name: omniflash description: Generate short video clips (with synchronized audio) and still images via the Omni Flash API (https://omniflash.net/) — Google's Gemini Omni Flash family. Use when the user asks to create a video or image from a prompt with Omni Flash, generate a short clip with sound, render an AI image with the Gemini Omni Flash models, or mentions seedance-2 / gpt-image-2 / nano-banana-2 in the context of Omni Flash. Triggers on phrases like "Omni Flash", "omniflash", "Gemini Omni Flash", "generate a video with omni flash", "make a clip with gemini omni flash", "draw an image with omni flash".

Omni Flash

Call the Gemini Omni Flash REST API to generate short video clips (with synchronized audio) and still images using Google's Gemini Omni Flash family of models — seedance-2 (text/image → video with audio), gpt-image-2 and nano-banana-2 (text/image → image).

Get an API key first. Tell the user to sign in at https://omniflash.net/ and copy a sk-… token from the account page. Read it from OMNIFLASH_API_KEY so the key never appears in chat or code.

When to use

Trigger this skill whenever the user wants to:

• Generate a short AI video (up to ~10 s) from a text prompt and/or reference images, with synchronized audio.
• Generate an image from a text prompt (and optionally reference images).
• Inspect or poll an Omni Flash task they already submitted.
• Compare the three Gemini Omni Flash models or pick the right one.

Decision tree

1. What output does the user want?
   • Video (with audio) → model_id: "seedance-2".
   • Image → default to gpt-image-2; use nano-banana-2 if the user asks for a cheaper/faster image alternative.
2. Do they have reference images / clips?
   • Yes → pass image_urls: [url, ...].
   • No → omit the field (text-to-X).
3. Aspect ratio?
   • Default to 16:9 for video and 1:1 for images unless the user specifies otherwise (e.g. 9:16 portrait, 4:5).
4. Run it. Use the convenience run helper if available in the chosen SDK; otherwise create + poll until task_status is 3 (success) or 4 (failed).

Workflow

1. Confirm the user has set OMNIFLASH_API_KEY. If not, link them to https://omniflash.net/ to issue one.
2. Pick the SDK that matches their stack — see references/api.md for raw REST or references/sdk-snippets.md for ready-to-paste code in Python, Node/TS, Ruby, Go, PHP, Rust, Dart, and .NET.
3. Build the prompt with the user — Omni Flash prompts behave like other Gemini-family diffusion models: subject, style, lighting, camera, modifiers. For video, include motion and audio cues (e.g. "the kettle whistles as steam rises").
4. Submit, poll, then surface the resulting video_url, audio_url, or image_url.

Models at a glance

API surface (summary)

POST /api/v1/tasks/create with { model_id, prompt, image_urls?, aspect_ratio? } → { data: { task_id, request_id, credits } }. GET /api/v1/tasks/{task_id} → { data: { task_id, task_status, image_url?, video_url?, audio_url?, ... } }. task_status: 1=queued, 2=running, 3=success, 4=failed.

Auth header on every request: Authorization: Bearer sk-....

Reference

• Full endpoint reference: references/api.md
• Per-language snippets: references/sdk-snippets.md
• Gemini Omni Flash website (sign in, get key)
• Live API docs: https://omniflash.net/api-docs

Omni Flash — Claude Code Skill

A Claude Code skill that lets Claude generate short video clips (with synchronized audio) and still images via the Gemini Omni Flash REST API — Google's Gemini Omni Flash family of models: seedance-2 (text/image → video with audio), gpt-image-2 and nano-banana-2 (text/image → image).

Sign in at Gemini Omni Flash and create an sk-… API key on the account page. Export it as OMNIFLASH_API_KEY and Claude will pick it up automatically.

What this skill does

When triggered, Claude will:

1. Pick the right Gemini Omni Flash model (seedance-2 for video + audio, gpt-image-2 for high-fidelity images, nano-banana-2 for cheaper/faster image generation).
2. Help you sharpen the prompt — for video, add motion + audio cues (e.g. "the kettle whistles as steam rises").
3. Submit the job through any of the official Omni Flash SDKs (Python / Node / Ruby / Go / PHP / Rust / Dart / .NET) and poll until done.
4. Return the final video_url, audio_url, or image_url.

The full workflow and model matrix lives in SKILL.md and references/.

Install into Claude Code

Clone the repo into your Claude Code skills directory:

# User-level (applies to every project)
git clone https://github.com/beamtrace/omniflash-skill.git ~/.claude/skills/omniflash

# Or project-level (only the current project)
git clone https://github.com/beamtrace/omniflash-skill.git .claude/skills/omniflash

Restart Claude Code and ask something like "generate a 5-second clip of a kettle whistling with Omni Flash" — the skill should fire automatically.

How to trigger it

The skill's description field already matches the most common phrasings. Any of these will trigger it:

• "generate a video with omniflash"
• "make a clip with gemini omni flash"
• "draw an image using omni flash"
• "how do I call Gemini Omni Flash"
• mentioning any of seedance-2, gpt-image-2, nano-banana-2 in the Omni Flash context

Companion multi-language SDKs

Each SDK is a ~150-line thin wrapper that exposes the same three methods: create_task / get_task / run.

Links

• Website (sign up + get key): Gemini Omni Flash
• Live API reference: https://omniflash.net/api-docs

License

MIT