Image-to-video is the bridge between static AI imagery and the dominant content format of 2026. A still image — a product photo, a portrait, a landscape, a brand visual — becomes a 6-10 second motion clip. The still is the anchor; the motion prompt is the change. The model is Grok Imagine.
This is the production guide: what Grok Imagine is, the 4 generation modes, the motion prompt structure, the 5 production patterns, the per-pattern prompts, the brand-safety rules, the iteration loop, and the workflow for shipping 20 short-form clips per week from your existing image library.
What Grok Imagine is (and isn't)
Grok Imagine is xAI's unified image + video + audio generation model. It supports five workflows:
- Text-to-image — generate a still from a text prompt
- Image editing — modify an existing image
- Text-to-video — generate a video from a text prompt
- Image-to-video — animate a still image into a short clip
- Image-to-video with native audio — animate a still with synchronized sound
For this post, the focus is image-to-video — the workflow that takes your existing image library (AI-generated or otherwise) and turns it into motion content. This is the highest-leverage use case for short-form video creators, social media managers, and ecommerce operators who already have a still-image pipeline.
The 4 generation modes
Grok Imagine has 4 generation modes that affect the visual style and content:
- Normal Mode — polished realism, balanced, the default. Use for most production work.
- Fun Mode — stylized color and energy, more dynamic. Use for creative / lifestyle / energetic content.
- Custom Mode — full control over the visual style. Use for branded content where you need to lock the look.
- Spicy Mode — mature, bold, NSFW-capable. Use only for adult content with appropriate consent and disclosure.
For most production work (social ads, ecommerce, brand content, lifestyle), Normal Mode is the default. Fun Mode is for content that needs more energy. Custom Mode is for brand-locked content. Spicy Mode is out of scope for commercial brand work.
The pricing and length
Grok Imagine's video generation as of June 2026:
- Length: 6-10 seconds per clip (depending on the endpoint and credits spent)
- Resolution: 720p standard, 1080p on higher credit tiers
- Audio: native synchronized audio on video output (background music, sound effects, dialogue)
- Pricing: credit-based. ~$0.07 per standard 6-second clip at 720p, ~$0.15 at 1080p with audio
- Aspect ratios: 16:9, 9:16, 1:1 supported
For a 6-second 720p social clip, the cost is roughly $0.07. For a 10-second 1080p clip with audio, the cost is roughly $0.15. At those prices, generating 20 clips per week for A/B testing is under $5/week.
The image-to-video prompt structure
The prompt structure for image-to-video has two parts:
- The reference image — the still that the model animates
- The motion prompt — the description of how the still should move
The format:
[Reference image: the still you want to animate]
Motion prompt: [description of how the still should move, what
should change, what should stay the same, and any audio cues].
Constraints: [duration, aspect ratio, mode, brand-safety rules].
The motion prompt is the key difference from text-to-image. The model already has the visual from the reference image. The motion prompt is only about change — what moves, how it moves, what stays still.
The 5 elements of a strong motion prompt
A strong motion prompt has 5 elements:
- Camera motion — what the camera does (push-in, dolly, pan, orbit, static, handheld)
- Subject motion — what the subject does (turns, walks, gestures, breathes, holds still)
- Environment motion — what moves in the scene (wind, water, particles, light flicker)
- Tempo — how fast / slow / energetic the motion is
- Audio cues — what the audio does (background music, sound effects, dialogue, ambient)
The format:
Motion prompt:
Camera: [camera motion].
Subject: [subject motion].
Environment: [environment motion].
Tempo: [tempo description].
Audio: [audio description].
Example:
Motion prompt:
Camera: slow push-in toward the subject's face, 1.5x over 6
seconds.
Subject: subtle smile widens, eyes blink once at the 2-second
mark, head turns 5 degrees to camera-right.
Environment: soft warm light flickers slightly as if a candle
in the background, no other motion.
Tempo: gentle, cinematic, slow.
Audio: soft ambient music, no dialogue.
Why it works: the model has explicit motion in every dimension. Camera, subject, environment, tempo, audio — all five are named. The model has no permission to invent motion that conflicts with the prompt.
The 5 production patterns for image-to-video
Pattern 1: Product hero turntable (the "360° product" pattern)
The single highest-ROI image-to-video pattern for ecommerce. Take a static product photo, animate it as a slow turntable or push-in. The result is a 6-10 second product clip that looks like a professional studio shoot.
The motion prompt structure:
[Reference image: the product on white background]
Motion prompt:
Camera: slow 180° orbit around the product at a constant
distance, 360° over 8 seconds.
Subject: the product rotates smoothly, no other motion, all
materials and colors remain exactly as in the reference image.
Environment: white background, no environment motion, soft
shadow follows the product as it rotates.
Tempo: slow, smooth, even.
Audio: soft ambient music, no dialogue.
Constraints: 8 seconds, 1080x1080, Normal Mode, product
preserved exactly, no fake text, no fake URLs.
Why it works: the camera does the work (orbit), the product stays the same (preserved), the audio is unobtrusive. The result looks like a $5,000 studio product turntable.
On Grok Imagine specifically: the orbit pattern is the strongest. The model handles smooth camera motion well.
The platform tweaks:
- Instagram Feed: 1:1, 6 seconds
- Pinterest: 2:3, 6 seconds
- TikTok / Reels: 9:16, 8 seconds
- Amazon / Shopify: 1:1, 6 seconds, no audio (autoplay restrictions)
Pattern 2: Portrait animation (the "subtle life" pattern)
Take a still portrait and add subtle motion. The head turns slightly, the eyes blink, the hair moves, the light shifts. The result is a portrait that feels alive without being uncanny.
The motion prompt structure:
[Reference image: a portrait of a person]
Motion prompt:
Camera: static, no camera motion.
Subject: subtle — eyes blink once at 2 seconds, head turns 3
degrees to camera-right between 3-5 seconds, soft smile widens
slightly.
Environment: soft natural light from window camera-left shifts
subtly as if a cloud passed.
Tempo: gentle, slow, cinematic.
Audio: no dialogue, soft ambient room tone.
Constraints: 6 seconds, 1080x1080, Normal Mode, preserve the
person's face and identity exactly, no fake expressions, no
fake emotions.
Why it works: the motion is minimal. A blink, a head turn, a slight smile. The brain reads the still image as a video without noticing the motion is artificial. The result is a portrait that feels alive.
On Grok Imagine specifically: the "subtle life" pattern is the strongest. The model handles small facial motion well.
The platform tweaks:
- LinkedIn: 1:1, 6 seconds
- Instagram Feed: 4:5, 6 seconds
- TikTok / Reels: 9:16, 8 seconds
Pattern 3: Lifestyle scene (the "moment" pattern)
Take a still lifestyle image (a person in a cafe, on a hike, in a kitchen) and add the motion that makes the moment feel real. Steam rising, leaves moving, the person gesturing, the light shifting.
The motion prompt structure:
[Reference image: a lifestyle scene]
Motion prompt:
Camera: static, no camera motion.
Subject: [specific subject motion, e.g. "the woman lifts the
coffee cup to her lips and takes a sip, then places it back on
the table"].
Environment: [specific environment motion, e.g. "soft steam
rises from the cup, leaves outside the window move slightly in
the breeze, the light shifts subtly"].
Tempo: natural, gentle, the speed of a real moment.
Audio: ambient cafe sounds, no dialogue.
Constraints: 8 seconds, 9:16, Normal Mode, preserve the scene
exactly, no fake people, no fake activities.
Why it works: the motion is the natural motion of the moment. Steam rising, leaves moving, the person doing a natural gesture. The result is a clip that feels like a real moment was captured.
Pattern 4: Brand visual (the "logo animation" pattern)
Take a static brand visual (logo, wordmark, hero image) and animate it. The logo assembles, the wordmark reveals, the hero image has subtle parallax. The result is a brand intro, a YouTube endcard, or a social media brand reveal.
The motion prompt structure:
[Reference image: the brand visual]
Motion prompt:
Camera: static or very slow push-in (0.5x over 6 seconds).
Subject: [specific brand motion, e.g. "the wordmark fades in
from left to right over 2 seconds, the symbol assembles from
discrete parts"].
Environment: clean background, no environment motion.
Tempo: smooth, professional, deliberate.
Audio: soft brand audio sting or no audio.
Constraints: 6 seconds, 16:9, Normal Mode, preserve the brand
mark exactly, no fake brand names, no fake taglines, all text
must be exactly as written.
The brand-safety rules:
- "preserve the brand mark exactly"
- "no fake brand names"
- "no fake taglines"
- "all text must be exactly as written"
Pattern 5: Reveal (the "before/after to motion" pattern)
Take a still that has a "before" state and a "during" state, and animate the transition. A box being opened, a curtain being pulled, a product being unwrapped. The result is a 6-second reveal clip.
The motion prompt structure:
[Reference image: the "after" state, with the "before" state
implied]
Motion prompt:
Camera: static, no camera motion.
Subject: [the reveal motion, e.g. "the box lid lifts open
slowly over 3 seconds, the product inside is revealed, the
lid continues to open until fully open by 5 seconds"].
Environment: clean background, no environment motion.
Tempo: slow, deliberate, building anticipation.
Audio: soft reveal sound, no dialogue.
Constraints: 6 seconds, 1:1, Normal Mode, preserve the
product exactly, no fake reveals, no fake products.
The 10-element motion prompt template (the master structure)
For complex clips, use the full template:
[Reference image]
Motion prompt:
Camera: [specific camera motion]
Lens feel: [specific lens, e.g. 35mm, 50mm, 85mm]
Subject: [specific subject motion]
Subject expression: [specific facial motion, if any]
Subject gesture: [specific gesture, if any]
Environment: [specific environment motion]
Lighting: [specific lighting motion, if any]
Tempo: [tempo description]
Audio: [audio description]
Constraints: [duration, aspect ratio, mode, brand-safety rules]
The format works for all 5 patterns — you fill in only the elements that apply.
The 12 brand-safety rules for image-to-video
Image-to-video adds new failure modes beyond still images. The rules:
- "Preserve the subject exactly as in the reference" — the reference is the source of truth
- "No fake people" — no invented faces or bodies
- "No fake actions" — no invented activities
- "No fake expressions" — no invented emotions
- "No fake text" — no invented captions, no fake URLs
- "No fake brand names" — no invented competitors or partners
- "No fake audio" — no invented dialogue, no fake music
- "No fake environments" — no invented scenes
- "No watermark" — clean output
- "All text must be exactly as written" — prevents paraphrasing
- "Preserve the brand mark exactly" — for branded content
- "Disclose AI generation per platform rules" — per the safety post
The iteration loop for image-to-video
Pass 1: Generate with the motion prompt. Run the reference + motion prompt. Watch the output.
Pass 2: Identify the motion drift. Common drifts:
- The subject moved too much (face changed, body position changed)
- The camera moved when it shouldn't have
- The environment motion was too aggressive
- The audio was wrong (music too loud, dialogue invented, sound effects off)
- The duration was wrong (too long, too short)
Pass 3: Refine the motion prompt. Add constraints. "Subject motion: only the eyes, no other motion." "Camera: static, no camera motion." "Environment: no motion."
Pass 4: Re-generate. Run the refined prompt. Watch again.
Pass 5: Adjust the audio. The audio often needs a separate pass. Specify the music, the volume, the sound effects.
Pass 6: Final check. Run the pre-flight checklist. Ship if clean.
The pre-flight checklist for image-to-video
Before you ship a motion clip:
- The reference is preserved (subject, scene, brand mark, text)
- The motion is intentional (matches the prompt, no random motion)
- The camera motion matches the prompt (or is static if specified)
- The subject motion is realistic (no uncanny valley, no extra limbs)
- The environment motion is appropriate (no excessive motion, no random objects)
- The audio is appropriate (matches the mood, no invented dialogue)
- The duration matches the platform (6-8 seconds for TikTok/Reels, 6 for Feed)
- The aspect ratio matches the platform (9:16 for vertical, 1:1 for Feed, 16:9 for YouTube)
- The brand-safety rules are applied (12 rules)
- The platform's AI label is applied (per the safety post)
Skip any of these and the clip is either uncanny, off-brand, or non-compliant.
The model pick by pattern
| Pattern | Best Grok mode | Why |
|---|---|---|
| Product turntable | Normal Mode | Polished realism, smooth camera motion |
| Portrait animation | Normal Mode | Subtle facial motion |
| Lifestyle scene | Fun Mode | Stylized energy, ambient motion |
| Brand visual | Normal Mode | Controlled, professional |
| Reveal | Fun Mode | Building anticipation, dynamic |
The 20-clips-per-week production cadence
Week 1: Establish the source library. Pick 20 still images (product photos, portraits, brand visuals, lifestyle scenes). These are the references.
Week 2: Generate 20 motion clips. One motion prompt per reference. Run the batch. ~30 minutes.
Week 3: QA and select the top 10. Watch all 20 clips. Identify the top 10 by visual quality and brand fit. Re-generate the bottom 10 with refined motion prompts.
Week 4: Generate 20 more from a different set. Different references, different motion patterns. The library compounds.
By week 12: 240+ motion clips, a tested library, and a continuously-improving production engine.
The cost is roughly $0.07-$0.15 per clip. 20 clips/week is under $5/week. The output is a month of social content for a fraction of the cost of reshooting.
The summary
Image-to-video is the bridge between still imagery and short-form video. Grok Imagine is the model. The 5 production patterns — product turntable, portrait animation, lifestyle scene, brand visual, reveal — cover ~90% of motion content needs.
- Use the 5-element motion prompt structure (camera + subject + environment + tempo + audio).
- Pick the Grok mode for the pattern (Normal for most, Fun for lifestyle and reveal).
- Apply the 12 brand-safety rules (preserve subject, no fake anything, all text in quotes).
- Run the 20-clips-per-week production cadence (~$5/week, 30 minutes of generation).
- Iterate with the motion-prompt refinement loop (identify drift, refine prompt, regenerate).
The model is not the bottleneck. The motion prompt discipline is. Pick the pattern, write the motion prompt, run the cadence, apply the brand-safety rules, and motion content becomes a continuously-improving, compounding production engine that turns your still-image library into short-form video.
