Don't stop here
Hand-picked guides our readers explore right after this one.
Master xAI Grok with real-time web access, deep reasoning, and X/Twitter integration prompts
Read the guideUnlock Google's Gemini with multimodal prompting strategies
Read the guideAI prompts for LinkedIn posts, profile optimization, outreach, thought leadership, and job search
Read the guideGrok Imagine limits are the most volatile in the industry right now, which is why the cap you hit today may be nothing like the one you had last month. After non-consensual deepfake imagery was generated at scale through the free tier in January 2026, xAI tightened limits across every tier, and a May 2026 support email revised paid video caps downward again (SuperGrok Heavy to more than 80 videos per 12 hours, and standard SuperGrok to more than 20 videos per 24 hours, both below earlier figures). On top of the published numbers sits a fair-use algorithm that throttles heavy users during peak hours, with reset windows reported anywhere from 2 to 24 hours depending on the feature. The detail that catches people out most: failed generations still count against your allowance, so a run of refusals or errors can burn a day's cap without producing a single usable image. Because xAI changes these numbers by email and in-app rather than in stable documentation, treat every figure here as a reported snapshot and verify against your own account.
'Limit reached' or a rate-limit notice blocks new generations
Your daily allowance is visibly lower than it was weeks ago
Failed or refused generations still reduce your remaining count
Limits bite harder at busy times of day than at quiet ones
Video generation is blocked while image generation still works, or the reverse
The reset takes hours longer than you expected
After the January 2026 deepfake incident on the free tier, xAI reduced limits everywhere. A May 2026 support email revised paid video caps down again, so figures published earlier in the year overstate what you actually get.
The one that surprises people most. A refusal, a moderation block, or an error still consumes quota, so an unlucky prompt run can exhaust your cap with nothing to show for it.
xAI applies a fair-use algorithm that throttles heavy users during peak hours. That is why two accounts on the same tier get different real-world throughput, and why the published number is a ceiling rather than a guarantee.
Image and video generation are metered separately, and reset windows are reported between 2 and 24 hours depending on the feature. Being blocked on video while images still work is normal, not a bug.
Free, Premium+, SuperGrok and SuperGrok Heavy have all been revised more than once in 2026, and older guides quote much higher numbers (SuperGrok was advertised at up to 200 images per day before January 2026). People plan against stale figures and hit the wall early.
When to try: First
Open Grok and look at the current limits shown for your tier in-app or in the most recent xAI email. Published numbers from earlier in 2026 are materially higher than what is live, so planning against them guarantees you hit the cap unexpectedly.
When to try: Immediately after hitting the cap
Reset windows run from roughly 2 to 24 hours depending on the feature, and video windows are the longest. If the app names a time, that is authoritative. Rapid retrying during a throttle does not help and can consume quota if the attempts register as generations.
When to try: Before every generation, especially video
Because failures still count, treat every generation as paid. Draft the prompt in a notes app: subject, style, camera or composition, lighting, and what to avoid. Then submit once. This single habit recovers more usable output per cap than anything else.
When to try: Any time your prompt involves a person
Requests involving real, identifiable people, public figures, or anything near the non-consensual imagery line are refused, and the refusal still costs you quota. Given the January 2026 incident, moderation on people prompts is strict. Describe fictional or generic subjects instead of naming real ones.
When to try: Before any video generation
Video caps are the tightest and reset slowest. Generate a still first to validate the subject, style, and framing, then commit a video generation once you know the concept holds. Iterating in video is the fastest way to burn a 12 or 24 hour window.
When to try: When throttled despite having allowance left
The fair-use throttle bites hardest when the service is busy. Shifting heavy batches to off-peak hours for your region measurably improves throughput even without changing tier.
When to try: If you hit the cap most days
If you generate daily, the tier difference is large: free is heavily restricted, Premium+ and SuperGrok sit in the middle, and SuperGrok Heavy carries the highest reported video caps. If upgrading is not worth it, run stills in another generator and save Grok's allowance for the work only it does well.
When to try: Ongoing, if you use Imagine heavily
Record each prompt and whether it succeeded, was refused, or errored. Since failures consume quota, knowing which phrasings get refused is directly worth allowance, and the log doubles as a reusable prompt library.
Draft and refine prompts outside the app, since failed runs still cost quota
Validate a concept as a still image before spending scarce video allowance
Keep real, identifiable people out of prompts, refusals burn your cap
Verify your tier's current limits in-app, published 2026 figures went stale fast
Contact xAI support through the Grok app or your account page if a paid tier shows free-tier limits, if the allowance never resets after the longest stated window, or if generations fail repeatedly with errors rather than refusals while still consuming quota. Include your tier, timestamps, and the number of attempts. The limit levels themselves are set by xAI and have been revised repeatedly in 2026, so those are feedback rather than a fixable ticket.
xAI tightened limits across all tiers in 2026. After non-consensual deepfake imagery was generated at scale via the free tier in January 2026, caps were reduced, and a May 2026 support email revised paid video caps down again, to more than 80 videos per 12 hours on SuperGrok Heavy and more than 20 per 24 hours on standard SuperGrok. Earlier figures, including the up-to-200-images-per-day SuperGrok number, are stale.
Yes. Refusals, moderation blocks, and errors still consume quota. That is the most common reason people run out with nothing usable produced, and it is why drafting prompts carefully before submitting matters more on Grok than on tools that only charge for successes.
Reported reset windows run from about 2 to 24 hours depending on the feature, with video the longest. A separate fair-use algorithm throttles heavy users during peak hours on top of that, so throughput varies by time of day. If the app shows a time, trust it over any published figure.
SuperGrok Heavy carries the highest reported video caps (more than 80 videos per 12 hours as of the May 2026 revision), with standard SuperGrok around more than 20 videos per 24 hours, Premium+ reported around 100 video renderings per day at points during 2026, and free heavily restricted. Numbers have been revised repeatedly, so verify in-app rather than trusting a guide.
Draft prompts fully before submitting, since failures cost quota. Validate concepts as still images before committing video generations. Keep real, identifiable people out of prompts to avoid refusals. Generate off-peak to dodge the fair-use throttle. Keep a log of which phrasings get refused.