai_reference_edit:
chains: joyai_image
display_name: "JoyAI Reference Edit"
description: " (JoyAI-Image-Edit recommended) For multimodal models, this feature enables powerful editing and combining capabilities. In txt2img mode, adding a single reference image performs an Image Edit (JoyAI-Image-Edit recommended), while adding multiple images performs an Image Combine (JoyAI-Image-Edit-Plus recommended with ZeroGPU Duration (s) set to 120)."
max_count: 2
usage_guideline: "Supply JoyAI reference image as Base64 Data URI or HTTP/HTTPS URL."
parameters_schema:
type: object
properties:
image:
type: string
description: "Input reference image encoded as Base64 Data URI or HTTP/HTTPS URL."
required:
- image
qwen_image_edit:
chains: qwen_image_edit
display_name: "Qwen-Image Edit"
description: " (lightx2v/Qwen-Image-Edit-2511-Lightning recommended) In txt2img mode, adding a single reference image performs an Image Edit, while adding multiple images performs an Image Combine."
max_count: 3
usage_guideline: "Supply reference image(s) (up to 3) encoded as Base64 Data URI or HTTP/HTTPS URL. Passing a single reference image performs an Image Edit, while passing multiple images performs an Image Combine."
parameters_schema:
type: object
properties:
image:
type: string
description: "Reference image encoded as Base64 Data URI or HTTP/HTTPS URL."
required:
- image
hidream_o1_smoothing:
chains: hidream_o1_smoothing
display_name: "HiDream O1 Smoothing Injector"
description: "HiDream O1 detail smoothing and artifact reduction injector."
max_count: 1
usage_guideline: "Configures smoothing factor for HiDream models."
parameters_schema:
type: object
properties:
factor:
type: number
default: 0.5
description: "Smoothing intensity (0.0 to 1.0)."
required: []
hidream_o1_reference:
chains: hidream_o1_reference
display_name: "HiDream-O1 Reference Edit"
description: " (HiDream-O1-Image-Dev recommended with resolution set to 4.0MP, e.g., 2048x2048) For HiDream-O1 models, this feature enables reference image editing and combining capabilities. In txt2img mode, adding a single reference image performs an Image Edit, while adding multiple images performs an Image Combine."
max_count: 9
usage_guideline: "Supply reference image(s) (up to 9) encoded as Base64 Data URI or HTTP/HTTPS URL. Passing a single reference image performs an Image Edit, while passing multiple images performs an Image Combine."
parameters_schema:
type: object
properties:
image:
type: string
description: "Reference image encoded as Base64 Data URI or HTTP/HTTPS URL."
required:
- image
sensenova_reference:
chains: sensenova_reference
display_name: "SenseNova Reference Edit"
description: "For SenseNova models, this feature enables reference image editing and combining capabilities. In txt2img mode, adding a single reference image performs an Image Edit, while adding multiple images performs an Image Combine."
max_count: 10
usage_guideline: "Supply reference image(s) (up to 10) encoded as Base64 Data URI or HTTP/HTTPS URL. Passing a single reference image performs an Image Edit, while passing multiple images performs an Image Combine."
parameters_schema:
type: object
properties:
image:
type: string
description: "Reference image encoded as Base64 Data URI or HTTP/HTTPS URL."
required:
- image
flux1_ipadapter:
chains: flux1_ipadapter
display_name: "Flux1 IP-Adapter"
description: "FLUX.1 model-specific IP-Adapter implementation."
max_count: 5
usage_guideline: "Supply reference image (Base64 Data URI or HTTP/HTTPS URL), optional weight (default 1.0), start_at (default 0.0), and end_at (default 1.0). Up to 5 images supported."
parameters_schema:
type: object
properties:
image:
type: string
description: "Reference image encoded as Base64 Data URI or HTTP/HTTPS URL."
weight:
type: number
default: 1.0
description: "Influence weight of the image prompt (0.0 to 2.0)."
start_at:
type: number
default: 0.0
description: "Start step percentage for IP-Adapter application (0.0 to 1.0)."
end_at:
type: number
default: 1.0
description: "End step percentage for IP-Adapter application (0.0 to 1.0)."
start_percent:
type: number
default: 0.0
description: "Alias for start_at."
end_percent:
type: number
default: 1.0
description: "Alias for end_at."
required:
- image
sd3_ipadapter:
chains: sd3_ipadapter
display_name: "SD3 IP-Adapter"
description: "SD3/SD3.5 model-specific IP-Adapter implementation."
max_count: 5
usage_guideline: "Supply reference image (Base64 Data URI or HTTP/HTTPS URL), optional weight (default 1.0), start_at (default 0.0), and end_at (default 1.0). Up to 5 images supported."
parameters_schema:
type: object
properties:
image:
type: string
description: "Reference image encoded as Base64 Data URI or HTTP/HTTPS URL."
weight:
type: number
default: 1.0
description: "Influence weight of the image prompt (0.0 to 2.0)."
start_at:
type: number
default: 0.0
description: "Start step percentage for IP-Adapter application (0.0 to 1.0)."
end_at:
type: number
default: 1.0
description: "End step percentage for IP-Adapter application (0.0 to 1.0)."
start_percent:
type: number
default: 0.0
description: "Alias for start_at."
end_percent:
type: number
default: 1.0
description: "Alias for end_at."
required:
- image
embedding:
chains: embedding
display_name: "Textual Inversion Embedding Injector"
description: "Downloads Textual Inversion embedding files from Civitai, Hugging Face, or uploads to the server. Note: This feature ONLY handles file downloading/preparation. To activate the embedding, manually add 'embedding:<filename>' (e.g. 'embedding:civitai/456' for Civitai ID 456, 'embedding:huggingface/lazypos' for Hugging Face, or 'embedding:upload_file/filename' for uploaded files) into your prompt or negative_prompt."
max_count: 5
usage_guideline: "Specify source ('Civitai', 'Hugging Face', or 'Upload File'), and embedding_value (Civitai Version ID, HF repo file path, or uploaded filename). The file is stored on server; manually enter 'embedding:<filename>' in prompt or negative_prompt to activate."
parameters_schema:
type: object
properties:
source:
type: string
enum:
- "Civitai"
- "Hugging Face"
- "Upload File"
description: "Download source for the Textual Inversion embedding file. Use 'Civitai' to download by Version ID, 'Hugging Face' to download by repo file path, or 'Upload File' for uploaded files."
embedding_value:
type: string
description: "For Civitai: the Version ID (e.g., '456' from civitai.com/models/123?modelVersionId=456, saved as 'civitai/456.safetensors'). For Hugging Face: repo_id/filename.extension (e.g., 'ilikebigturtles/lazypos/lazypos.safetensors', saved as 'huggingface/lazypos.safetensors'). For Upload File: uploaded filename (saved as 'upload_file/filename.safetensors'). Manually reference embedding:<filename> in prompt or negative_prompt."
required:
- source
- embedding_value
prompt_enhancer:
chains: prompt_enhancer
display_name: "Prompt Enhancer Settings"
description: "Enhance or upsample prompts with any registered prompt-enhancer model (Qwen-Image-2.1 PE T2I/I2I, Qwen3.8-27B, Qwen3.5-9B, Qwen3-VL-4B-Instruct, FLUX.3 Prompt Upsampler, P-H3-Image Prompt Upsampler), optional reasoning CoT thinking mode, and optional system prompt. The clip and system_prompt registries are independent, so they are freely mixable across architectures."
max_count: 1
usage_guideline: "Set enable to true to activate prompt enhancement. Pick model from the clip registry and system_prompt from the system_prompt registry in yaml/prompt_enhancer.yaml; both lists are global, so any architecture may select any clip paired with any system prompt. Defaults for each framework come from the architecture section and reset when the base model changes. architecture[framework].clip_type is the framework's own encoder and applies to whatever clip is selected, so a prompt upsampler built for one vendor still runs under another framework's encoder. Nothing is auto-detected from images, so an explicit system prompt always wins. Enable thinking for reasoning CoT (reasoning tags are stripped from the result), and set max_length for generation length."
parameters_schema:
type: object
properties:
enable:
type: boolean
default: false
description: "Enable prompt enhancement."
thinking:
type: boolean
default: false
description: "Enable thinking (reasoning CoT). Thought text inside <think> tags is automatically stripped before the prompt reaches downstream text encoding."
model:
type: string
description: "Prompt Enhancer Model name from the clip registry in yaml/prompt_enhancer.yaml (e.g. Qwen-Image-2.1-Prompt-Enhancer-T2I, Qwen3.8-27B, Qwen3.5-9B, Qwen3-VL-4B-Instruct, Qwen3.5-9B-FLUX.3-Prompt-Upsampler, PrunaAI/P-H3-Image-Prompt-Upsampler). Any model may select any registered clip regardless of architecture. The CLIPLoader type is the framework's, not the checkpoint's: architecture[framework].clip_type applies to whatever clip is selected (same Qwen3.5-9B weights load as `minimax` for MiniMax-H3-Image and `moss_image` for MOSS-Image), so a Qwen-Image-2.1 upsampler still runs as `moss_image` under MOSS-Image. A clip entry may also declare its own clip_type for calls without framework context. Defaults to the architecture entry for the active framework."
system_prompt:
type: string
description: "System Prompt name from the system_prompt registry in yaml/prompt_enhancer.yaml (e.g. none, Ming-Image-PE-Template, FLUX.3-T2I-PE-Template, FLUX.3-I2I-PE-Template, FLUX.3-T2I-Bbox-PE-Template, FLUX.3-I2I-Bbox-PE-Template). Independent of the selected clip and never inferred from images; an empty prompt means no system prompt."
max_length:
type: integer
default: 4096
description: "Maximum generation length for prompt enhancement (default 4096)."
moss_image_control:
chains: moss_image_control
display_name: "MOSS-Image Color/Depth Control"
description: "Controls MOSS-Image generation using a color region map (object/region spatial layout) and/or a depth map (3D spatial depth, near is white) via Comfy-Org/MOSS-Image color/depth control LoRA and MOSSImageControl node."
max_count: 1
usage_guideline: "Supply color_map (RGB region ID map) and/or depth_map (near=white grayscale map) as Base64 Data URI or HTTP/HTTPS URL. Set strength (default 1.0) and timestep range (start_percent 0.0, end_percent 0.5)."
parameters_schema:
type: object
properties:
color_map:
type: string
description: "Color region map image encoded as Base64 Data URI or HTTP/HTTPS URL. Each region uses a flat RGB color as a region ID."
depth_map:
type: string
description: "Depth map image encoded as Base64 Data URI or HTTP/HTTPS URL (near is white, far is dark)."
strength:
type: number
default: 1.0
minimum: 0.0
maximum: 2.0
description: "Control influence strength multiplier (default 1.0)."
start_percent:
type: number
default: 0.0
minimum: 0.0
maximum: 1.0
description: "Sampling percentage at which control begins (default 0.0)."
end_percent:
type: number
default: 0.5
minimum: 0.0
maximum: 1.0
description: "Sampling percentage at which control ends (default 0.5, use 0.5 for 3D mannequin)."
moss_image_edit:
chains: moss_image_edit
display_name: "MOSS-Image Edit"
description: "Edits images or generates variants using MOSS-Image Edit LoRA (officially packaged by Comfy-Org as moss_image_edit_lora_rank_64_bf16.safetensors) and native TextEncodeMOSSImageEdit node."
max_count: 10
usage_guideline: "Upload 1-10 reference images and specify prompt. Automatically uses the official MOSS-Image Edit LoRA layout."
parameters_schema:
type: object
properties:
images:
type: array
items:
type: string
description: "List of 1 to 10 reference images encoded as Base64 Data URI, HTTP/HTTPS URL, or local file path."