Venice AI Nodes (unofficial)

ComfyUI Venice AI Nodes (unofficial)

ComfyUI nodes for the Venice AI API: every Venice image, image edit and video model in one image node, one edit node and one video node, with the price and the censored/uncensored flag right in the model list. Plus three save/load helper nodes.

Unofficial. This project is not affiliated with, endorsed or sponsored by Venice.ai. You need your own Venice API key; every generation is billed to your Venice account. 🇩🇪 Deutsche Anleitung: README.de.md

Nodes

| Node | What it does | |---|---| | Venice AI · Image | Text to image with every Venice image model | | Venice AI · Image Edit | Edit one image, or build a new one from several reference images (up to 10, depending on the model) | | Venice AI · Video | Text / image / reference / video to video with every Venice video model | | Venice AI · Video Save Plus 🎬 | Save MP4 + audio, workflow PNG/JSON and prompt .txt to any folder (from ComfyUI-CustomVideoSave v2.1.0) | | Venice AI · Save Image + Prompt TXT | Save images with the prompt as .txt, reveal / copy / delete buttons (from ComfyUI_SaveImagePlusPrompt v1.4.1) | | Venice AI · Load & Resize Image + Filename | Load, resize, filename output, folder series, rectangle/freehand mask (from ComfyUI-LoadResizeImageWithFilenameV02 v2.0.0) |

The three bundled nodes have their own node IDs, names and server routes, so they run side by side with the original packs. All nodes are in the node menu under Venice AI.

Model overview and example workflows

• docs/MODELS.md: what every Venice image, edit and video model can do (lengths, resolutions, aspect ratios, audio, input images, price, strengths, uncensored, models ending soon). German: docs/MODELS.de.md. Regenerate with python tools/generate_model_docs.py.
• example_workflows/: six ready workflows with notes (text → image, image edit with two references, image → video, text → image → video, reference images → video, video → video). In ComfyUI: Workflow → Browse Templates → this pack, or drag a .json onto the canvas.

Installation

• ComfyUI Manager: search for Venice AI Nodes (unofficial) and install, then restart ComfyUI.
• Manually: clone into ComfyUI/custom_nodes and restart:

  cd ComfyUI/custom_nodes
  git clone https://github.com/Slartibart23/ComfyUI-Venice-AI-Nodes
  

No extra Python packages are needed.

API key

Create an API key at venice.ai (Settings → API). The nodes look for it in this order:

1. the api_key field of the node: not recommended, it is saved inside every workflow and every PNG/MP4 that embeds the workflow,
2. the environment variable VENICE_API_KEY,
3. the file Venice_AI_API.txt in your ComfyUI folder (next to main.py; the portable root one level above works too). Only the key, one line. VENICE_API_KEY=... is accepted as well.

Model lists

Each entry shows name, type, price, censorship flag, privacy and the model ID:

★ MiniMax H3 Max · I2V · $0.31/5s · UNCENSORED · private · minimax-h3-max-image-to-video
Seedance 2.5 · I2V · $1.44/5s · censored · seedance-2-5-image-to-video-basic

• ★ favourites come first: put model IDs (one per line) into Venice_AI_favourites.txt in your ComfyUI folder, see docs/Venice_AI_favourites.example.txt.
• T2V / I2V / R2V / V2V = text / image / reference / video to video.
• Price: video at about 5 s and 720p (the actual price is quoted before every run), image per picture at the smallest size (+ = larger sizes cost more).
• UNCENSORED / censored is how Venice flags the model. "censored" only means "not flagged": models and providers may still refuse prompts.
• private = Venice states it does not pass your data on; other models are "anonymized" (sent to the model provider without your identity).
• Filters above the model list narrow it down: type (text / image / reference images / video to video) and takes audio input in the video node, combines several images in the edit node, uncensored only in all three. They are saved with the workflow but never sent with the prompt.
• Below the settings each node shows a fact sheet of the selected model (length, resolution, audio, inputs, price, strengths, removal date); after a run the run info appears above it.
• ENDS YYYY-MM-DD = Venice removes the model on that date (see the replacement in docs/MODELS.md).
• The list is shipped with the pack and refreshed in the background: models once a day, video prices once a week (no API key needed). Reload the browser tab to see changes. Workflows keep working when a price changes, because the model ID at the end of the entry counts.

Venice AI · Video

• duration / resolution / aspect_ratio only offer what the selected model can do, up to its maximum length (e.g. Seedance 2.5: 4–30 s, 480p–1080p; Wan 3.0: 2–30 s in 5 s steps). Values a model does not offer are replaced by the nearest one and noted in info.
• Inputs: image (start image, I2V), end_image, reference_images (R2V, image batch, up to 30), video (V2V), audio (models with audio input).
• preview: cheap test run with the lowest resolution and a short duration; models that cannot go below 1080p use their cheaper sibling (Wan 3.0 Pro → Wan 3.0). Venice has no draft mode, so the preview is not the same video as the full run, it is for testing prompt and motion.
• price_only: only asks Venice for the price, nothing is generated or billed.
• max_price_usd: the run stops before anything is billed if the quoted price is higher.
• seedance_consent: Seedance may require you to confirm the provider terms, that you have the legal right to the input media (faces!) and that it may be screened. Off by default.
• keep_raw_copy: the downloaded MP4 is kept in output/venice_ai (paid results are never lost).
• Outputs: video (VIDEO), video_path, last_frame (to chain the next clip), prompt, info.
• Venice video has no seed. Change variation (or set it to randomize) for a new take.

Typical wiring without re-encoding:

[Venice AI · Video] --video_path--> [Venice AI · Video Save Plus] source_video_path
                    --prompt------>                               prompt_text

Venice AI · Image

• aspect_ratio / resolution / quality follow the selected model (e.g. Grok Imagine: 1:1 … 2:3, 1K/2K, low/medium). Models with free sizes (e.g. Lustify) use width/height: pick a ratio and the node computes the size (max. 1280 px per side), or custom for your own width/height.
• safe_mode is off by default (on = Venice blurs adult content). variants 1–4 images per run.
• Outputs: images, prompt (connect to Save Image + Prompt TXT to store it as .txt), info.

Venice AI · Image Edit

• image_1 is the image to edit; image_2 … image_6 and extra_images (batch) add references, e.g. "the woman from image 1 sits on the sofa from image 2". Images may have different sizes.
• The model list shows how many input images each model combines (max 6 img, 1 img = only one, multi img = several, no limit stated by Venice). More images than allowed stop the run before anything is billed. Some models charge extra per additional input image (+ in the price).
• aspect_ratio auto = taken from the input image; resolution / quality follow the model.
• One image uses Venice's edit endpoint, several images (or a quality level) the multi-edit endpoint.
• Edits have no seed: change variation for a new result with the same settings.
• Outputs: images, prompt, info.

Legal notes

• Use your own key only. Venice does not allow sharing or reselling API access.
• You are responsible for what you generate. Do not use a real person's face, voice or likeness without their consent, and follow the Venice terms and the law of your country.
• Prices are shown for information; the amount Venice bills is authoritative.

License and credits

GPL-3.0, see LICENSE.

• Load & Resize Image + Filename is based on LoadAndResizeImage from ComfyUI-KJNodes by kijai (GPL-3.0); this is why the whole pack is GPL-3.0.
• Save Image + Prompt TXT was inspired by LayerUtility's SaveImagePlus from ComfyUI_LayerStyle.
• Venice and Venice.ai are names of their respective owner and are used here only to describe compatibility.