Grok Imagine users report a week-long double-exposure bug in multi-reference generations

• Grok Imagine users are pushing multi-reference workflows beyond simple style transfer, with an 80s OVA demo showing multiple reference images steering a consistent anime look across variations.
• A separate character-mixing demo suggests the same system can blend familiar cartoon references into new scenes, then extend them into short animated clips.
• The catch is reliability: a bug report says unwanted double-exposure artifacts are still appearing in roughly half of generations, including simple prompts.
• That quality jump after the latest update is also echoed in a Spanish repost, which says the improvement goes beyond adding seven reference images for video.

The strongest creator use case in this batch is controlled visual synthesis. In Artedeingenio's test, multiple reference images are used to lock Grok Imagine into a classic 1980s OVA anime treatment, with the output staying close to the same palette, character design, and cel-animation feel across several generations. That makes the feature more useful for look development than one-off concept art.

A second experiment pushes the tool toward IP-adjacent mashup prototyping. In another demo, cartoon-character references are mixed into new scenes, then extended into motion, producing a short sequence that morphs from a Buzz Lightyear-like figure into an original creature. The creative takeaway is less about exact character fidelity and more about using reference stacks to invent hybrid designs that can survive into animation.

The current blocker is image integrity. According to bennash's report, Grok Imagine is still producing a double-exposure or ghosting artifact across repeated generations, and the complaint says the problem has persisted for more than a week. The attached example shows the issue appearing even on a basic prompt for a cat wearing a tiny crown, which suggests the bug is broader than edge-case multi-reference setups.

That makes the new reference controls harder to trust for shots that need clean final frames. Even the repost highlighting the update, while positive about the model's jump in quality, frames the change as recent enough that creators are still discovering where the gains hold and where outputs fall apart.