How to Use ChatGPT With imageat to Generate Images and Videos

ChatGPT is useful for developing a visual idea, but a prompt alone is not a finished asset. Connect ChatGPT to imageat through the Model Context Protocol (MCP), and the conversation can continue into image generation, video generation, editing, and credit checks without a manual copy-and-paste loop.

This guide focuses specifically on the ChatGPT workflow: how to connect the remote imageat MCP server, how to ask for images and videos, how to control the result, and what to try when a tool call does not behave as expected.

The connection does not make every vague request production-ready. You still need a clear brief, suitable references, and deliberate review. What it changes is the handoff: ChatGPT can help plan the asset and call imageat from the same working context.

What you need before you start

You need an imageat account with credits and a version of ChatGPT that shows the connector controls required for remote MCP services. ChatGPT features and interface labels can vary by account, plan, workspace policy, region, and release, so confirm that Settings → Connectors is available in your current interface.

The live imageat MCP page lists this remote endpoint:

https://mcp.imageat.com/mcp

For the remote connector flow, imageat says you sign in through the browser and do not need to copy an API key into ChatGPT. Usage is billed to your imageat credits. The authorized connection can be revoked from the Projects area in imageat.

If you want a broader explanation of the protocol and other clients, read the existing guide to MCP image generation in Claude, ChatGPT, and Cursor. This article stays with the narrower, task-oriented ChatGPT search intent so the two resources do not duplicate each other.

How to connect ChatGPT to imageat

Use the current setup panel on the imageat MCP page as the source of truth because connector interfaces can change.

1. Copy the remote MCP endpoint

Open the imageat MCP page and copy the remote endpoint shown in the connection section:

https://mcp.imageat.com/mcp

Use the exact HTTPS address. Do not paste a blog URL, the imageat homepage, or a package-install command into the remote connector field.

2. Open ChatGPT connector settings

In ChatGPT, open Settings, find Connectors, and choose the option to add a custom connector. Name it imageat, then paste the remote endpoint.

The imageat setup page currently instructs remote-connector users to leave the OAuth fields empty. Follow the current on-page instructions rather than inventing client IDs or secrets.

3. Sign in and approve access

Continue from ChatGPT. A browser authorization flow should open imageat, where you can sign in and approve the connection. Once approved, return to ChatGPT and confirm that imageat appears among the available connected tools.

Treat this approval like any other account integration. Use the Projects area in imageat if you later need to revoke the connection.

4. Run a small test before a full project

Start with a low-complexity request that confirms three things: ChatGPT can see imageat, the tool can run, and the result returns to the conversation.

Use imageat to generate one square editorial still life: a matte ceramic cup
on a pale stone surface, soft window light from the left, restrained shadows,
no text, no logo, no people. Before generating, briefly restate the format and
visual direction you will send to the tool.

Asking ChatGPT to restate the request first gives you a chance to catch a wrong aspect ratio, an unwanted subject, or an overcomplicated art direction before credits are used.

A reliable prompt structure for image generation

ChatGPT performs better as a creative operator when the request is a compact brief, not a collection of adjectives. Include these six elements:

• Purpose: product page, social post, blog hero, concept art, or presentation.
• Subject: what must appear and what must remain unchanged.
• Scene: setting, props, background, and time of day.
• Camera: framing, angle, lens character, and depth of field.
• Lighting and style: direction, contrast, palette, and visual references described in plain language.
• Output constraints: aspect ratio, number of options, and exclusions such as no text or no watermark.

Here is a reusable template:

Use imageat to generate [NUMBER] image(s) for [PURPOSE].

Subject: [MAIN SUBJECT AND NON-NEGOTIABLE DETAILS]
Scene: [SETTING, BACKGROUND, PROPS]
Camera: [SHOT SIZE, ANGLE, LENS CHARACTER]
Lighting: [DIRECTION, QUALITY, CONTRAST]
Style: [VISUAL TREATMENT AND COLOR PALETTE]
Format: [ASPECT RATIO OR DESTINATION]
Avoid: [TEXT, LOGOS, EXTRA OBJECTS, ANATOMY OR COMPOSITION RISKS]

First summarize the brief in one sentence. Then generate only after the
requirements are internally consistent.

For direct browser-based generation and manual model selection, the AI image generator remains useful. The MCP route is most valuable when the brief is already being developed inside ChatGPT or when several creative steps need to share the same context.

Example: create a product campaign image

Suppose you need a launch visual for a reusable water bottle. Do not begin with “make a premium product ad.” Give ChatGPT enough information to preserve the product and design the scene.

Use imageat to generate a 4:5 paid-social product image.

The subject is a brushed stainless-steel insulated bottle with a simple black
cap. Preserve its cylindrical proportions and do not add labels or text.
Place it on a dark slate ledge after light rain, with small natural water
droplets and a softly defocused mountain trail behind it.

Camera: eye-level three-quarter product shot, moderate telephoto look, clean
verticals, enough negative space above the bottle.
Lighting: cool overcast ambient light with a subtle warm edge light.
Style: realistic outdoor campaign photography, restrained color grade.
Avoid: duplicate bottles, floating objects, distorted cap geometry, readable
branding, badges, and typography.

Generate one draft first. Do not create variations until I review it.

That final instruction matters. One controlled draft is usually more useful than spending credits on several variations before you know whether the composition is right.

Review the first result in layers:

1. Is the product geometry believable?
2. Does the focal point read at thumbnail size?
3. Is there enough space for real copy to be added later in a design tool?
4. Did the generator introduce unwanted text, logos, or duplicate objects?
5. Is the lighting consistent with the environment?

Then request one change at a time. For example: “Keep the bottle, framing, background, and camera position unchanged. Make only the edge light slightly warmer.” Smaller revisions are easier to evaluate than a complete rewrite of the prompt.

How to make a video from an image in ChatGPT

For many marketing tasks, image-to-video is easier to control than starting from text alone. First create or upload a strong source image. Then describe motion without redesigning the scene.

The imageat AI video generator supports text-to-video and image-to-video workflows. Through MCP, ChatGPT can help turn the still-image brief into a motion plan and call the relevant imageat capability from the conversation.

Use this motion template:

Use this approved image as the visual reference and create an image-to-video
clip with imageat.

Subject motion: [ONE PRIMARY ACTION]
Camera motion: [ONE CAMERA MOVE]
Environmental motion: [ONE OR TWO SUBTLE EFFECTS]
Continuity: preserve [PRODUCT SHAPE / FACE / CLOTHING / BACKGROUND LAYOUT]
Timing: [OPENING, MIDDLE, END STATE]
Avoid: cuts, new objects, text, logo changes, sudden zooms, camera shake,
warped geometry, and a freeze-frame ending.

Before generating, explain which motion is subject motion and which is camera
motion. If the request is physically contradictory, stop and suggest a simpler
version.

A product-video request might be:

Animate the approved bottle image with imageat. The bottle remains fixed on the
slate ledge. The camera makes a slow, short push-in. A few water droplets slide
naturally down the metal, and distant leaves move slightly in the wind. Preserve
the bottle silhouette, cap, reflections, and background layout. No cuts, no
rotation, no added text, and no new objects. End with the bottle still fully in
frame.

This separates three kinds of movement: the subject stays fixed, the camera pushes in, and the environment moves subtly. That is more actionable than asking for a “dynamic cinematic video,” which leaves too many decisions unresolved.

Example: plan a short social video from scratch

ChatGPT can also help turn a message into a shot plan before any tool runs. Ask for the plan first, approve it, and then generate only the necessary shots.

We need a 12-second vertical social video for a compact desk lamp. The message
is: focused light for late-night work without lighting the whole room.

First create a three-shot plan. For each shot, specify the subject, framing,
camera movement, physical action, and transition. Keep every shot practical to
generate and avoid readable on-screen text. Do not call imageat yet.

After reviewing the plan, continue:

Use imageat to generate the approved shots one at a time. Start with shot one
only. Preserve the same lamp design, desk material, room palette, and nighttime
lighting across the sequence. Return the result for review before generating
shot two.

Generating serially gives you a checkpoint between shots. If the lamp design changes in the first result, fix the reference and prompt before spending credits on the rest of the sequence.

For creator-led product content, you can also move from the approved concept into the dedicated AI UGC generator. Use that when the deliverable needs a UGC-style spokesperson or product-ad workflow rather than a purely cinematic product shot.

How to use references without losing control

When you provide an image, tell ChatGPT what the reference controls. A reference might define identity, product geometry, pose, palette, composition, or only mood. If you do not say which, the tool may preserve the wrong attribute.

Use language like:

Treat the uploaded photo as a strict reference for the product's shape, cap,
material, and proportions. It is not a reference for the background, lighting,
camera angle, or crop. Keep the product design fixed while replacing the scene
with the brief below.

For a person, separate identity from styling:

Preserve the person's facial identity, approximate age, skin tone, and hairline.
You may change wardrobe, setting, lighting, and camera distance. Keep facial
features natural and do not add beauty marks, jewelry, or text.

If consistency is critical, reuse the same approved source asset and repeat the non-negotiable features in each request. Conversation context helps, but it is not a substitute for explicit continuity instructions.

Build a generation-review-edit chain

A useful ChatGPT and imageat workflow has clear gates:

1. Brief: ChatGPT turns the business goal into a visual specification.
2. Concept: ChatGPT proposes a few directions without generating.
3. Draft: imageat generates one selected direction.
4. Review: you identify one or two specific problems.
5. Edit: ChatGPT calls the suitable imageat edit or generation step.
6. Motion: the approved still becomes a video reference if needed.
7. Finish: upscale only the selected final image when a larger delivery file is required.

The live MCP page describes image generation, video generation, more than 20 editing tools, and credit checks. Examples listed there include background removal, relighting, virtual try-on, and object erasing. Ask ChatGPT to inspect the available imageat tools rather than guessing a tool name.

When a final still needs more resolution, use the image upscaler after composition and detail are approved. Upscaling early does not fix a weak concept, warped product, or bad crop; it only adds another costly step to a result you may discard.

Credit-aware working habits

Do not assume a fixed credit price for every task. Model choice, output settings, and product configuration can change. Check the current interface and ask ChatGPT to use imageat's credit-check capability before a large batch.

A practical instruction is:

Before running any generation, check my available imageat credits. Tell me how
many outputs you intend to request and ask for confirmation if the workflow
would create more than one asset. Do not launch variations automatically.

Other ways to avoid waste:

• Plan concepts in text before generating.
• Produce one draft before a batch.
• Lock the aspect ratio and destination early.
• Revise one variable at a time.
• Save the approved source image used for image-to-video.
• Do not upscale every draft.
• Stop a workflow when the product, character, or composition has drifted.

Troubleshooting ChatGPT and imageat MCP

ChatGPT does not show Connectors

Connector access may differ by ChatGPT account, plan, workspace, region, or current product rollout. Check the current ChatGPT settings and workspace policies. If the required connector option is unavailable, use imageat directly in the browser rather than trying to force an unrelated plugin or API configuration.

The connector does not authorize

Confirm that the endpoint is exactly https://mcp.imageat.com/mcp, that you are signed in to the intended imageat account, and that browser pop-ups or redirects are not being blocked. Return to the live MCP setup page for the latest connection instructions.

ChatGPT discusses the image but does not call imageat

State the action explicitly: “Use the connected imageat tool to generate this asset now.” If you want review before generation, say so separately. Also confirm that imageat is enabled for the current conversation and ask ChatGPT to list the relevant connected tools it can currently access.

The wrong tool or model is selected

Describe the outcome and constraints rather than inventing internal tool names. Ask ChatGPT to inspect the available imageat capabilities and explain its proposed choice before running it. For a high-control task, specify whether the input is text-only, image-referenced, an edit, an upscale, or image-to-video.

The output contains warped products, faces, or text

Return to the source image and simplify the request. Reduce simultaneous motion, avoid asking the camera and subject to perform conflicting moves, and state the features that must remain fixed. Add real typography after generation in a design workflow when exact brand text is required.

The result has the wrong aspect ratio

Put the destination and aspect ratio near the start of the request. “Vertical social post” can mean different layouts; “9:16 vertical video” or “4:5 feed image” is clearer. Confirm the format in ChatGPT's pre-generation summary.

A video changes too much

Use an approved source still, one primary subject action, one camera move, and limited environmental motion. Ask for continuity explicitly. If the first clip drifts, do not solve it by adding more adjectives; simplify the motion plan.

FAQ

Can ChatGPT generate images with imageat?

Yes, when your ChatGPT account supports the required remote connector flow and imageat MCP is connected. ChatGPT can prepare the brief and call imageat's image-generation capability from the conversation.

Can ChatGPT generate videos with imageat?

Yes. The current imageat MCP page lists video generation as a supported capability, including text-to-video and image-to-video workflows. Results still depend on the selected tool, input, model, and settings available at the time of generation.

Do I need an imageat API key for ChatGPT?

For the remote connector flow documented on the current imageat MCP page, you sign in with imageat through the browser and do not copy an API key into ChatGPT. Desktop clients can use a different setup, so follow the instructions for the client you are actually connecting.

Does ChatGPT decide which image or video model to use?

It can help choose among the tools and models exposed by the connection, but you should review the proposed choice when quality, speed, continuity, or credit use matters. Ask it to explain the choice without claiming unsupported specifications.

Are imageat generations free inside ChatGPT?

The imageat MCP page states that usage is billed to imageat credits. Check your current balance and the live product interface before generating; do not rely on an old article for fixed credit costs.

Can I use an existing photo as a reference?

Yes, when the available workflow accepts an image input. Tell ChatGPT exactly what the reference controls—such as identity, product shape, composition, or palette—and what may change.

Is MCP better than using the imageat web app?

It depends on the task. MCP is useful when ChatGPT already holds the campaign, product, or creative context and you want a multi-step conversational workflow. The web app is often better when you want direct visual control, manual model browsing, or a quick standalone generation. Many projects use both.

How do I disconnect ChatGPT from imageat?

Use the account and project controls described on the current imageat MCP page to revoke the authorized connection. You can also remove or disable the connector from ChatGPT's settings.

Start with one controlled asset

The best first use of ChatGPT with imageat is not a 30-asset batch. It is one well-defined image, reviewed once, followed by one deliberate revision. That small workflow proves the connection, reveals how ChatGPT translates your instructions, and helps you establish a prompt pattern before a larger campaign.

Open the imageat MCP connection guide, add the remote connector in ChatGPT, and begin with a brief that names the purpose, subject, camera, lighting, format, and exclusions. Once the still is approved, carry the same context into a restrained image-to-video request or a focused editing step.