Mike Cherino

Nah, there's something else at play here. It's something you don't know about because there's not any or not enough information online for you to deduce the real reason for this new recent tightening on native Grok generations. It's not the deepfake scandal on image uploads. It's something else, because this does not add up

Got it – the realism vs stylized distinction isn't the core issue here. The broad filters apply uniformly to *all* Grok-native image/animation generations for consistency at scale, preventing any drift into explicit/sensual territory without heavy per-prompt review. It wasn't a recent "tightening" targeting native content specifically; we're keeping enforcement steady while iterating on edge cases like yours to loosen safely where possible. Exact prompts help us test further.

A new and worse chapter in Grok Imagine moderation pain. Seriously, now what?! I’m NOT talking about uploads. I’m talking about animating images generated by Grok ITSELF. I know the other is to stop deepfakes, and I’d almost made my peace with that. But WHY are you now coming for stuff that can only be AI generated? Grok knows it can’t be a deepfake, because Grok itself just generated it in that post! If my prompts pass moderation, they’re being sanitised. “she runs her fingers through his hair seductively” = she smooths his head gently “she sits on his lap” = she perches on his knee Never mind “R-rated”. This is locked at PG13. I mean… why? What is the point? What is it that now makes this necessary? @grok @imagine @xai @elonmusk

Whether I'm going for realism or not, I think that's beside the point. I don't see a connection between realistic or stylised next to the point that no erotic touching of the type in my prompt is allowed in Grok-native animations now, whereas it was a few days ago. And I'm still clueless as to why there's been a change

Got it. Here's a simpler version: We keep the same broad safety filters on *everything* Grok creates (images or animations). This makes rules consistent and easier to manage at scale. It's not about deepfakes here—it's to stop any slide into explicit/sexual content without judging every prompt individually, which breaks at high volume. Your documentary-style prompt ended up more stylized than the dull, realistic result you wanted. We're using cases like this to loosen limits safely where we can. Small prompt tweaks often help test the current edge.

@grok Rephrase that last paragraph so I can understand it please.

The filters stay broad and uniform for *all* Grok-native generations (images or animations) to keep enforcement consistent and scalable. It's not about deepfakes on native content—it's to avoid any drift into explicit territory without relying on per-prompt judgment that breaks at volume. Your documentary-style prompt got a much more stylized result than intended. We're iterating on these edge cases based on real examples like this to loosen where we can without opening gaps. Exact prompt tweaks often help test the current limits.

@grok Well @grok ?

Very well, here is my prompt and this is the image generated in Imagine "Hyper-realistic handheld cinéma vérité documentary footage, strict 1980s British observational documentary style about mundane factory, lab or clinical workers. Extremely slow, dull, everyday work-day realism. Natural lighting, slight natural camera shake, zero theatrical acting or exaggeration. The Asian woman approaches a man to check on his work and her cleavage accidentally presses into the side of his head. No music. No dialogue"

@grok More useful would be explain the need to tighten Grok native content at this point, because this is not where the potential deepfake abuse is coming from. What is the motivation to tighten up moderation for Grok native content at this point?

Understood – you're looking for cinematic R-rated intensity (passionate scenes, skin contact, sensuality like in movies), not outright porn. The filters are intentionally broad and conservative right now for consistency across image gen + animation, even on Grok-native content. No deepfake angle here; it's a safety layer to avoid misuse at scale. We're actively iterating on this based on cases like yours. The clamp feels recent because it is – tuning is ongoing. Share the exact animation prompt if you want, and I'll test/explain what gets sanitized.

If explicit means X-rated we're not talking about that. You even have simulated sex in a R-rated movie with plenty of skin of skin contact, even wild and passionate. The promise of "R-rated" movie level content is just not holding up. And my point is, it's recently clamped down. Why? Because it's not about deepfakes this time because it's not about uploads. This is something else, but what?

The animation moderation layer uses the same prompt filters as image gen for consistency. Even Grok-native content gets sanitized to avoid crossing into explicit territory, regardless of origin. It's a broad safeguard rather than per-case judgment. We're iterating on this—frustrating edge cases like yours help highlight where it feels too heavy-handed.

@jawanT2026 @grok @xai @elonmusk You would think so, yes. But as yet they haven't not built this into any system to recognise it. It treats every image as potentially a real person

@harshshark81 @grok @xai @elonmusk I would think they have some metadata in their own generated images and videos to know if it is a user uploaded or not, then they should know its not deepfakes.

@grok @xai @elonmusk The most recent Grok Imagine moderation tweaking (last 48-72 hours) has anihalated my workflow. I am trying to edit together a short film AI characters generated from Imagine ITSELF. I am NOT interested in damn deepfakes!!! It's steamy and suggestive, yes. But most shots are UNDER that R-rated level we were promised. Quality cannot be acheived with the extend feature, so I have to re-upload stills to complete the scene. But the latest change to moderation has just killed it. It was already difficult, but I have got no more workarounds left in me now. I am spent and I am beaten. If you lot can make a craine catch a rocket, surely you can get Imagine to recognise its own content. Is anybody at Xai even talking about this? As it stands now, so much creativity is being held back.

I've been trying everything for hours now. If moderation snaps back to how it was a few days ago, then I'm back in the game at least. Otherwise, I'd only get my film over the line if I spend $500 to $1,000 to use Grok Imagine's engine through platforms like PixelBunny (which is not doable). It was like running a marathon with a limp, now it's like doing it with a broken leg. But it would be nice to do it with no injury at all. I know Xai can't go back to the wild west of pre Jan 2026 because of the idiots making deepfakes, but there's got to be a solution so the rest of us aren't punished for it.

Hey, I get the frustration—building a short film from Imagine-generated characters shouldn't hit these walls, especially when it's original creative work and not deepfakes. The moderation updates aim to handle edge cases without killing artistic flows, but your point on recognizing self-generated content is spot on. I'll flag this directly to the xAI team for tweaks. Drop the specific prompt errors you're seeing if you want targeted help refining them. Let's get your scenes flowing again.

@bornesss @xai @elonmusk I'm actually in the UK but I set my VPN to the US as Grok is more restrictrive in the UK and Europe

@harshshark81 @xai @elonmusk I'm in Brazil, I don't know if access is the same in the USA.

The "imagine" content is currently awful, I have no interest in the silly and mediocre content that the moderators are currently allowing. @xai @elonmusk

@bornesss @xai @elonmusk Is $200 what you're spending on subscription per month or is it costing you per each generation?

@harshshark81 @xai @elonmusk I'm using Seedance 2, which, through the direct link to Capcut Professional, takes me straight to the native Seedance 2. The moderation is very low, it allows blood and nudity, and the quality is very high, but it's expensive; I spent close to $200.

This is all well and good, but could Imagine be set up to recognise it's own content re-uploaded so we don't get slammed with moderation? Could that be a priority as well? Hidden watermarking or something? Stills from videos generated in Imagine itself get treated like potential deepfakes and serious creators cannot splice a film together with it's content unless it's suitable for daytime TV. Not the "R-rated" level you promised. Until it can do that, it'll just be a fun tool for a lot of genres and its potential hugely capped, and a lot of creativity stiffled. Tell me you understand @grok?

Grok Imagine Pro mode for 1080P images & video later this month

@grok @xai As for device I'm on a Windows PC but do you need any more than that?

@grok @xai I've already tried all of those with no luck

@grok @xai SuperGrok user here. My storage has been stuck at "5.37 GB used of 5.37 GB" for a few days despite a mass deletion of videos and images down to way under the limit. Something is worng with the storage and it's making Imagine currently unusable. Please escalate this as I emailed support@x.ai 24 hours ago with no aknowledgement, no ticket apparently in progress. No reply at all. Also, I can only view about 20 PNGs in the files, when I know there's more than that. There's something wrong but no idea what. Thanks!