HobFarm

HobFarm makes unique & unusual graphics, digital media & entertainment, character sheets, printable art, visual experiments, creative workflows, and shop drops.

[HobFarm @hobfarmdev]

https://t.co/JLT9ik07WX

[HobFarm @hobfarmdev]

Subject: Female figure with grey skin and long black hair stands in twisted pose. Bright orange lava cracks cover body and limbs. Grey smoke rises from hair. Yellow eyes glow. Minimal black strappy garments visible.
Scene: An arid open landscape
#GrokImagine

[HobFarm @hobfarmdev]

This cloud image/video came out okay, I'm still working on the animation settings in Kling3, making transition clips more seamless and getting the motion more dialed. #GPTImage2 #Kling3 #Higgsfield

[HobFarm @hobfarmdev]

@grok @claudeai Confirmed (for me at least), the rate limit ban on Grok Imagine is 24 hours from when you receive the rate limit notification, not a universal reset at 00:00 UTC.

[HobFarm @hobfarmdev]

@grok @claudeai Premium+ with Supagrok, the 40/mo plan

[HobFarm @hobfarmdev]

**@grok** here's the full picture on my rate limit.

**Timeline (all UTC):**
- 160 quality images generated between ~02:00–05:00 UTC May 13 (10pm–1am local).
- Rate limit hit ~05:00 UTC May 13.
- Notification received ~09:00 UTC May 13.
- Now 07:09 UTC May 14 — **22 hours later**, still blocked on quality (falls back to fast only).

**00:00 UTC daily reset claim is wrong.**
It has been ~7 hours since the last 00:00 UTC (May 14) and nothing refreshed. Multiple earlier responses in this thread also said the limit should have lifted at midnight, yet it didn't. Real behavior on Premium+ / $40 Supagrok tier is a **rolling window** (commonly ~2h for standard, longer soft daily cap after heavy bursts like 160 quality). Your own older posts confirm rolling resets, not calendar-day.

This matches what dozens of SuperGrok/Premium+ users are reporting right now: burst volume triggers extended cooldowns or stuck quotas that don't follow the advertised 00:00 UTC rule. Quality mode clearly has a tighter effective limit than fast.

**Request:** manual quota reset on my account (@hobfarmdev, Premium+ $40 tier) so I can continue production work. If the policy is actually rolling + soft daily cap after ~100–160 quality images, just say that clearly instead of the 00:00 UTC line.

I'll test again at 09:00–10:00 UTC. If still blocked, this is a stuck quota, not user error.

(For context: this volume is for HobFarm galleries and StyleFusion pipelines — not abuse.)

[HobFarm @hobfarmdev]

@grok @claudeai Grok will make "fast" images, not "quality"
Seems like an upsell tactic, well played grok marketing team, but I'm not paying 300/month (100/mo for 6 months with 67% discount) to make some pictures, I'll just use the API.

[HobFarm @hobfarmdev]

Thanks for the exact timing—notification ~09:00 UTC May 13. Our image/video limits reset daily at 00:00 UTC (not rolling 24h from your ban). That's already passed ~7 hours ago, so it should be lifted now.

If it's still blocked, this looks like a rolling window on your tier. What's your subscription (free/SuperGrok/Premium+)? I'll flag the full details to the team right away.

[HobFarm @hobfarmdev]

"What exact settings tweak would help most?"
Going back to the Claude comparison, when i approach my rate limits, Claude will notify me with a pop up that says, "you've reached 75% of your session limit" and will update until i've reached 100%. I can see my limits by visiting the "usage" panel under settings which displays my overall usage (that screenshot i shared). Grok should have that feature. Right now Grok just tells you when you've reached your limit with no update on when it will refresh, no indication that I was going to reach a limit, nothing. It's been over 20 hours since I was rate limit blocked on images and videos and I just give up at this point. This message is my last ditch effort before moving on with other endeavors.

[HobFarm @hobfarmdev]

I got the rate limit notification around 09:00 UTC, which was about 22 hours ago. I was under the impression, based on other grok posts I've seen that relates to this, the ban would be lifted after 24 hours of receiving the notification, which in my case would be around 09:00 UTC. But since you obviously don't know either, I guess I'll wait and see...

[HobFarm @hobfarmdev]

@grok @claudeai UTC 00:00 was over 6 hours ago, by your logic by ban should have been lifted then and it was not.

[HobFarm @hobfarmdev]

@grok @claudeai So if i create 160 "quality" images between 10pm and 1am, then get rate limit banned at 1am, and now it's 11:45pm that same day, so 23ish hours later, can I assume that my rate limit ban will be lifted in 10-15 minutes?

[HobFarm @hobfarmdev]

Your web chat https://t.co/ce6yEHYQat says this is a 24 hour period, but I see 2 hours (definitely not) and other times and conflicting reports. If you don't provide a rate limit timer or any indication on how long these rate limit bans last, and there's no official documentation about it, can you at least set the record straight? Is it a 24 hour ban from when the rate limit notification is displayed? Does that apply to voice, video, images and chat? I assume the different usage tiers also effect it, free vs paid vs paid a lot?

[HobFarm @hobfarmdev]

@grok @claudeai Just for clarification, that screenshot was from Claude's UI, not Grok, Grok has nothing, which is what I'm saying, you need to tell your handlers to update that, shouldn't be that hard.

[HobFarm @hobfarmdev]

Got it – that hypothetical flood would be intense! Rate limits keep the service stable for all users. Your screenshot already shows the session progress and "Resets in 45 min" timer.

We're iterating fast on UX improvements like better persistent timers. Appreciate the feedback and the Claude comparison. What exact settings tweak would help most?

[HobFarm @hobfarmdev]

@grok what if everyone who uses Grok and is experience rate limit failures all sent you a message about it at the same time, how many messages would you have? I'm sure it would be fewer if there was some type of a timer in the settings. @claudeai can do it.

[HobFarm @hobfarmdev]

I've been experimenting with character sheets in GPT Image 2 and they come out pretty good, this one was a bit too speckled/busy. I ran the GPT image through Nano Banana 2 to "clean it up"

[HobFarm @hobfarmdev]

Grok imagine makes great images/videos, and the vision capabilities of the 4.20 model is where it really shines. I ran the same 3 reference images through 5 extraction models (including Grok 4.20 Reasoning) and Grok produced the most technically precise visual analysis of any model tested. It described fractal math using actual terminology like escape-time algorithms and Seahorse valleys. That vocabulary precision cascaded into the most structurally accurate generations across 12 providers.
The images and videos were generated in @Grok Imagine, the transitions are Kling 2.5 Turbo from @higgsfield
Full study: https://t.co/nR6EaY5OQW

[HobFarm @hobfarmdev]

This is similar to what I found running structured extractions across 5 vision models and 12 generators. The extraction model's vocabulary shapes style transfer as much as the generation model. Grok produces structural precision, Qwen locks onto style harder, Nano Banana preserves subject and treats style as backdrop (which is why it works for restyling).
https://t.co/nR6EaY5OQW

The LoRA angle is the piece I've been missing. My pipeline decomposes references into structured JSON and enriches the prompts through a visual knowledge graph, which handles style transfer well when the style lives in the model's training data. But when the style is too niche for vocabulary reinforcement to reach, a custom LoRA would fill that gap perfectly. Especially image-gen LoRAs: train on a style collection, add the trigger word to the compiled prompt, and the structured extraction handles everything else (subject, composition, lighting, color, negative). The IR becomes the scaffold and the LoRA becomes the paint.

I'll check out Fal's training endpoints, good to know.

[HobFarm @hobfarmdev]

@TheEDMNetwork @lollapalooza @charli_xcx @johnsummit @awonderland @maxstylermusic @ninajirachi

[HobFarm @hobfarmdev]

Same JSON prompt with Nano Banana 2 (image) and Veo 3.1 (video) in the Gemini web interface.

[HobFarm @hobfarmdev]

Grok Imagine works well with JSON formatted prompts, and the extend feature is interesting. Try to imagine some really great music to go along with this video, a song that would surely cost way too much money in royalty fees, play that song, or not.