Comfyui

ComfyUI Workflows in OpenMontage

Use this skill before calling comfyui_image, comfyui_video, or comfyui_music, and when converting a community ComfyUI workflow into an OpenMontage tool call.

Server Contract

• ComfyUI must be running before the tool can generate. The default server is http://localhost:8188; override it with COMFYUI_SERVER_URL.
• Running separate ComfyUI instances per capability (different GPU, different model set)? COMFYUI_IMAGE_SERVER_URL / COMFYUI_VIDEO_SERVER_URL / COMFYUI_MUSIC_SERVER_URL each override COMFYUI_SERVER_URL for that one tool only. Optional -- a single-server setup needs none of these.
• Health and hardware status come from GET /system_stats.
• Jobs are submitted to POST /prompt, completed outputs are read from GET /history/{prompt_id}, and artifact bytes are downloaded with GET /view.
• Long waits (video, music) prefer ComfyUI's websocket feed for immediate completion/error detection and transparently fall back to REST polling if websocket-client isn't installed. Either way, a timeout is recoverable: pass the error's prompt_id back in as resume_prompt_id to resume waiting on the same job instead of resubmitting it.
• Export workflows with ComfyUI's API-format JSON, not the UI layout format. If a downloaded workflow will not submit, re-export it from ComfyUI with API format enabled.

Partner Nodes are hosted

• gemini_omni_flash, seedance_2.5, and minimax_h3_api in comfyui_video are official ComfyUI Partner Nodes. They call hosted APIs and require network access, a logged-in Comfy account, and prepaid credits.
• Do not describe Partner Nodes as local, offline, or free merely because the graph runs in a local ComfyUI process.
• minimax_h3_local is a separate open-weight path. It requires the official MiniMax H3 workflow exported in API format, its output_node, and the model stack reported by the tool.

Choosing a Workflow

• Use bundled workflows when the requested operation matches and the local machine has the required models and VRAM.
• Use a custom workflow_json or workflow_path when the user needs a community recipe, a lower-VRAM model, a different style family, or custom nodes.
• For 8GB-12GB GPUs, prefer lower-footprint workflows such as Wan 2.1 1.3B, LTXV FP8 or quantized workflows, or Wan 2.2 GGUF/quantized community workflows. The bundled Wan 2.2 14B FP8 video workflows are a 16GB-class path, not a provider-wide floor.
• Do not promise that arbitrary custom workflows will fit a machine. The workflow, quantization, resolution, frame count, and offload settings determine the real resource envelope.

Output Node Contract

• Custom workflows must pass output_node.
• Pick the node that writes the artifact, usually SaveImage, SaveVideo, VHS_VideoCombine, or another terminal saver node.
• Pass the node ID as a string, for example "108". Do not pass the class name.
• If a workflow has multiple savers, choose the final deliverable node, not previews or intermediates.

Templated vs Fixed Nodes

• Identify templated nodes before execution: prompt text, seed, dimensions, frame count, source image, sampler settings, and output filename prefix.
• Fixed nodes are model loaders, VAEs, text encoders, LoRA loaders, schedulers, and graph wiring. Do not mutate those unless the workflow author intended that customization.
• For community workflows, inspect each loader node and note every required model or custom node before running. Missing models should be handled through the tool's structured missing_models payload when available.

Video workflows need a temporal latent

• A video workflow's empty-latent node must be a video latent -- EmptyHunyuanLatentVideo for Wan 2.2 t2v, Wan22ImageToVideoLatent or WanImageToVideo for the image-conditioned variants. The frame count goes in length; batch_size is how many separate clips to generate and stays at 1.
• EmptyLatentImage with batch_size set to the frame count is a trap: it asks for N unrelated images, and the resulting MP4 has the right frame count, duration and codec and passes ffprobe cleanly -- it just strobes. Check the latent node before running any unfamiliar video workflow.