First check: are you on a paid tier?
The March 19, 2026 change is the single biggest cause of "Grok Imagine not working" reports since spring. xAI removed the free tier for image and video generation with no email announcement and no in-app notice — users found out when generation just stopped responding. If you can still sign in and chat with Grok normally but Imagine either silently fails or prompts you toward a subscription, this is what you're hitting.
The paid plans that unlock Grok Imagine as of 2026:
X Premium. Basic paid X (formerly Twitter) subscription. Includes Grok access with limited image gen.
X Premium+. Higher X tier. More generation headroom.
SuperGrok. xAI's dedicated Grok subscription. Highest image + video quotas.
The change was made in response to reporting on deepfake images of public figures generated on the free tier — xAI paywalled the feature and added safety filters to the aurora model that runs image generation. The paywall is unlikely to reverse; if you want to keep using Grok Imagine, budget for one of the three paid tiers.
If you're paid and it still fails: the five common causes
Paid tier users hit "not working" states too. Five patterns cover almost every failure.
Server overload. Since the March paywall funnel-ed all image gen through paid capacity, peak-hour saturation became a bigger problem. Grok's image servers throttle when demand spikes, and even paid users see failures during peak windows (US evening hours in particular). Waiting an hour and retrying often works.
Daily quota exhausted. Each paid tier has a daily generation cap. Hitting it silently stops generation for the day. There's no clear "quota reached" error message in the UI — it looks like a failure. The fix is waiting for the daily reset.
Stale browser cache or cookies. If Grok updated its frontend and your browser is running cached JavaScript, image gen calls can fail even when your account is fine. Hard refresh (Cmd/Ctrl + Shift + R) or clearing cache for x.com fixes this.
Ad blocker or VPN blocking the API. Grok Imagine's frontend calls the aurora API endpoint separately from the main chat. Aggressive ad blockers, privacy extensions, or VPN configurations that break API calls can cause silent failures. Temporarily disabling extensions for x.com or turning off the VPN isolates whether this is your issue.
Expired login session. Long-running browser tabs can hold an expired session that appears authenticated but fails on API calls. Log out fully and log back in.
The fix sequence
Work through these in order — most cases resolve in the first two steps.
1. Verify your subscription tier is active. Go to X/Grok settings and confirm the paid plan is current and hasn't expired or been declined. If the subscription lapsed, image gen stops even if chat still works.
2. Check daily quota. If you've generated a lot today, you may have hit the tier's cap. There's no explicit quota-reached indicator in the UI — a rough test is trying again in 6-12 hours after a daily reset.
3. Hard refresh the browser. Cmd/Ctrl + Shift + R clears cached frontend code that might be stale after an xAI deploy. If that doesn't work, clear all cookies for x.com and log in again.
4. Disable VPN and ad-blocker extensions for x.com specifically. Extensions that break API calls will produce silent failures on image gen while chat still works (chat uses a different endpoint).
5. Log out fully and back in. Resolves the expired-session pattern where you appear logged in but API calls fail authorization.
6. Wait for server capacity. If steps 1-5 don't help and it's peak hours (US evening), the issue is server-side saturation and waiting 1-2 hours often works.
If all six fail, xAI is likely having an incident — check the xAI status page or downdetector for a spike in reports.
For AI companion photos specifically: a different model
Grok Imagine is a general image-generation tool — you prompt it, it renders, you get back an image. Post-paywall, that's pay-per-generation via the tier's quota. This model works well when what you want IS general image generation and it fails predictably on companion-style use ("photos of the same character over time") because there's no persistent character to render — every generation starts from prompt.
Sloane's photo model is structurally different because the persona is durable.
Photos are part of the persona identity, not per-request generation. When Kaya sends a photo, it's rendered by a model trained on her specific visual identity. Same face, same look across every photo. Not a fresh generation from prompt each time.
Cadence-driven, not credit-driven. Kaya sends photos on her own schedule as part of the relationship — you don't burn credits per photo. Free-tier users get her cadence-photos included. Custom photo requests (asking for a specific scene) are what use credits, on Plus and Premium tiers.
No per-image quota that stops generation cold. The model that renders Kaya's photos is dedicated to Sloane's persona rendering rather than shared across a public image-gen API. Peak-hour saturation isn't part of the product.
The honest tradeoff: Sloane isn't a general image-gen tool. If you want to prompt "cyberpunk city at dusk in the style of Blade Runner," Grok Imagine is the right product; Sloane doesn't do that. What Sloane does is companion-shaped: consistent persona identity, photos that fit the character, no burn-credits-per-request pattern. Kaya is a warm-anchor persona a lot of former Grok Imagine users start with when they realize what they wanted was a specific character rather than a general prompt-to-image tool. Free tier is 50 messages per day with any persona, no card required.