How to Turn an Image into Video with AI Assistants
To turn a still image into a short video, pass the image URL to a connected video skill and describe the motion. ChatGPT, Claude, and Grok can talk about camera moves. They do not, on their own, return an MP4 of your product photo. ToolRouter’s Generate Video skill is image-to-video on most clients. On the Claude connector, text-to-video is also available. Otherwise make or upload the first frame, then animate it.
This matches how people actually search (image to video in ChatGPT, Claude, Grok) and how the catalog works. If you wanted a native Sora-style prompt box, use that product. If you wanted the still you already have to move inside the assistant you already use, stay here.
Native limits versus a first-frame workflow
| Job | Assistant alone | Assistant + ToolRouter |
|---|---|---|
| Describe a camera move in words | Yes | Yes |
| Animate a JPEG/PNG/WebP you host | No catalog clip | Yes, image_to_video |
| Invent a video from text only in this catalog | Some clients have their own video products | No. Generate the still with Generate Image, then animate |
| Pick duration, resolution, audio | Client-specific | list_models first; cost and limits vary by model |
| Download an MP4 | No | Yes, as a catalog asset |
Generate Video requires image_url. Duration, resolution, and audio depend on the model. Call list_models rather than hard-coding names; the live matrix changes. Typical clips are a few seconds, not a two-minute ad.
This is a paid skill. Check the model list before you generate. Failed paid calls are not supposed to charge; free skills never charge.
Related how-tos if your client is Claude-only: Can Claude generate video? and How to make videos in Claude. Grok-only: Can Grok generate video?. This page is the job that is the same across clients: still in, clip out.
How to connect
The Model Context Protocol is the shared protocol. Each host still has its own switch.
- Claude: custom connector
https://api.toolrouter.com/mcp/anthropic. Connect Claude. Anthropic connectors. - ChatGPT: Developer mode app,
https://api.toolrouter.com/mcp. OpenAI Developer mode. Connect ChatGPT. - Grok (grok.com): custom MCP at grok.com/connectors. xAI Connectors. Connect Grok.
- Grok Bot: Settings → Plugins → Add. HTTPS only. How to add MCP to Grok Bot.
- Cursor: Cursor MCP and Connect Cursor.
Grok (grok.com) is not Grok Bot. Do not paste a plugin flow into grok.com, or the reverse.
The prompt to type
You already have a still:
Animate this image into a 5-second clip.
https://example.com/product.jpg
Slow camera pull-back, product stays sharp, no extra objects.
List models and estimated cost before you generate.You do not have a still:
Generate a first-frame product photo of a ceramic mug on a
wooden table, window light, then animate a slow orbit around it.
Confirm the model and cost before the video step.The second prompt should hit Generate Image then Generate Video. If the assistant skips the still, stop it. Image-to-video without a frame is how you get a random clip that does not match the product.
Motion prompts that behave: one camera move, or one subject action. “She turns and smiles.” “Steam rises from the cup.” Stacking five events in five seconds usually smears.
Give your AI superpowers. Copy one instruction and paste it into your AI.
Cost, time, and what will go wrong
Video is slower and dearer than a still. Budget 60–90 seconds for a run, longer if the provider queues. Ask for list_models so you see duration caps, audio support, and price before you spend.
Audio is optional and only on some models. Turning it on can roughly double cost. If you need a silent loop for a site hero, keep audio off.
Rights: you need permission to animate the still. A scraped competitor photo is a bad first frame. A shot you generated, or a product image you own, is the intended path.
This catalog does not turn a URL of a website into a screen recording, and it does not clone a TikTok by pasting a link. Those are different products.
When ToolRouter is not the answer
- You already have Sora, Veo, or Grok Imagine video in the client and the native clip is enough.
- You need a 60-second narrative with dialogue and cuts. This skill returns a short clip from one frame.
- You need frame-accurate product demo of a live UI. That is a screen recorder.
Connect ToolRouter, attach the still, and ask for one motion.
Frequently Asked Questions
Can ChatGPT turn an image into video?
**Not as a generic “animate this JPEG” in every plan.** Use the client’s own video product if you have it, or attach ToolRouter in Developer mode and pass the image URL to [Generate Video](/tools/generate-video).
Can Claude generate video from a photo?
**Not natively as a downloadable MP4.** Connect ToolRouter. See also [Can Claude generate video?](/blog/can-claude-generate-video).
Can Grok do image to video?
**Grok has its own media products, and grok.com can also call a custom MCP connector.** For the ToolRouter catalog path: [Connect Grok](/docs/connect-grok). For the downloadable app: [Grok Bot](/blog/how-to-add-mcp-to-grok-bot).
Do I need a first frame?
**Yes in this catalog.** Text-only video was removed so every model shares the same still-then-animate path.
How much does it cost?
**Depends on the model.** Call `list_models` on [Generate Video](/tools/generate-video) the same day you generate. Do not reuse an old price from a blog.
