Image and video generation: parameters, limits, and use cases
The official description of the Grok Bot's design for image and video generation boils down to one sentence: it only creates when you ask, starting from your description or the images you share, and every image and video it makes is sent back to the conversation. The previous lesson covered getting materials in and out; this lesson covers how it "creates something from nothing": how parameters like aspect ratio, duration, and resolution are set, how to provide reference images, and what jobs it can't do.
The whole thing takes about 10 minutes, and there isn't a single command to run—every operation is just "talking to it."
What you'll be able to do after this lesson
- Use one sentence of your own to spell out the subject, style, and mood, and have the Bot output an image at the specified aspect ratio—that's the entirety of Grok Bot image generation's "parameters"
- Share existing images into the conversation as reference images, so it can keep creating on top of your visuals
- Generate videos: set duration (1–15 seconds, default 6 seconds), resolution (480p, can request 720p), and aspect ratio, and use a first and last frame to make a looping clip
- Use up to 14 reference images to lock down people, objects, clothing, and scenes, then use up to 4 keyframe images to specify "must appear at the Nth second"
- Tell at a glance which kinds of requests to take elsewhere: charts and tables go to Pages, for real people and real objects it finds you real photos, and trademarks and characters often get blocked
Your current struggles
- Getting AI to draw images feels like a slot machine: wrong aspect ratio, off-style results, and re-rolling the same prompt ten times still doesn't hit the mark
- You want "the backpack must appear on the table at the 4-second mark" in a video, but don't know how to tell it that
- You ask it to draw a data chart and the numbers turn into a blur; you name a well-known character and get blocked outright by content rules
When to use this technique
When you need:
- A cover, illustration, or concept image: aspect ratio, style, and mood on your terms
- A 1–15-second live-action-feel short clip: starting from a scene description, a single first-frame image, or a first and last frame
- Multiple outputs that stay consistent: use reference images to lock down the look of characters, clothing, and scenes
And you don't want to:
- Charts and tables "drawn" into existence—the numbers are simply unreadable
- Naming trademarked characters or real people and getting blocked over and over without knowing why
- Losing track of files after generation, unable to send them anywhere
🎒 Before you start
- Completed 2.1 Messages and conversations: you can send messages with attachments—both image and video generation start with "talking to it"
- Know the desktop attachment limits: 20 attachments per message, images up to 25 MB each and 200 MB total (see 2.2 Files and reviewable results)
- Have at least one Bot on hand and have assigned it a task once (see 1.5 Your first Bot)
- A cup of water ☕—this lesson has no commands to run; every operation is just "talking to it"
Core idea
The way the Grok Bot generates images and videos is completely different from "opening an image-generation website and tweaking parameters": there's no parameter panel, no model selection—all parameters live in your words.
Three main threads:
- It only creates when you ask. It starts from your description, or from the images you share into the conversation; if you don't ask, it doesn't generate.
- Output comes back to the conversation. Every image and video it makes is posted to the conversation, and you can find them later in the Bot's Library; generated images load normally even when its cloud computer is asleep.
- Its capabilities have hard boundaries. Image generation isn't for charts, graphs, or tables—those belong to Pages; videos are live-action-feel and don't do motion graphics, title cards, slides, product demos, or precise text; for real people, real places, and real products, it finds real photos instead of generating them.
A side-by-side of image and video generation parameters, to get familiar first:
| Image generation | Video generation | |
|---|---|---|
| How to start | Describe the subject, style, mood; reference images optional | Describe the scene; or provide a first-frame image / a first-and-last frame pair |
| Aspect ratio | 1:1, 4:3, 3:4, 16:9, 9:16 | 1:1, 16:9, 9:16, 4:3, 3:4, 3:2, 2:3 |
| Duration / Resolution | — | 1–15 seconds (default 6 seconds); 480p, can request 720p |
| Reference images | just share them into the conversation | up to 14 reference images + 4 keyframe images |
| Input images | subject to attachment limits (each image 25 MB) | JPEG / PNG / WebP, about 35 MB total per video |
| Output | images, sent to the conversation | MP4 files, sent to the conversation |
Follow Along
Step 1: Use one sentence to have the Grok Bot generate an image: subject, style, mood, and aspect ratio
Why: Grok Bot image generation has no parameter panel—subject, style, mood, and aspect ratio all depend on how you describe them in your own words; state them clearly to get it right the first time.
- Three elements of a description: subject (what to draw), style (what art style), mood (what atmosphere)—just write them in your own words
- For a specific frame, name the aspect ratio directly: 1:1 (square), 4:3 / 16:9 (landscape), 3:4 / 9:16 (portrait)
- The Bot only creates when you ask; every finished image is sent back to the conversation
Sample wording (swap in your own content and send it directly):
Help me generate a 16:9 cover image: a city rooftop at dawn, a person facing away from the camera watching the sunrise, watercolor style, a quiet atmosphere.
You should see: the image appear in the conversation, openable in full screen and zoomable; even if its cloud computer is asleep, already-generated images still load normally.
- Used a one-sentence description to have the Bot generate an image, and specified the aspect ratio
Step 2: Generate an image with reference images: have it draw based on your picture
Why: To keep creating based on an existing image—say, "draw in the feel of this photo"—share the image into the conversation and state clearly how each image should be used; the Bot will treat them as references.
- Paste the image into the input box, or send it with the attachment control (on desktop, each image 25 MB, 200 MB total per message)
- State each image's purpose in the same message—"this is a product photo," "follow its color palette"—don't make the Bot guess
Sample wording:
The attachment is a product photo. Reference its color palette and composition, and generate a new 1:1 image, replacing the subject with an iced coffee.
You should see: the Bot combines the reference images to generate a new image whose picture matches your description, rather than improvising out of thin air.
💡 Reference images aren't just for image generation
Video also accepts reference images, with finer limits: up to 14 reference images + 4 keyframe images—see Step 4.
Step 3: Generate a video: start from a description or an image, set duration, resolution, and aspect ratio
Why: Grok Bot's video generation parameter limits are spelled out officially—know them before you ask, say it all in one go, and save yourself back-and-forth revisions.
Videos can be generated from three kinds of starting points:
| Starting point | What to say |
|---|---|
| Scene description | Describe the picture directly: content, camera, atmosphere |
| One image | Share an image, and it becomes the video's first frame |
| Two images | One start frame, one end frame; using the same image for both = a looping clip |
Parameter limits (all specified in your message):
- Duration: 1–15 seconds, default 6 seconds—for a different duration, just say so
- Resolution: 480p; for sharper output, explicitly request 720p
- Aspect ratio: 1:1, 16:9, 9:16, 4:3, 3:4, 3:2, 2:3; when generated from your image it keeps the original image's shape unless you specify otherwise
- Input images: JPEG, PNG, or WebP; all input images for one video total about 35 MB
- Output: each clip is saved as an MP4 file and sent back to the conversation
- Generating a video can take a few minutes, so don't rush it
Sample wording:
Use the image I just sent as the first frame, and generate a 9:16 8-second video, with the camera slowly pulling back, 720p.
You should see: after a few minutes, the MP4 appears in the conversation; click the thumbnail for full-screen playback, and the player has a progress bar and 2× / 1.5× / 1× / 0.5× playback speed.
- Generated a video (starting with the default 6 seconds is fine), or can recite the limits for duration, resolution, and aspect ratio
Step 4: Control video frames with reference images and keyframe images
Why: Video is harder to get right in one shot than images—the official docs give you two sets of images to lock down the picture: reference images govern "what it looks like," keyframe images govern "when it appears".
- Reference images (up to 14): show what people, objects, clothing, and scenes should look like
- Fixed-point images (up to 4): specify frames that must appear at specific moments in the clip
Sample wording:
The first 3 attached images are reference images of the character; follow the outfit and hairstyle shown in them. The 4th image (the one with the red backpack) must appear on the desk at around the 4th second of the video. Generate a 6-second 16:9 video.
You should see: the person and outfit in the video match the reference images, and the fixed-point image content appears at the moment you specified.
⚠️ Input images have format and size thresholds
Images for a video must be JPEG, PNG, or WebP, and all input images for a single video should total about 35 MB. Check the format before submitting, so you don't discover the images are rejected only at the generation step.
- Know the respective purposes of reference images (up to 14) and fixed-point images (up to 4)
Step 5: Understand the four ways the Bot "doesn't do what you said"
Why: several kinds of "not doing what you said" are officially documented design behaviors and limitations—identify which one you're hitting first, then decide the next step, instead of treating it as a bug and retrying repeatedly.
| What you're seeing | Official stance | What you should do |
|---|---|---|
| You want to see a real person, real place, or real product, but the Bot returned a real image | Design behavior: it finds real images rather than creating them | If you want the real thing, use it directly; if you want a generated result, describe the scene in your own words and don't name real objects |
| Your request was blocked by content rules and the Bot told you | Naming trademarked characters, brands, or real people often triggers the block | Describe the subject in your own words and try again |
| The Bot tells you to "try again later" | Image generation is busy | Wait a while and send again |
| The Bot says your account doesn't have image or video generation | It's been turned off for your account; in a team it may be caused by the admin's model settings or privacy settings | Ask your admin |
You should see: each kind of "not doing what you said" has a clear cause—it's a boundary, not a malfunction.
Step 6: View, save, and forward the Bot's output
Why: images and videos are both posted in the conversation, but to actually use them—save them locally or send them to another chat—you need to know the entry points on each platform.
Desktop:
- Click an image or video thumbnail: view full screen; click again to zoom in on images; switch among multiple outputs with left/right arrow keys or the thumbnails; the download button at the top saves;
Escape, the close button, or clicking outside the image closes it - Video player: play / pause, progress bar, playback speed menu 2× / 1.5× / 1× / 0.5×
- Right-click: image → Copy image or Save image…; video → Save video…
- Hover over a message and click the three dots: Download; when a message has multiple outputs, use Download all; images also have Copy
Mobile:
- Tap an image to view full screen → viewer menu: Forward (forward to another chat), Share, Save
- Long-press an image in the conversation: Copy Image, Save to Photos
- Tap a video to play it directly in the conversation; long-press → Save to Photos
Finding them again: every image and video the Bot has made is recorded in its Library (desktop: Bot name at the top of the chat → View conversation details → Library; mobile: Bot name → profile → Library). For details on the operations, see 2.2 Files and reviewable results.
You should see: images and videos can be viewed full screen, saved locally or to your photo library, and forwarded to other chats; the Library lets you find them again by time.
- Saved an image or video generated by the Bot
Checkpoint ✅
- Described the subject, style, and mood in your own words and had the Bot generate an image at a specified aspect ratio
- Generated an image with reference images, or know that reference images require "clearly stating how to use them"
- Generated a video, or can recite: 1–15 seconds with a 6-second default, 480p can be requested as 720p, 7 aspect ratios, output is MP4
- Know that charts and tables shouldn't be generated as images—use Pages; for real people and things the Bot finds real images
- Saved and forwarded the Bot's output on desktop or mobile
Pitfall reminders
❌ Asking the Bot to draw charts or tables with image generation
Symptom: you ask it to draw a data bar chart or a table, and the numbers come out blurry and don't match the data at all.
Cause: the official documentation is explicit—image generation is not for charts, graphs, or tables.
The right approach: Have the Bot turn this kind of content into a page (Pages): it can hold text, tables, charts, and controls—reports and dashboards both rely on it—see C2 Delivering Reports and Content Dashboards with Pages.
❌ Naming trademarked characters, brands, or real people—request gets blocked
Symptom: You ask it to draw a well-known character or a real person, and the Bot says the request was blocked by content rules.
Cause: The official guidance is clear—naming trademarked characters, brands, or real peopleoften triggerscontent blocks; the Bot will tell you it was blocked rather than failing silently.
The right approach: Describe the subject in your own words (appearance, clothing, vibe) without naming it, then try again.
❌ You want to see what a real person or landmark looks like, but the Bot “found a photo”
Symptom: You ask it to “draw” a real place, and it replies with an existing photo instead of a generated image.
Cause: By design—when showing what a real person, place, or product looks like, the Botfinds real photosinstead of creating them.
The right approach: If you want the real look, just use the real photo it found; if you want a stylized image, describe the scene itself and don't name the real subject.
❌ Making video do “precision work”: motion titles, slides, product demos
Symptom: You want a motion video with exact text, exact timing, or brand assets, and the result doesn't match.
Cause: The official guidance is clear—video has a filmed look and isn't suited to motion graphics, title cards, slides, product or app demos, charts, or anything requiring exact text, timing, or brand assets.
The right approach: Use video within its strengths: filmed-style scenes, animating still images, characters from reference images. Leave precision-text work to Pages or documents.
Advanced tips (optional)
Work handed to a task can also generate images and video. When the Bot hands work to a task, that task can also create images and videos, and the Bot sends them back to the conversation—you don't need to change how you assign work.
Cloud computer sleep doesn't affect viewing images. Images the Bot generated still load normally in the conversation while its cloud computer is asleep—you can browse earlier outputs without waking it.
Charts and dashboards go through Pages. For content that needs charts, tables, or controls, have the Bot make it as Pages—for version requirements and sharing options, see C2 Delivering Reports and Content Dashboards with Pages.
Lesson recap
✅ Grok's main thread for image and video generation: it only creates when you ask, starts from a description or an image you share, and outputs go back to the conversation and are recorded in the Library.
✅ Image generation: describe the subject, style, and mood; aspect ratios 1:1, 4:3, 3:4, 16:9, 9:16; share existing images into the conversation as references.
✅ Video generation: 1–15 seconds (default 6 seconds), 480p can be upgraded to 720p, 7 aspect ratios; start from a first-frame image or first and last frames, same image for first and last = loop; up to 14 reference images + 4 keyframe images; input images JPEG / PNG / WebP totaling about 35 MB; output MP4.
✅ Boundaries: use Pages for charts and tables, not image generation; video has a filmed look and doesn't do motion graphics, exact text, or brand assets; for real people and real things the Bot finds real photos; trademarks and real people are often blocked—describe in your own words and try again.
✅ Outputs are usable anytime: full-screen on desktop / download / right-click copy, forward or save to gallery on mobile; find them again in the Library by time.
Key parameters recap:
| Parameter | Value | Notes |
|---|---|---|
| Image aspect ratios | 1:1, 4:3, 3:4, 16:9, 9:16 | Square / landscape / portrait—name it in your description |
| Video duration | 1–15 seconds | Default 6 seconds |
| Video resolution | 480p | Ask explicitly for 720p for more clarity |
| Video aspect ratios | 1:1, 16:9, 9:16, 4:3, 3:4, 3:2, 2:3 | Keeps the original image's shape when generated from your image |
| Reference images | Up to 14 images | Lock in people, objects, clothing, scenes |
| Keyframe images | Up to 4 images | Appear at specified moments |
| Input images | JPEG / PNG / WebP, totaling about 35 MB | All input images for one video |
| Video output | MP4 | Generation may take several minutes |
Next lesson preview
It can now “create from nothing”; the next lesson shows how it “remembers you.” Next lesson: 2.4 Memory.
You will learn:
- The Bot decides on its own what to remember: a small set of facts about you is shared across all Bots
- Memory persists across chats: it still remembers your preferences in a different conversation
- View, correct, and remove what it remembers right in the chat
- Reset resets the conversation but keeps memory; you can also import from other assistants