ComfyUI custom nodes for the Agnes AI API — a free, cloud-based AI generation platform. Generate images, create videos, enhance prompts, and analyze visuals — all without a local GPU. Zero extra Python dependencies.
- 2026/09/18: Update ComfyUI-Agnes-AI to v1.4.0 ( updates.md )
- New (External Presets): Decoupled prompt presets in
presets/styles.jsonwith dynamic discovery and 8 production-grade creative assistants (Script Writing,Storyboard Creation,Image Analysis, etc.). - New (Native Media Connect): Direct ComfyUI
AUDIOandVIDEOnode inputs onAgnesVideo(automatically converted to base64 Data URIs). - New (Real-time Progress & Cancellation): Native 0–100% progress tracking bar during cloud video polling and immediate task interruption upon cancelling prompts.
- New (Multi-Key Failover): Memory-level multi-key rotation and 120s cooldown handling with redesigned Settings management.
- New (External Presets): Decoupled prompt presets in
- 2026/09/13: Update ComfyUI-Agnes-AI to v1.3.0 ( updates.md )
- New: Transparent node-level model dropdowns with real-time cloud synchronization, workflow migration engine (
agnes-compat.js), and configurable timeouts.
- New: Transparent node-level model dropdowns with real-time cloud synchronization, workflow migration engine (
- 2026/09/03: Update ComfyUI-Agnes-AI to v1.2.0 ( updates.md )
- New: Full support for
agnes-video-2.5andagnes-video-2.5-flashwith the modern OpenAI Videos-compatible API. - New: Upgraded image model support to
agnes-image-2.5-flash(1K / 2K / 3K / 4K tiers & multi-image composition). - Architecture: Re-architected with Model Adapter Pattern & Dynamic Registry for clean separation of concerns and automatic compatibility with future Agnes AI models.
- Fixed: Fixed video polling query parameters (
model_name) and duration (seconds) parameter mapping.
- New: Full support for
- 2026/07/29: Update ComfyUI-Agnes-AI to v1.1.0 ( updates.md )

Initial release with Agnes-AI Image, Video, Text, and Config nodes.
| Free | No token billing, no credit system — get a key and use it |
| No GPU needed | All computation runs on Agnes servers, your ComfyUI stays lightweight |
| One API, many models | Image gen, video gen, chat, vision — all through a single key |
| Multi-image compose | Up to 4 reference images for img2img composition |
| Video generation | Text-to-video, image-to-video, and keyframe interpolation |
| Smart text presets | Prompt enhancement, translation, art style extraction, image description |
Get a free API key: platform.agnes-ai.com
- Image Generation — Text2img / img2img with up to 4 reference images powered by
agnes-image-2.5-flash. Auto-detects mode based on input connections. 1K / 2K / 4K resolution with ratio matching. - Video Generation — Text-to-video, images-to-video, first-and-last-frame keyframe animation powered by
agnes-video-2.5-flash,agnes-video-2.5, andagnes-video-v2.0. Supports 720p / 960p / 2K, 3–18 seconds duration, configurable frame rate, negative prompt, and full frame/audio extraction. - Prompt Enhancement — 4 built-in presets: enhance, translate, extract art style, describe image. Custom system prompt override. Vision-based presets with image input.
- Settings Panel — Configure API key, select default models, and pull latest models directly from ComfyUI's built-in Settings Panel. Supports multiple API keys for load balancing.
Search Agnes-AI in ComfyUI-Manager and install.
cd ComfyUI/custom_nodes
git clone https://github.com/1038lab/ComfyUI-Agnes-AIRestart ComfyUI.
Download the latest release, extract to ComfyUI/custom_nodes/ComfyUI-Agnes-AI/, restart ComfyUI.
No pip install needed — zero additional dependencies.
Open ComfyUI Settings (⚙️ gear icon) → search "Agnes-AI" → enter your API key.
- Click "✨ Get The Latest Models" to fetch and refresh the available model list anytime
- Multiple keys can be comma-separated for round-robin load balancing
- Alternatively, set the
AGNES_API_KEYenvironment variable - Key is saved to
agnes_config.jsonand persists across restarts
Priority: env var (AGNES_API_KEY) > saved config (Settings Panel).
Generate images from text, or compose new images from up to 4 reference images. Auto-detects text2img (no image input) vs img2img (images connected). When images are connected without a prompt, defaults to merging them into one cohesive composition. Powered by agnes-image-2.5-flash with 1K / 2K / 4K size tiers and multiple aspect ratios.
Supports agnes-video-2.5-flash (default), agnes-video-2.5, and agnes-video-v2.0.
Three generation modes:
- Text To Video — generate video from text
- Images To Video — animate or reference from start frame
- First and Last frame — interpolate between two images
Inputs:
- prompt — Description of the video to generate.
- negative_prompt — (Optional) Describe what to avoid in the generated video.
- image — Start frame (for Images To Video / First and Last frame).
- end_frame — End frame (for First and Last frame mode).
- quality, aspect_ratio, duration, frame_rate, seed.
Outputs:
- video — The generated video path or video object.
- last_frame — Automatically extracted last frame of the video (
IMAGE). - frames — The full sequence of extracted video frames as a batched
IMAGEtensor (requires localffmpeg). - audio — Extracted audio track as a standard ComfyUI
AUDIOwaveform (requires localffmpeg).
Process prompts through 4 built-in presets:
| Preset | Needs Image | Use Case |
|---|---|---|
| Prompt Enhance | No | Expand brief prompts with vivid visual context |
| Translate to English | No | Translate prompts while preserving visual details |
| Extract Art Style from Image | Yes | Analyze artistic style from a reference image |
| Image Detailed Description | Yes | Generate detailed AI-ready prompts from images |
Custom system prompt override supported for all presets. Supports agnes-2.5-flash (default) and agnes-2.5-pro configured via the Settings Panel.
Coming soon — check the example_workflows directory.
- Agnes AI — Free API and model infrastructure. Get your key at platform.agnes-ai.com
- Created by AILab
GPL-3.0
If this custom node helps you, please ⭐ the repo!
