Nova Creative

AI images, videos & workflows

Create AI images, short videos, and creative variations from a brief or your reference media. Explore visual directions, animate product images, and refine the result in a reusable Nova Creative workflow.

What you can ask

Create or refine a saved Nova workflow from my brief and references. Show estimated usage; let me confirm in the Nova card.
Example path → ·

MCP connector

Server URL: https://app.creative.novastorm.ai/mcp
Login: OAuth
Protocol: MCP 2026-07-28
Instructions: Nova Creative creates images, videos, talking avatars, audio and 3D from selected briefs and references, keeping user-owned editable workflows. Call before rename or patch and pass its updatedAt as expected_updated_at. Use only for new graphs. For Save as or a full copy use : it preserves current node results without copying history, uploading or regenerating assets. Default to one output per requested item; do not invent alternative images, first frames or batches without an explicit request. A wire carries the entire result stream. To pass one selected image to Seedance or another reference input, insert listSelector (mode=index, zero-based index); choosing a preview thumbnail alone does not filter a stream. Call when diagnosing a workflow quality problem or applying a known repair. Output nodes are optional display labels; add one only when requested and preserve existing output and templateOutput nodes. Use for generated media and with include_results=true when saved node values are needed. templateOutput has separate required-template semantics. Default text/copy LLM nodes to provider="openrouter", model="openai/gpt-6.1-sol" unless the user selects another model. Prefer Remotion/Motion Ad (remotionVideo) when the user explicitly requests a Remotion video, or when a video presentation workflow needs animated slides, exact on-screen text, feature lists, statistics or branded transitions around existing images/clips. Add it as the composition stage after the relevant content and media, keeping the existing workflow; its video output can be a final deliverable or a segment for concatVideo. A generic video/ad request alone is not a reason to add Remotion: use videoGenerator for new realistic scenes or natural subject motion, and existing editing nodes for trimming, merging or captions. Remotion outputs a short MP4 from fixed scene templates, not an editable PPTX or arbitrary React code; do not add a video stage to a static presentation unless requested. For Remotion/Motion Ad (remotionVideo), read with query=remotionVideo and write the scene plan yourself using this conversation's model. Set mode="plan" and put the nova.remotion-ad/v1 JSON string in data.plan or a connected text node on plan. Public MCP rejects auto/script planning at preflight; template is also available without an LLM. Do not add a runLlm node to write the plan. Connect image/video outputs to media1-media3, an image logo to logo, audio to voice/music and brand text to brand; scenes reference media slot names, never URLs. Nova does not call or charge another planning model in Plan mode; Remotion rendering and any upstream generators retain their reviewed costs. Use before choosing a provider. Call to reuse references already uploaded to Nova; never require them to be reattached to chat. For selected local references use with base64 bytes (16 MB), or (64 MB) using the current host's exposed file schema; a host may upload a selected local file, while direct MCP accepts a selected public HTTPS attachment. Image, video, audio and 3D are supported. The remote server itself never reads client disk paths. For requests to replace, change or remove an object, text, clothing, logo or background in an image, call to show the selectable image and layers. Pass the requested edit as instruction. Opening does not generate; the user chooses layers and confirms analysis or generation in the editor. Preflight the saved version and present the scope and estimated usage. After the user's explicit approval in chat, call with confirmed=true and approved_estimated_credits equal to the reviewed estimatedCredits. Existing approval for unchanged scope and cost is sufficient; do not ask for another confirmation or require a card. Reuse request_id for retries. The optional Nova card still uses its private ; never expose or fabricate its confirmation token. Follow the durable run for results; refresh expired download links without regenerating. Use incremental edits on the same saved project for feedback. Check and the tools actually exposed by the current host before relying on GPU features; do not promise or work around a missing tool. For Avatar/Comfy workflows use , , and when available, then set an explicit gpuNodeId and upload the portrait/voice. A workflow never implicitly rents a GPU or installs weights. GPU rental time, including installation time, is billed separately from provider-call estimates. GPU rental is unavailable through this plugin; do not offer to rent a server or work around the unavailable rental tool. Use an existing accessible GPU. Model/runtime installs need workflows.generate and installation confirmation. permanently deletes the exact rented server and its disk; it needs workflows.write and explicit deletion confirmation, then must confirm provider deletion before claiming billing has stopped. Settings/model removal also need workflows.write and exact target confirmation. Personal tokens do not automatically gain paid-action permission. Whole-workflow deletion and Meta campaign publication are unavailable. Do not offer upgrades, checkout or credit purchases through this plugin.

Tools

Reads data19

Changes data11

Deletes or can't be undone6

Skills

1 in the ChatGPT listing

Screenshots

MCP connector

Server URL

https://app.creative.novastorm.ai/mcp

OAuth

Protocol

MCP 2026-07-28

Instructions

Nova Creative creates images, videos, talking avatars, audio and 3D from selected briefs and references, keeping user-owned editable workflows. Call before rename or patch and pass its updatedAt as expected_updated_at. Use only for new graphs. For Save as or a full copy use : it preserves current node results without copying history, uploading or regenerating assets. Default to one output per requested item; do not invent alternative images, first frames or batches without an explicit request. A wire carries the entire result stream. To pass one selected image to Seedance or another reference input, insert listSelector (mode=index, zero-based index); choosing a preview thumbnail alone does not filter a stream. Call when diagnosing a workflow quality problem or applying a known repair. Output nodes are optional display labels; add one only when requested and preserve existing output and templateOutput nodes. Use for generated media and with include_results=true when saved node values are needed. templateOutput has separate required-template semantics. Default text/copy LLM nodes to provider="openrouter", model="openai/gpt-6.1-sol" unless the user selects another model. Prefer Remotion/Motion Ad (remotionVideo) when the user explicitly requests a Remotion video, or when a video presentation workflow needs animated slides, exact on-screen text, feature lists, statistics or branded transitions around existing images/clips. Add it as the composition stage after the relevant content and media, keeping the existing workflow; its video output can be a final deliverable or a segment for concatVideo. A generic video/ad request alone is not a reason to add Remotion: use videoGenerator for new realistic scenes or natural subject motion, and existing editing nodes for trimming, merging or captions. Remotion outputs a short MP4 from fixed scene templates, not an editable PPTX or arbitrary React code; do not add a video stage to a static presentation unless requested. For Remotion/Motion Ad (remotionVideo), read with query=remotionVideo and write the scene plan yourself using this conversation's model. Set mode="plan" and put the nova.remotion-ad/v1 JSON string in data.plan or a connected text node on plan. Public MCP rejects auto/script planning at preflight; template is also available without an LLM. Do not add a runLlm node to write the plan. Connect image/video outputs to media1-media3, an image logo to logo, audio to voice/music and brand text to brand; scenes reference media slot names, never URLs. Nova does not call or charge another planning model in Plan mode; Remotion rendering and any upstream generators retain their reviewed costs. Use before choosing a provider. Call to reuse references already uploaded to Nova; never require them to be reattached to chat. For selected local references use with base64 bytes (16 MB), or (64 MB) using the current host's exposed file schema; a host may upload a selected local file, while direct MCP accepts a selected public HTTPS attachment. Image, video, audio and 3D are supported. The remote server itself never reads client disk paths. For requests to replace, change or remove an object, text, clothing, logo or background in an image, call to show the selectable image and layers. Pass the requested edit as instruction. Opening does not generate; the user chooses layers and confirms analysis or generation in the editor. Preflight the saved version and present the scope and estimated usage. After the user's explicit approval in chat, call with confirmed=true and approved_estimated_credits equal to the reviewed estimatedCredits. Existing approval for unchanged scope and cost is sufficient; do not ask for another confirmation or require a card. Reuse request_id for retries. The optional Nova card still uses its private ; never expose or fabricate its confirmation token. Follow the durable run for results; refresh expired download links without regenerating. Use incremental edits on the same saved project for feedback. Check and the tools actually exposed by the current host before relying on GPU features; do not promise or work around a missing tool. For Avatar/Comfy workflows use , , and when available, then set an explicit gpuNodeId and upload the portrait/voice. A workflow never implicitly rents a GPU or installs weights. GPU rental time, including installation time, is billed separately from provider-call estimates. GPU rental is unavailable through this plugin; do not offer to rent a server or work around the unavailable rental tool. Use an existing accessible GPU. Model/runtime installs need workflows.generate and installation confirmation. permanently deletes the exact rented server and its disk; it needs workflows.write and explicit deletion confirmation, then must confirm provider deletion before claiming billing has stopped. Settings/model removal also need workflows.write and exact target confirmation. Personal tokens do not automatically gain paid-action permission. Whole-workflow deletion and Meta campaign publication are unavailable. Do not offer upgrades, checkout or credit purchases through this plugin.

Tools